10068

2022
November 2022

Populism and the Skill-Content

of Globalization: Evidence
from the Last 60 Years
Frédéric Docquier, Lucas Guichard, Stefano Iandolo, Hillel Rapoport,
Riccardo Turati, Gonzague Vannoorenberghe

Electronic copy available at: https://ssrn.com/abstract=4273442

Impressum:

CESifo Working Papers

ISSN 2364-1428 (electronic version)
Publisher and distributor: Munich Society for the Promotion of Economic Research - CESifo
GmbH
The international platform of Ludwigs-Maximilians University’s Center for Economic Studies
and the ifo Institute
Poschingerstr. 5, 81679 Munich, Germany
Telephone +49 (0)89 2180-2740, Telefax +49 (0)89 2180-17845, email office@cesifo.de
Editor: Clemens Fuest
https://www.cesifo.org/en/wp
An electronic version of the paper may be downloaded
· from the SSRN website: www.SSRN.com
· from the RePEc website: www.RePEc.org
· from the CESifo website: https://www.cesifo.org/en/wp

Electronic copy available at: https://ssrn.com/abstract=4273442

 CESifo Working Paper No. 10068

Populism and the Skill-Content of Globalization:

Evidence from the Last 60 Years

Abstract

We analyze the long-run evolution of populism and explore the role of globalization in shaping
such evolution. We use an imbalanced panel of 628 national elections in 55 countries over 60
years. A first novelty is our reliance on both standard (e.g., the ”volume margin”, or vote share of
populist parties) and new (e.g., the ”mean margin”, a continuous vote-weighted average of
populism scores of all parties) measures of the extent of populism. We show that levels of
populism in the world have strongly fluctuated since the 1960s, peaking after each major
economic crisis and reaching an all-time high – especially for right-wing populism in Europe –
after the great recession of 2007-10. The second novelty is that when we investigate the ”global”
determinants of populism, we look at trade and immigration jointly and consider their size as well
as their skill-structure. Using OLS, PPML and IV regressions, our results consistently suggest that
populism responds to globalization shocks in a way which is closely linked to the skill structure
of these shocks. Imports of low-skill labor intensive goods increase both total and right-wing
populism at the volume and mean margins, and more so in times of de-industrialization and of
internet expansion. Low-skill immigration, on the other hand, tends to induce a transfer of votes
from left-wing to right-wing populist parties, apparently without affecting the total. Finally,
imports of high-skill labor intensive goods, as well as high-skill immigration, tend to reduce the
volume of populism.
JEL-Codes: D720, F220, F520, J610.
Keywords: elections, populism, immigration, trade.

Frédéric Docquier Lucas Guichard

LISER, Luxembourg Institute of LISER, Luxembourg Institute of
Socio-Economic Research / Luxembourg Socio-Economic Research / Luxembourg
frederic.docquier@liser.lu lucas.guichard@liser.lu
Stefano Iandolo Hillel Rapoport
DISES, Università degli Studi di Salerno / Italy PSE, Paris School of Economics / France
siandolo@unisa.it hillel.rapoport@psemail.eu
Ricardo Turati Gonzague Vannoorenberghe
UAB, Universitat Autónoma de Barcelona IRES-LIDAM, Université catholique de
Barcelona / Spain Louvain / Belgium
riccardo.turati@uab.cat gonzague.vannoorenberghe@uclouvain.be

Electronic copy available at: https://ssrn.com/abstract=4273442

This version: October 26, 2022
This paper is part of the INTER project on “Globalization, Inequality and Populism across
Europe” supported by a grant from the Luxembourg FNR (EUFIRST, n.13956644) and by the
Belgian FNRS. We wish to thank Shuai Chen, Gabriel Facchini, Sergei Guriev, Dorothee
Hillrichs, François Maniquet, Massimo Morelli, Eugenio Peluso, Ariell Reshef, Jérôme Valette
and conference participants at the 2022 Meeting of the European Economics Association (Milan),
the 2022 ETSG (Groningen), the 2022 ITSG (Salerno), and seminars participants at University of
Sheffield, UAB and Université de Paris 1 Panthéon-Sorbonne for helpful comments and
suggestions. The usual disclaimers apply.

Electronic copy available at: https://ssrn.com/abstract=4273442

1 Introduction
The recent surge of populism is often portrayed as a rebellion of the losers from globalization. The
fall of the Communist Block and the ensuing opening of EU markets to trade and immigration from
Eastern Europe, China’s entry into the WTO in 2001, or the generalization of offshoring practices
toward low-wage countries since the 1990s have exposed workers and firms in industrialized countries
to a global competition that some (and certainly the populists) characterize as unfair. The same ’unfair
competition’ argument is used to describe the effects of low-skill immigration from poor countries on
rich countries native workers’ labor market outcomes. In this context, globalization has gradually
become a salient issue in the political discourses and public debates of most Western democracies.
This is best illustrated by the 2016 Brexit referendum in the UK, the election of Donald Trump in
the U.S. that same year, or by the electoral agenda and performance of populist parties in recent
elections in virtually all Western European countries. Besides their anti-establishment and anti-media
rhetoric, populist-nationalistic parties have long tried to gain popular support by tapping on people’s
concerns about the economic and social implications of globalization. And indeed, the link between
populism and globalization seems to cross the ages. As recalled by Guriev and Papaioannou (2021),
the late-19th-century American People’s Party, one of the first populist parties in the modern sense,
had a clear anti-globalization agenda. This link seems more relevant than ever, as evidenced by the
recent anti-globalization campaigns of La Lega and Movimento 5 Stelle in Italy, the Front National
and Reconquête in France, AfD in Germany, FPö in Austria, Podemos and Vox in Spain, the Vlaams
Belang in Belgium, etc. Anti-globalization stances are more and more frequent during and between
election campaigns (Colantone et al., 2021) and are voiced by political parties from the right as well
as from the left (Funke et al., 2020).
As noted by Rodrik (2018, p.12), ”the term [populism] originates from the late nineteenth century,
when a coalition of farmers, workers, and miners in the US rallied against the Gold Standard and
the Northeastern banking and finance establishment. Latin America has a long tradition of populism
going back to the 1930s, and exemplified by Peronism.” Several definitions of populism have been
used though, combining concepts such as anti-elite and anti-pluralism rhetoric (Mudde, 2004), identity
politics (Müller, 2016), authoritarianism (Eichengreen, 2018), anti-globalization view (De Vries, 2018;
Algan et al., 2018), communication style (Campante et al., 2018), or shortsighted political agenda
(Guiso et al., 2020).
This paper discusses the measurement of populism and documents its evolution over the last sixty
years; it then studies its determinants, focusing on the role of globalization shocks. Its contribution is
fivefold.
First, while we rely on standard measures of populism such as the sum of the vote shares of parties
classified as ”populists” (which we refer to as the ’volume” margin’), we note that populist ideas are
not restricted to populist parties but can spillover to traditional (or non-traditional) parties not defined
as populist. To reflect this and based on their political platforms, we propose to assign a continuous
populism score to all political parties competing in the elections in our data set (i.e., 628 national

Electronic copy available at: https://ssrn.com/abstract=4273442

elections in 55 countries during the period 1960-2018 Second, thanks to this continuous measure we
can study changes in populism not just along the ’volume margin’ but also along the ’mean margin’
(i.e., the vote-weighted scores of populism for all parties running in an election). The mean margin
does not rely on a dichotomous classification of parties into populist or not, and captures the overall
exposure of voters to populist ideas in a given election. Third, we conduct a unified analysis of the
effects of imports and immigration competition on populism, which we disentangle according to the
skill-content of immigration and import flows. Fourth, we implement an instrumentation strategy that
predicts changes in the bilateral and skill structure of imports and immigration using origin-specific
factors, generalizing the approach used in the trade and migration literature in a long panel setting
(Autor et al., 2020; Munshi, 2003; Boustan, 2010; Klemans and Magruder, 2018; Monras, 2020). And
fifth, we document and identify different evolution patterns and relations to globalization for left-wing
and right-wing populism. We relate and contribute to a growing literature on globalization and the
formation of political preferences in general, and on the political economy of populism in particular.
As far as trade is concerned, several papers focusing on the exposure to the “China trade shock”show
that the rise in Chinese imports triggered growing support for radical-right parties in a number of
OECD countries (Autor et al., 2020). These studies exploit variability in regional exposure to trade
with China. While looking at a well-identified shock, they use a relatively narrow time span (Becker
et al., 2017). Other studies show that populism tends to flourish in contexts of economic uncertainty
(Rodrik, 1997; Swank, 2003; Algan et al., 2017), which is itself partly generated by globalization shocks
(Di Giovanni and Levchenko, 2009; Vannoorenberghe, 2012; Caselli et al., 2015).
Similarly, the political economy of immigration literature has grown tremendously in the last ten
years. It includes explorations of the link between immigration and attitudes toward immigrants (e.g.,
Mayda, 2006; Card et al., 2012) or toward redistribution (e.g, Moriconi et al., 2019; Alesina et al.,
2021, 2022) as well as many studies identifying a causal positive effect between immigration and voting
for far-right, populist parties in contexts as various as the United States (Mayda et al., 2022), France
(Malgouyres, 2017), the United Kingdom (Colantone and Stanig, 2018; Becker and Fetzer, 2016; Becker
et al., 2017), Germany (Dippel et al., 2015), Italy (Barone et al., 2016), Spain (Mendez and Cutillias,
2014), Austria (Halla et al., 2017), Denmark (Harmon, 2018; Dustmann et al., 2019), Switzerland
(Brunner and Kuhn, 2018), in the city of Hamburg (Otto and Steinhardt, 2014), or more broadly
Western Europe (Guiso et al., 2017).1 These effects are often rationalized by the fear of adverse labor
market or of fiscal effects of immigration, or by identity/cultural factors, which in both cases depend
on the skill structure of the immigrant population (Edo et al., 2019; Moriconi et al., 2022, 2019).
Beyond trade and immigration, other key drivers of populism have been explored; these include
the role of automation and de-industrialization (e.g. Frey et al., 2018; Anelli et al., 2018), Gallego
et al. (2018) or the role of economic and financial crises (Funke et al., 2016; De Bromhead et al.,

1
By contrast, using the exogenous deployment of refugee centers during the 2015 crisis, Steinmayr (2021)
finds the opposite effect in Austrian neighborhoods. Along similar lines, Schneider-Strawczynski (2021) finds a
negative effect of the opening of a refugee center at the municipality level in France on votes for the National
Front, and disentangles a number of mechanisms such as ’contact’ and ’white flight’).

Electronic copy available at: https://ssrn.com/abstract=4273442

2013; Algan et al., 2017). The surge of populism has also been related to cultural factors and to the
perception that the elites are neglecting people’s concerns about identity, fairness, political distrust
(Norris and Inglehart, 2019; Mukand and Rodrik, 2018; Algan et al., 2018). Lastly, it has been shown
that populism benefited from the expansion of internet and social media (Zhuravskaya et al., 2020;
Campante et al., 2018; Guriev et al., 2019). We account for those other determinants and explore
interactions between them and globalization shocks.
Overall, we extend the literature by considering new measures of populism, over a longer period,
in a larger sample of countries, and by looking jointly at trade and at immigration while at the same
time accounting for their heterogeneous effects on the left-right spectrum of populism.
The paper is organized as follows. In Section 2, we construct a new continuous and time-varying
populism score for 3,860 party-election pairs (involving 1,206 unique parties in 628 political elec-
tions) using data on political manifestos across election campaigns. We rely on two criteria that are
well established in the political science literature to measure populism: the anti-establishment and
commitment-to-protect stances. We show that our new populism score is comparable across countries
and election periods, and describe its correlation with existing measures.
In Section 3, we use our populism score to describe the long-run trends in the volume and mean
margins of populism, the distance between populists and non-populists, and the comparative evolution
of right-wing v. left-right populism. We show that the mean level of populism has been fluctuating
since the sixties, with peaks during major economic crises such as the oil shocks of the seventies, deep
crises in the nineties (hitting Nordic countries, Mexico, South-East Asia, Russia, Brazil and Turkey),
and after the financial crisis of 2007-08. The surge of populism is not a pure European phenomenon
per se, but has become a widespread “pathology” in the European Union. The rise in the volume
and mean margins observed in European countries after 2005 is more pronounced than in the rest of
the world, a phenomenon that is not solely caused by the recent evolution in Eastern Europe. The
average populism score of right-wing populist parties has increased drastically since 2005, suggesting
a return to more authoritarian positions towards established elites, open markets, and protection of
minorities.
In Section 4 we then empirically link the trends in the volume and margins of populism to the size
and structure of import and immigration shocks. Exploiting dyadic data on import, migration and on
their skill intensities (Feyrer, 2019; Hausmann et al., 2007), we distinguish between shocks that are
likely to adversely affect low-skill voters and income inequality (such as imports of goods intensive in
low-skill labor or low-skill immigration), and those that are likely to adversely affect high-skilled voters
and decrease inequality. The surge in populism appears closely linked to the skill structure of imports
and immigration. Higher imports of low-skill intensive goods increase total and right-wing populism
along the volume and mean margins, with no effect on left-wing populism. As far as immigration
is concerned, low-skill immigration induces a transfer of votes from left-wing to right-wing populist
parties, without affecting the total volume or mean margin of populism. Interestingly, imports of
goods intensive in high-skilled labor and high-skilled immigration reduce the volume of populism.
These findings are typically stronger when using instrumental methods, thereby supporting a causal

Electronic copy available at: https://ssrn.com/abstract=4273442

interpretation of our results. Our results thus only partially align with Rodrik (2018)’s hypothesis
that globalization fosters right-wing populism when it takes the form of immigration shocks (as in
European countries), and left-wing populism when it takes the form of trade shocks (as in Latin
America). Section 5 concludes.

2 A Continuous Populism Score

Existing studies measuring populism typically classify political parties (or leaders) as either populist
or not based on experts’ opinions, as in Van Kessel (2015) or Rodrik (2018), or on an analysis of
political speeches and agendas. Such dichotomous definitions of populist parties neither capture the
“extent” of populism (Sikk, 2009) nor the fact that non-populist parties – potentially responding to
the populist “pressure” – may become more or less distant to the populist ones (Inglehart and Norris,
2016). In this section, we develop a continuous populism score for each political party that is time-
varying (parties can become more or less populist across elections) and consistent over time and across
space for a large set of countries since the early 1960s. Relying on political manifestos, our continuous
populism score can be used not only to document changes in the volume margin of populism – the vote
share of so-called populist parties – but also to characterize changes in the average level of exposure to
populism to which voters are exposed to at each election, what will be referred to as the mean margin
of populism. We first describe the methodology and data that we use to construct a populism score
(Section 2.1). We then confront our continuous populism score with existing studies covering different
sets of periods and countries (Section 2.2) and discuss our methodological choices in Section 2.3. We
present some stylized facts in Section 3.

2.1 Populism Scoring Methodology

For each party-election pair in our sample, we construct a populism score based on a content-analysis
p
of its political manifesto. We denote it by Si,e,t for party p ∈ (1, ..., P ) from country i ∈ (1, ..., I), in
election e ∈ (1, ..., E) at year t ∈ (1960, ..., 2018). Our scoring methodology is theory-based and relies
on two standard dimensions of populism, the anti-establishment and commitment-to-protect stances.
In Section 2.3, we show that deviating from this parsimonious definition of populism creates additional
noise and reduces comparability with existing measures and classifications.

Data. – We rely on the Manifesto Project Database (MPD), which characterizes a party’s political
preferences by counting the number of quasi-sentences associated with a specific issue compared to
the length of the party’s manifesto (salience). For some variables, the MPD reports separately the
salience of both positive and negative statements about an issue. In such a case we construct the net
position as the difference between the two. The MPD covers several political issues such as the position
on external relations (e.g., European Union and/or internationalism), the economic system (e.g., free
market economy v. market regulation), the welfare system (e.g., welfare state and public education

Electronic copy available at: https://ssrn.com/abstract=4273442

expansion), the fabric of society (e.g., the relevance of traditional morality and law enforcement) and
on specific social groups (e.g., working class and minorities). The MPD captures the positioning of
parties in the campaign, when parties are seeking to attract electors and before accepting possible
post-election compromises with other parties. The MPD covers all parties that won at least one seat
in an election campaign. Although debates can be engaged on selection issues, the one-seat constraint
excludes many independent candidates, and implies that parties that are very small or politically
insignificant are excluded from the sample. Figure A-I and Table A-I in the Appendix document the
geographic coverage of the MPD database and of our sample, respectively. The MPD also provides an
overall synthetic index positioning the party over the right-left political spectrum (Budge and Laver,
2016), as discussed below.

Dimensions of Populism. – Populism is a multi-faceted concept that involves different trends

and heterogeneous ideologies. To provide a consistent measure over space and time, we rely on
a parsimonious definition of populist parties, which is based on existing literature and associates
populism with two main characteristics.2
First, the anti-establishment stance (AES) is the key characteristic that recurs in all definitions of
populism. Populist parties build on the premise that high ethical and moral values are the hallmark of
the people, and not of the ruling class (Shils, 1956; Wiles, 1969). They highlight the divide between the
good, pure and homogeneous people, and the corrupt and self-centered elite (Taggart, 2000; Mudde,
2004; Van Kessel, 2015). Mudde (2004), a key reference in this literature, defines populism as “an
ideology that considers society to be ultimately separated in two homogeneous and antagonist groups:
the pure people against the corrupt elite, and which argues that politics should be the expression
of the general will of the people.” Such an antagonistic view implies that populists advocate the
sovereignty and protection of the people against the political establishment as well as against internal
and external threats (Stanley, 2008), which leaves no room for pluralism, diversity of opinions, and
even for the protection of minorities (Guriev and Papaioannou, 2021). We use two variables from
the MPD to proxy for the AES: the salience of, and position towards (i ) political corruption, which
include mentions related to the need to eliminate political corruption, power abuses and “clientelist”
structures; and (ii ) political authority, which proxies for anti-pluralism views and measures parties’
own statements about their relative competences and abilities.
Second, populism involves a strong commitment to protect (CTP) the people against threats driven
by external or alien entities (Morelli et al., 2021). Populists tap on the fear of people and base
communication on cleavages that go beyond the anti-elite rhetoric (Guiso et al., 2017; Rodrik, 2018).
Populists’ communication style is sometimes perceived as “chameleonic” (Taggart, 2000), and consists
in exacerbating feelings of resentment already present in the society to get support from followers.3
Pointing out economic inequality in income and wealth, left-wing populists tap on the economic
2
The exact description of these characteristics is provided in Appendix B.
3
Populist leaders simplify their discourse, and provide sound-bite and catchy solutions to real or imaginary
problems (Moffitt and Tormey, 2014). Their cleavage-based discourse is aggressive, authoritarian and critical
of the positions defended by other politicians, journalists and scientists (Guriev and Papaioannou, 2021).

Electronic copy available at: https://ssrn.com/abstract=4273442

cleavage between social classes or between capitalists and workers. Such a version of populism has
been widespread in Latin American and is still present in Venezuela, Ecuador, Bolivia or in the context
of a few developed countries (March and Mudde, 2005), as evidenced by the rise of Syriza in Greece,
of Le Parti du Travail de Belgique in Belgium, La France Insoumise in France, or Podemos in Spain.
By contrast, right-wing populists tap on the ethno-national or cultural cleavage, stressing the threat
of losing one’s national identity (from an ethnic, religious or cultural viewpoint) due to increased
immigration. Growing right-wing populism is evidenced by rise of the Tea Party and Trump’s election
in the U.S., the Lega Nord and Fratelli d’Italia in Italy, the Law and Justice Party in Poland, by the
growing success of the Front National in France, Alternative for Germany (AfD) in Germany, UKIP
and other partisans of Brexit in the UK, Vlaams Belang in Belgium, or by the re-elections of Victor
Orban in Hungary or of Recep Tayyip Erdogan in Turkey.
We rely on four variables in the MPD to proxy for the commitment-to-protect stance: the salience of
and position towards (i ) protectionism, which captures parties’ favorable statements towards the pro-
tection of the internal market, (ii ) internationalism, which refers to parties’ mentions of international
cooperation and national sovereignty, (iii ) European Community/Union, which includes mentions of
its expansion and increase in its competences, and (iv ) nationalization, which reflects mentions of
government ownership of land and industries.4

Populism score. To obtain a populism score based on the 6 dimensions of the MPD identified
above, we perform two stages of dimensionality reduction. In the first, we perform a Principal Compo-
nent Analysis of the variables belonging to each populism dimension (AES and CTP), and construct
a synthetic indicator for each of them. Panel I of Table 1 shows the results of the PCA for the two
dimensions of populism. Col. (1) gives the eigenvalues associated to each variable. Following the
so-called Kaiser’s criterion, we focus on the first component only, which retains a sizeable amount
of variance and exhibits eigenvalues above one (Preacher and MacCallum, 2003). Col. (2) gives the
score of the first component associated to each variable, and Col. (3) shows the correlation between
the estimated first component and each of the underlying variables. This first stage gives rise to two
synthetic indicators capturing political parties’ positions with respect to anti-establishment (AES) and
commitment-to-protection (CTP) stances.
In panel II of Table 1, we estimate the partial correlations between our two synthetic indices AES
and CTP after controlling for country and year fixed effects and for parties right-leaning ideology
(RW) (available in MPD (Budge and Laver, 2016)). The results are reported in Cols. (4) to (7), and
the R-squared of the regressions are provided in Col. (8). These regressions suggest that AES and
CPT are positively and highly significantly related one to the other. Finally, in Cols. (9) to (11) of
panel III, we provide the standard deviation (SD), the minimum (Min) and the maximum (Max) of
the two synthetic indices.
4
MPD documents positive attitudes towards nationalisation. As for dimensions (i), (ii) and (iii), it provides
net favorable positions corresponding to the difference between positive and negative mentions. Finally, for
parties belonging to non-European countries, component (iii) is set to zero. Similar variables have been adopted
in Colantone et al. (2021) to build a measure of parties’ autarky stance.

Electronic copy available at: https://ssrn.com/abstract=4273442

 p
Table 1: Construction of the populism score (Si,e,t ) using a two-stage PPCA

I. PPCA (AES/CTP) II. Corr. btw. AES & CTP III. Descriptives
EV Score Corr. AES CTP RW R2 SD Min Max
(1) (2) (3) (4) (5) (7) (8) (9) (10) (11)
Anti-establish- .09† .01†
- .27 1.03 -.72 8.27
ment (AES): (.02) (.00)
- Pol. corruption 1.07 .71 .73‡
- Anti-pluralism .93 .71 .73‡
Commitment to .13∗∗ -.01∗
- .11 1.13 -5.81 10.94
protect. (CTP): (.04) (.00)
- Protectionism 1.29 .41 .48‡
- Internationalism .96 -.41 -.46‡
- EU institutions .92 -.60 -.67‡
- Nationalization .83 .55 .63‡
Populism score .81 -3.27 5.61
Notes: Panel (I) shows the results of the polychoric principal component analysis (PPCA). Cols. (1), (2) and (3)
give eigenvalues (EV) associated to each variable, their scoring, and the correlation between the first component
of the PPCA and the variables in the analysis. Panel (II) shows the partial correlations between dimensions after
controlling for a left-to-right index of parties’ position over the political spectrum, country and year fixed-effects.
Standard errors are clustered at the country level. Panel (III) provides some descriptive statistics. Level of sig-
nificance: * p<0.05 ; ** p<0.01 ; † p<0.001 ; ‡ p<0.00001.

In a second stage of dimensionality reduction, we perform a weighted average of the two synthetic
indicators extracted from the first stage, and identify a general populism score for each election-party
pair. In our context, performing a PCA would provide identical results, with the same weights assigned
to the two synthetic indicators. In the bottom panel of Table 1, we show the descriptive statistics
p
associated to the populism score, Si,e,t . By construction, each index has a zero mean, while the
standard deviation equals 0.81.

Right-wing vs. left-wing populism. – Populism is a “thin” ideology, which can be combined
with other political views and can easily adapt its position on salient political issues at stake (Taggart,
2000; Mudde, 2004; Rooduijn et al., 2014). In particular, populism is usually identified as right-
wing or left-wing populism based on the type of cleavage used to create two antagonist groups in the
society. Mobilization of voters along income/social class lines is associated with left-wing populism.
By contrast, tapping on the ethno-national/cultural cleavages is associated with right-wing populism.
Based on the work of Budge and Laver (2016), we position parties over the left-right political
scale using the left-right index (rile) available in the MPD. We consider as left-wing (as right-wing,
respectively) those belonging to the first tercile (third tercile, respectively) of the left-right political
scale distribution. Those in the second tercile are classified as centrist. It is worth emphasising that

Electronic copy available at: https://ssrn.com/abstract=4273442

this classification along the left-right spectrum is governed by several factors such as parties’ attitudes
towards redistribution and political preferences that are related to moral values (e.g. on law and order,
traditional morality, importance of military forces, anti-imperialism, etc.). In unreported regressions
(available upon request) we show that on average, the highest populism scores are associated with
radical right and, to a lesser extent, radical left parties – the classification by political family is
provided in the Chapel Hill Expert Survey for the 1994-2014 period. By contrast, the least populist
family is that of the “green” parties, followed by traditional (liberal, Christian-democratic and socio-
democratic) parties.

2.2 Comparison with existing measures of Populism

Other populism indices and classifications have been developed in the political science literature. The
most commonly used classifications heavily rely on the anti-establishment stance proposed by Mudde
(2004); they cover different sets of countries and periods.

As a first step, we focus on four databases providing a dichotomous classification of parties, and
investigate whether our continuous populism score is a good predictor of a party’s probability to be
classified as populist. The four existing databases are:

 Van Kessel. Van Kessel (2015) identifies populist parties based on their manifesto and political
discourses in 31 countries over the 2000-2013 period. A party is defined as populist if it portrays
people as virtuous and homogeneous, if it claims popular sovereignty and positions itself against the
political elite. This data set identifies 57 populist parties. It has been used as a relevant reference
point for alternative populism measures (e.g., Guiso et al., 2017).

 Swank. Based on the definition of right-wing populism provided by Betz (1994), Swank (2018)
identifies about 30 right-wing populist parties in 21 countries over the 1950-2015 period. Betz
(1994) defines right-wing populist parties as those providing a mixed political stance based on
economic liberalism, questioning of the legitimacy of democracy, and fueling xenophobic views.5
Left-wing populist parties are not included.

 PopuList. The PopuList dataset developed by Rooduijn et al. (2019) identifies a list of populist
parties over the 1989-2020 period for 31 developed countries. Validated by more than 80 academic
scholars, it includes parties that have won at least one seat or at least 2% of the votes in an election.
The information for the 212 parties available in the PopuList data set has been frequently used in
recent studies of populism (e.g., Guiso et al., 2020; Morelli et al., 2021).6
5
A few parties identified by Swank (2018) as right-wing populist are not available in our sample due to the
low percentage of votes received during their national elections (e.g., Démocratie Nationale in Belgium or the
National Renovator Party in Portugal).
6
The sample of countries includes: Austria, Belgium, Bulgaria, Croatia, Cyprus, Czech Republic, Den-
mark, Estonia, Finland, France, Germany, Greece, Hungary, Iceland, Ireland, Italy, Latvia, Lithuania, Luxem-
bourg, Malta, Netherlands, Norway, Poland, Portugal, Romania, Slovakia, Slovenia, Spain, Sweden, Switzerland,
United Kingdom.

Electronic copy available at: https://ssrn.com/abstract=4273442

 GPop 1. The Global Populism data (Grzymala-Busse and McFaul, 2020) from the Freeman Spogli
Institute for International Studies provides information on populist parties (only) for 40 developed
and developing countries over a long period (1916-2018).7 This data set is particularly relevant
for our analysis, since it allows us to cross-validate our time-variant measure over a time-invariant
definition of populist parties for the whole 1960-2018 period.

In Panels I to IV of Table 2, we regress classifications of populist parties provided in existing

p
studies on our continuous populism score (Si,e,t ) and on its two components (AES and CTP). We
estimate Probit models (denoted by PRB). Partial correlations are provided for Van Kessel in Panel
I, for Swank in Panel II, for the PopuList database in Panel III, and for the GPop 1 database in
Panel IV. In all cases, we control for country and election-year fixed effects, to capture countries’
time-invariant unobserved heterogeneity and common year trends. The estimates suggest a positive
and highly robust correlation between our populism score and the probability to be classified as a
populist party in the existing literature.
To better grasp the quality of the fit of our Probit models with respect to the different binary
definitions of a populist party, we first compute the predicted probability of being defined a populist
party using the estimated models, and we define the set of predicted populist parties as the ones
characterized by a predicted probability of being populist above 0.5. Following Naik and Leuthold
(1986) we then compute the ratio of accurate forecasts (RAF), which is the percentage of predicted
populist identifiers (either 0 or 1) corresponding to the actual data set of reference. The ratio of
accurate forecasts takes value between 80% to 91%, suggesting that our predictions nicely fit alternative
classifications. Interestingly, the highly significant correlation levels obtained for Global Populism data
(GPop 1) over the 1960-2018 minimize concerns related to comparability and consistency issues over
our long period of analysis.8

In a second step, we produce our own classification of parties using our continuous and centered
(i.e., zero-mean) score of populism. This classification is needed to define the volume margin of
p
populism. We classify a party as populist when its populism score Si,e,t exceeds a certain threshold,
which can be expressed as a multiplying factor η of the standard deviation of the distribution (SD).
We define a dummy 1pi,e,t equal to 1 if the party p from country i is classified as populist in election e
at year t, and 0 otherwise: (
p
1 if Si,e,t ≥ η × SD
1pi,e,t = (1)
0 otherwise.

7
The sample of countries includes: Albania, Argentina, Austria, Belgium, Bolivia, Bulgaria, Chile, Croatia,
Czech Republic, Denmark, Estonia, Finland, France, Georgia, Germany, Greece, Hungary, Ireland, Israel, Italy,
Japan, Latvia, Lithuania, Mexico, Moldova, Netherlands, Poland, Portugal, Romania, Russia, Serbia, Slovakia,
Slovenia, South Korea, Spain, Sweden, Switzerland, Turkey, Ukraine, United Kingdom.
8
Controlling for the left-right index hardly affects the correlations between alternative definitions of populism
and our populism score or its commitment-to-protect component. The correlation with the anti-establishment
index is less robust, suggesting that parties’ ideological orientation captures part of the anti-establishment
stance.

10

Electronic copy available at: https://ssrn.com/abstract=4273442

 Table 2: Correlation with existing classifications of populist parties

I. Van Kessel (2000-2013) II. Swank (1960-2015) III. PopuList (1989-2018)

Populist party (PRB) RW Populist party (PRB) Populist party (PRB)
(1) (2) (3) (4) (5) (6) (7) (8) (9)
p 0.699∗∗∗ 0.460∗∗∗ 0.550∗∗∗
Si,e,t
(0.161) (0.112) (0.094)
AES 0.247∗∗∗ 0.252∗∗ 0.156∗∗∗
(0.091) (0.100) (0.054)
CTP 0.474∗∗∗ 0.234∗∗∗ 0.428∗∗∗
(0.093) (0.045) (0.069)
Obs. 650 650 650 1658 1658 1658 1635 1635 1635
Countries 25 25 25 16 16 16 28 28 28
Country FE 3 3 3 3 3 3 3 3 3
Year FE 3 3 3 3 3 3 3 3 3
Pseudo R2 0.18 0.08 0.18 0.16 0.12 0.14 0.17 0.09 0.19
RAF (%) 82.3 81.5 82.6 91.4 91.6 91.4 86.2 86.1 86.4

IV. GPop 1 (1960-2018) V. GPop 2 (1998-2017) VI. CHES (1998-2018)

Average Populism
Populist party (PRB) People vs. Elite (OLS)
Speeches (OLS)
(10) (11) (12) (13) (14) (15) (16) (17) (18)
p 0.376∗∗∗ 0.120∗∗ 1.262∗∗∗
Si,e,t
(0.081) (0.052) (0.210)
AES 0.093∗ 0.057∗ 0.933∗∗∗
(0.050) (0.032) (0.257)
CTP 0.277∗∗∗ 0.087∗ 0.668∗∗∗
(0.053) (0.046) (0.130)
Obs. 2847 2847 2847 100 100 100 176 176 176
Countries 36 36 36 31 31 31 28 28 28
Country FE 3 3 3 7 7 7 3 3 3
Year FE 3 3 3 3 3 3 3 3 3
Pseudo-R2 0.16 0.12 0.17
RAF (%) 88.9 88.6 88.7
R2 0.22 0.19 0.22 0.37 0.21 0.33
Notes: In Cols. (1) to (12), we provide partial correlations between parties’ political induces and the prob-
ability of being coded as populist party or right wing populist party following the definition of Van Kessel
(2015), Swank (2018), Rooduijn et al. (2019) and Grzymala-Busse and McFaul (2020) and adopting a probit
model. Each regression controls for country and year fixed effects. We also provides the ratio of accurate fore-
casts (RAF) between our estimated model and actual data, using a predicted probability of 0.5 as threshold
to define a party as populist. In Cols. (13) to (15), we provide partial correlations between political indices
and party leader’s speeches (Hawkins et al., 2019) after controlling for year fixed-effects. In Cols. (16) to
(18), we provide partial correlations between political indices and expert evaluations of parties degree of pop-
ulism (Bakker et al., 2015). Standard errors are clustered at country level. Level of significance: * p<0.1, **
p<0.05, *** p<0.01.

11

Electronic copy available at: https://ssrn.com/abstract=4273442

 The classification depends on the populism threshold determined by η. To identify a relevant
threshold, we compare our classification of populist parties with those of existing studies when we
gradually increase η from 0 to 2 (i.e., 0 to 2 standard deviations above the mean). As our database
includes more parties and elections than alternative databases, the statistics are computed for the
party-election pairs included in each alternative database – namely those of Van Kessel, Swank, Pop-
uList and GPop 1. We investigate the capacity of our populism score to predict the probability to be
classified as populist in these databases. We estimate new Probit models for each of the four dependent
variables with three sets of explanatory variables, a dummy 1pi,e,t equal to one if the party is classified
as populist according to our criteria (η),9 country and year fixed effects.
Figure 1 shows that η = 1 determines a relevant threshold, maximizing the partial correlation
with three existing classifications. A more restrictive threshold might be desirable to maximize the
partial correlation with the GPop 1 database. However, Figure B-I in the Appendix shows that η = 1
also maximizes the rate of accurate forecasts for the overall set of parties and for populist parties
only, whatever the classification used as a reference (even when using the GPop 1 classification).
Consequently, when using a dichotomous classification of parties to compute the volume of populism,
we classify parties with a populist score exceeding one standard deviation as populist in the rest of
the paper.10

In a third step, we also investigate whether our populism score and its two components are well
correlated with other continuous measures of populism from the existing literature. The latter are
provided in two additional databases covering a limited number of years:

 GPop 2. The Global Populism Data (Hawkins et al., 2019) provides a continuous measure of pop-
ulism based on textual analysis of the political discourses of parties’ leaders who won the national
election. The analysis is limited to presidents or prime ministers (depending on the institutional
context). The measure is based on four types of speeches – campaign speeches (usually closing
or announcement speech), ribbon-cutting speeches, international speeches and famous speeches.
Speeches are categorized between 0 (containing few populist elements) and 2 (extremely populist).
The sample includes 31 countries over the 1998-2017 period.11

 CHES. The Chapell Hill Expert Survey (Bakker et al., 2015) provides a continuous index of pop-
ulism, based on expert surveys and following the definition of Mudde (2004). By asking whether
parties believe that the people should have the final say on political issues against the elite, CHES
provides a continuous measure of populism (from 0 for pro-elite views, to 10 for pro-people views).

9 p
In Table 2, we regressed existing populist dummies on our continuous score (Si,e,t ).
10
When describing trends and exploring the determinants of populism, we will assess the robustness of our
findings when considering threshold levels equal to 0.9 standard deviations (referred to as the lax threshold, η)
and to 1.1 standard deviations (referred to as the strict threshold, η). As shown in the Appendix, all stylized
facts highlighted in the subsequent sections are highly robust to the choice of the threshold level.
11
It includes Austria, Bulgaria, Canada, Croatia, Czech Republic, Estonia, Georgia, Germany, Greece, Hun-
gary, Ireland, Italy, Japan, Latvia, Lithuania, Moldova, Montenegro, Netherlands, North Macedonia, Norway,
Poland, Romania, Russia, Serbia, Slovakia, Slovenia, Spain, Sweden, Turkey, Ukraine and UK.

12

Electronic copy available at: https://ssrn.com/abstract=4273442

 Figure 1: Populist parties – Threshold definition

1.4
2

1.2
1.6

1
Van Kessel ID (beta)

Swank ID (beta)
1.2

.8
.6
.8

.4
.4

.2
0

0
0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0 0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0

Populism ID (Threshold, SD) Populism ID (Threshold, SD)

(a) Van Kessel (b) Swank

1.6

1.2
1
1.2
PopuList ID (beta)

.8
GPop 1 ID (beta)
.8

.6
.4
.4

.2
0

0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0 0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0
Populism ID (Threshold, SD) Populism ID (Threshold, SD)

(c) PopuList (d) GPop 1

Note: The Figure shows the partial correlations between a dummy which defines a party as populist based
on different threshold of the populism score (x-axis) and a populist identifier based on: Van Kessel (2015)
(Panel a), Swank (2018) (Panel b), Rooduijn et al. (2019) (Panel c) and Grzymala-Busse and McFaul (2020)
(Panel d). The partial correlations are estimated from a probit model, including country and year fixed
effects. The rate of accurate forecasts for the overall set of parties and for populist parties only are provided
in Figure B-I in the Appendix.

13

Electronic copy available at: https://ssrn.com/abstract=4273442

 However, this index is available in the last wave of the survey (2019) only, and for a reduced number
of parties (247). To have a proper comparison with our dataset, we match CHES observations with
parties participating in the last electoral event available in the MPD. Since MPD includes parties
that won at least a seat during the elections, the matched sample includes 176 parties over 28
countries over the 1998-2018 period.12

By estimating standard OLS, Panels V and VI of Table 2 show positive and significant correlations
p
between our populism score Si,e,t and its components as well as with the average index of populist
speeches from GPop 2 and the CHES pro-people indicators. These results highlight a convergence in
identifying populist parties using as proxies leader’s speeches and/or expert surveys.

2.3 Discussion
Although our populism score is a good predictor of existing continuous measures and classifications,
its construction relies on a parsimonious definition of populism, and its validation is based on a
comparison with existing measures that are taken as ground truth. We assess here whether these
working hypotheses are relevant in our context.
First, we investigate whether a better predictor of existing measures can be obtained when de-
parting from the parsimonious (bi-dimensional) definition of populism. By focusing on two main
characteristics identified in the literature (i.e., anti-establishment and commitment to protect), our
populism score abstracts from a significant amount of relevant information available in the MPD. In
Appendix B.6, we use similar dimensionality reduction techniques to construct two extended populism
scores that exploit additional potential characteristics of populist parties, and check whether these
p
extended scores (say Ŝi,e,t ) better correlate with existing measures. Our first extended score accounts
for the fact that populist parties are sometimes characterized by their shortsighted and opportunistic
research agenda, which guides their political strategy (Guiso et al., 2017). We combine two additional
MPD variables covering aspects which are primarily influenced by policies with a long-term perspec-
tive such as education and environmental issues. Our second extended score accounts for the whole
set of information available in MPD. We construct synthetic indices of political preferences using the
remaining set of 44 variables available from the MPD. We then compute correlations with existing
measures and classifications. Although the extended populism scores account for a larger number of
political characteristics, they do not provide significantly better proxies for populism, as suggested by
the smaller magnitudes of the estimated partial correlations presented in Table B-VIII in the Appendix
(in other words, adding more information to the populism score can create additional noise).
Second and in the same vein, our parsimonious definition of populism voluntarily disregards MPD
statements that directly capture the salience of cultural and immigration-related aspects. The reason
is that right-wing parties (cleavage based on cultural identify) and left-wing populist parties (cleavage

12
The list of countries includes Austria, Belgium, Bulgaria, Croatia, Cyprus, Czech Republic, Denmark,
Estonia, Finland, France, Germany, Greece, Hungary, Ireland, Italy, Latvia, Lithuania, Luxembourg, Malta,
Netherlands, Poland, Portugal, Romania, Slovakia, Slovenia, Spain, Sweden and United Kingdom.

14

Electronic copy available at: https://ssrn.com/abstract=4273442

 10
Figure 2: Unsupervised clustering analysis on two dimensions of populism

10
8

8
FPÖ,1994 FPÖ,1994
Anti-establishment index

Anti-establishment index
6

6
4

4
FPÖ,1986
2

2
FPÖ,1990

FPÖ,2013
DF,2005 FPÖ,2013
DF,2005
FPÖ,1983 FN,2007
AfD,2017 FN,2007
AfD,2017
UKIP,2017 DF,2011 UKIP,2001
DF,2011
UKIP,2001
FN,1986 FPÖ,2017 FN,2002
FN,1993 FN,1997
FN,1988 FN,2002
0

0
FN,2017 UKIP,2015 FN,1997
FN,2017 UKIP,2015
FPÖ,1970
FPÖ,1966
FPÖ,1962 FPÖ,2002
FPÖ,1999
FPÖ,1979
FPÖ,1971
FPÖ,1975
FPÖ,2008
FPÖ,2006
FN,2012
DF,2007
DF,2001
AfD,2013
DF,1998
DF,2015 DF,2001
AfD,2013
DF,1998
DF,2015
-2

-2
-6 -4 -2 0 2 4 6 8 10 12 -6 -4 -2 0 2 4 6 8 10 12
Commitment to Protect index Commitment to Protect index

(a) All parties (b) Populist parties only

Notes: We perform a clustering analysis using the two dimensions associated to the standard populism score:
anti-establishment and commitment-to-protection stances. The left panel presents the space including all
the parties, while the right panel presents the space once we focus on populist parties only.

based on social classes) are likely to differ drastically on these issues. Controlling for country and
p
year fixed effects, we computed partial correlations between our populism score Si,e,t and four MPD
variables capturing preferences for immigration and multiculturalism.13 Note that these variables are
not available for the years prior to 2006, which also explains why they cannot be accounted for when
constructing our extended populism score. In line with intuition, we find that the populism score of
centrist and right-wing parties positively and significantly correlates with negative attitudes towards
immigration and multiculturalism. The correlation is insignificant when the sample is restricted to
left-wing parties. We also computed pairwise correlations between our populism score and proxies for
(i) cultural conservatism, and (ii) preferences for government intervention and economic planning. We
find that the populism score of centrist and right-wing parties is positively and significantly correlated
with cultural conservatism; this is not the case among left-wing parties. Interventionism and populism
are positively and significantly correlated on both sides of the left-to-right spectrum (and more so for
left-wing populism). Results are provided in Appendix B.2.
Third, instead of considering existing databases as a reference basis, we stick to our parsimonious
selection of political dimensions, and check whether an unsupervised machine-learning algorithm can
validate our dichotomous classification of parties (1pi,e,t ). Remember that classifying parties with a
populism score exceeding one standard deviation as populist matches well alternative definitions of
populism from existing literature. As an alternative approach, we also perform a cluster analysis over
the two dimensions of populism identified in the left panel of Table 1 (i.e., AES and CTP). We use

13
Namely, (i) immigration is negative for country’s national way of life, (ii) immigration is positive for
country’s national way of life, (iii) immigration positively contributes to multiculturalism, and (iv) immigrant
should assimilate to the country culture.

15

Electronic copy available at: https://ssrn.com/abstract=4273442

the unsupervised k-means clustering method (with the Euclidean distance as dissimilarity measure),
which does not require an a priori classification or measurement of populism. On Figure 2, the left
panel considers all election-party pairs and identifies three clusters of parties colored in grey, red and
blue in the two-dimensional space. On the right panel, we select election-party pairs with populism
score above the one standard deviation threshold. The clustering approach clearly shows that parties
above the one standard deviation threshold belong to a specific cluster in the two-dimensional space,
which means that they are both anti-establishment and committed to protect, or that they exhibit a
very large index along one of those two dimensions.14

3 Trends in Populism over 60 Years

In this section, we analyze the evolution of the distribution and mean level of populism focusing on 55
countries over almost six decades. As stated above, we distinguish between the mean level of populism
of all political parties – a concept that captures voters’ exposure to (and the extent of) populism
without requiring a dichotomous classification of parties – and the vote share of populist parties –
a concept that has been abundantly used in cross-country and case studies. Overall, our analysis
confirms that (i) populism is not a recent phenomenon (ii) both margins of populism have waxed and
waned over the last decades, and (iii) populism in general, and right-wing populism in particular has
become much stronger in Europe over the last decade. Appendix B.5 shows that very similar trends
are obtained when using a balanced sample of countries from 1960 to 2018, suggesting that those
evolutions are not driven by changes in the composition of our sample.

Distribution of, and trends in populism scores. – We first abstract from the dichotomous
classification of parties and aggregate the populism scores of all parties included in the sample by
period. Figure 3 describes the distribution of populism scores across parties (top panel) and shows
different measures of the evolution of the average level of populism over time (bottom panel).
Panels (a) depicts the changes in the density of the populism score across all political parties and
countries. The populism score is normally distributed in all decades. We observe a slight increase in
the mean, variance, and right skewness (at least, an increase in the density in the range of 1 to 2)
during the last decade. Panel (b) depicts the evolution of the Theil index of inequality in populism,
and of its between-country and within-country components. Inequality in populism declined between
the sixties and early eighties, peaked in the early nineties before declining again, and increased between
the financial crisis of 2008 and 2015. The between-country component has been rather stable until the
mid-eighties, and has gradually decreased since then. On the contrary, the within-country component
– the dominant component in most periods – has shown greater variations and significant increased
during the eighties and after 2008, which may reflect a polarization of populist stances and/or vote
shares in these periods.
14
It is worth emphasizing that this pattern is less clear-cut when applying the same unsupervised machine-
learning algorithm to extended populism scores (see Appendix B.6).

16

Electronic copy available at: https://ssrn.com/abstract=4273442

 Figure 3: Stylized facts I – Distribution of populism scores and mean margin of populism

.04
.8

.03
.6
Density

Theil index
.4

.02
.2

.01
0

-6 -4 -2 0 2 4 6

0
Populism index 1960 1970 1980 1990 2000 2010 2020

1960-69 1970-79 1980-89 1990-99 2000-09 2010-18 Theil all Theil within Theil between

(a) Density of populism score by period (b) Inequality across parties (Theil)

.4
.4

Adj. average populism index (weighted)

.2
Average populism index
.2

0
0

-.2
-.2

-.4

1960 1970 1980 1990 2000 2010 2020 1960 1970 1980 1990 2000 2010 2020

Full sample EU28 RoW Full sample EU28 RoW

(c) Mean populism score of all parties (unweighted) (d) Mean margin of populism (ΠM
i,e,t )

Notes: Fig. (a) shows the kernel-density of the populism score by decade. Fig. (b) depicts the Theil index
of inequality in populism across parties, and gives its between-countries component and the within-countries
components (Cadot et al., 2011). Fig. (c) plots the average populism score of all parties running for election
in a given year. Fig. (d) plots the mean margin of populism, a weighted average of the populism scores
with weights equal to the party’s share in votes. Fig. (c) and (d) show moving averages including 3 years
before and 3 years after each date. The vertical lines indicate shifts in our sample size: inclusion of Greece,
Portugal and Spain around 1975, and inclusion of Latin American and former soviet union countries around
1990. Similar trends are obtained in the balanced sample (see Fig. B-V in Appendix B.5).

Panels (c) and (d) characterize the evolution of the mean level of populism since the early sixties.
In Panel (c), we compute the mean populism score of all parties running for election in all years,
PI PP p
disregarding their electoral success (i.e., i=1 p=1 Si,e,t /I/P ). This mean level can be seen as a
(continuous) proxy for the supply of populism. However, one needs to be very careful with this
interpretation as the populism stance of parties is endogenous to the potential demand for populism.

17

Electronic copy available at: https://ssrn.com/abstract=4273442

The populism score has fluctuated since the early sixties, with peaks aligned with major economic
crisis – the oil crisis of the seventies, the deep crises of the nineties, and the years after 2005. The
average level observed in 2018 is larger than the level observed in 1960, but smaller than the peak
of the late seventies. This masks disparities between European (EU28) and non-European (RoW)
countries. In the European Union, the level observed in 2018 is way larger than the level observed
in 1960, and slightly greater than the level of the late seventies. It is worth emphasizing that this
evolution is not solely driven by the rise of radical right parties in Eastern European countries. In
Appendix C.3, we show that very similar trends are observed when focusing on the EU15 countries.
In non-European countries, current levels are lower than those observed in the seventies.
Finally, Panel (d) accounts for the vote shares and depicts the “post-election” mean level of
exposure to populism. We define this weighted average as the mean margin of populism, which is
computed at the aggregate level as:
PI PP p p
i=1 p=1 Si,e,t πi,e,t
ΠM
e,t = PI PP p , (2)
i=1 p=1 πi,e,t

p
where πi,e,t is the vote share for party p in election e of country i at year t. Note that the mean margin
can also be computed at the level of each country (ΠM
i,e,t ) by removing the summation over i in the
above equation. The latter variable will be used as a dependent in our regression framework.
Panel (d) shows that the evolution of the mean margin of populism is very similar to that of
the unweighted average level (i.e., peaks aligned with economic crisis). The rise observed in European
countries after 2005 is more pronounced, and the European and non-European mean levels are currently
almost identical. Hence, the surge of populism is not a pure European phenomenon per se, but has
become a widespread “pathology” in both Western and Eastern Europe.
In Appendix C.1, we provide stylized facts for five types of countries, namely Western European
countries (France, Germany and the UK), other old members of the European Union countries char-
acterized by rising votes for radical parties (Austria, Greece and Italy), Eastern European countries
(Czech Republic, Hungary and Poland), traditional settlement countries (Australia, Canada and the
U.S.), and Latin American countries (Argentina, Chile and Mexico). We point to large variations
in the mean populism across elections in many countries (such as Austria, Italy, Hungary, Poland,
Australia, Mexico, etc.). These are the sources of variation that we will use in the next section to
assess the effect of globalization on populism.

Trends in the presence and success of populist parties. – We now account for the
dichotomous classification of parties and focus on the presence and electoral success of populist parties,
defined as in the previous section as those with a populism score exceeding the one standard deviation
threshold (1pi,e,t = 1). Stylized facts are presented in Figure 4.
Panels (a) and (b) illustrate the increasing presence of populist parties in political elections. Panel
(a) shows the evolution of the average number of populist parties per election, conditional on obtaining
one seat (to be part of our sample). The total number of populist parties in the 55 countries included

18

Electronic copy available at: https://ssrn.com/abstract=4273442

in our sample has increased steadily since the sixties, with peaks observed in the late seventies, mid-
nineties and in the recent years. This suggests that changes in the mean level of populism highlighted
in the bottom panel of Figure 3 have been governed, at least partly, by changes in the number of
populist parties. The trends are similar in European and non-European countries, except for the
recent years. The last peak is clearly determined by the rising number of populist parties in the
European Union. As a corollary, Panel (b) shows that the share of elections with a least one populist
party has also increased steadily since the early nineties. Populist parties are present in about 55
percent of contemporaneous elections, and in more than 70 percent of European elections.
Turning our attention to the success of populist parties, we define the volume margin of pop-
ulism as the vote share of populist parties, and compute it at the aggregate level as:
PI PP p p
i=1 p=1 1i,e,t πi,e,t
ΠVe,t = PI PP p , (3)
i=1 p=1 πi,e,t

p
where, as before, πi,e,t denotes the vote share. Note that the volume margin can also be computed at
the level of each country (ΠVi,e,t ) by removing the summation over i in the above equation. The latter
variable is also used as a dependent in the regression analysis.
Panel (c) depicts the evolution of the volume margin of populism over time. The evolution of the
volume margin is by and large similar to that of the mean margin, suggesting again that changes in
the mean margin have been strongly governed by the number and electoral success of populist parties.
Stylized facts for five types of countries are provided in Appendix C.1. Variations in the volume of
populism are way greater than variations in the mean margin. This is due to the fact that parties
frequently enter or exit the populist group either by changing their political discourses, or by exiting
or entering our sample (remember that our sample only includes countries with at least one seat in
the Parliament). Hence, in line with the trade literature, changes in the volume of populism (total
share of votes won by populist parties) can be studied along the extensive margin (number of populist
parties running for election) and the intensive margin (average share of votes won by each populist
party). Changes in the extensive margin are illustrated in Panel (a) and appear stronger than those
identified in the volume margin of populism.
Other variables of interest are the average populism score of populist parties (Sikk, 2009) and its
difference with the score of traditional/non-populist ones (Inglehart and Norris, 2016). In Panel (d),
we compute the average populism score of traditional or non-populist parties (gray crosses), and of
populist parties (black diamonds). The figure shows that the populism score of traditional parties has
been rather stable over time. As far as populist parties are concerned, their average score has had its
ups and downs. Before the year 2000, the mean score of populist parties was negatively correlated with
the volume margin of populism. This can be due to the fact that more parties becomes “moderately”
populist (changes along the extensive margin) or that “moderately” populist parties start obtaining
seats when there is a window of opportunity for sanction votes (i.e., times of crisis). As a mirror
effect, the score of traditional parties decreases in these periods. Perhaps more worrisome is that the

19

Electronic copy available at: https://ssrn.com/abstract=4273442

 Figure 4: Stylized facts II – Presence, electoral success and score of populist parties

70
1.4

60
Election with populist party (%)
1.2
# Populist Parties/# Elections

50
1

40
.8

30
.6

20
.4

10
.2
0

0
1960 1970 1980 1990 2000 2010 2020 1960 1970 1980 1990 2000 2010 2020

Full sample EU28 RoW Full sample EU28 RoW

(a) Number of populist parties per elections (b) Elections with a populist party (%)
25

2015

2010
Votes gained by populist parties (%)
20

2000
15

1990
10

1980

1970
5

1963

-.5 0 .5 1 1.5 2
0

1960 1970 1980 1990 2000 2010 2020 Populism index

Full sample EU28 RoW Not populist party Populist party

(c) Volume margin of populism (ΠVi,e,t ) (d) Mean score of populist vs. non populist

Notes: Fig. (a) shows the total number of populist parties per elections. Fig. (b) gives the percentage
of elections with at least a Populist party. Fig. (c) depicts the average share of votes for populist parties
(the volume margin). Fig. (d) presents the average populism score of populist and non populist parties.
Populist parties are defined as those with a score exceeding 1 standard deviation (0.81). Fig. (a), (d),
(e) and (f) show moving averages including 3 years before and 3 years after each date. The vertical lines
indicate shifts in our sample size: inclusion of Greece, Portugal and Spain around 1975, and inclusion of
Latin American and former soviet union countries around 1990. Similar trends are obtained in the balanced
sample (see Fig. B-VI in Appendix B.5).

correlation between the mean score of populist parties and the volume of populism has turned positive
after 2005. The gap with traditional parties has widened since then, which is in line with the recent
evolution of the within-country component of inequality illustrated in Figure 3.

20

Electronic copy available at: https://ssrn.com/abstract=4273442

Trends in left-wing vs. right-wing populism. We finally decompose the trends above along
the left-right spectrum. Remember that we position parties over the left-right political scale using the
rile index available in MPD (Budge and Laver, 2016), and we consider parties as left-wing, centrist or
right-wing when their left-right index belongs to the first, second or third tercile of the distribution,
respectively. We combine this with our dichotomous classification of populist parties and identify the
extent of left-wing populism (often associated with radical left parties), right-wing populism (often
associated with far-right parties), and the residual category of centrist populism. Stylized facts are
depicted in Figure 5.
When aggregating all countries, Panel (a) shows that the recent rise in the number of populist
parties (extensive margin) is driven by parties belonging to the centre and left-wing terciles of the
distribution. This might be surprising at first glance, but it is worth reminding that the political
success of parties (i.e., their vote share) is not taken into account at this stage. The number of
right-wing populism increased drastically between the second half of the eighties and the early 2000s.
Panel (b) shows that the average populism score of left-wing populist parties has decreased since
the financial crisis of 2008 (it reaches 1.4 – i.e., 1.75 standard deviations in 2018). On the contrary, the
average populism score of right-wing populist parties has increased since 2005 (it reaches 1.7 – i.e., 2.1
standard deviations in 2018). This suggests that the financial crisis of 2008 and the resulting economic
inequalities have probably allowed a return to more authoritarian positions towards established elites,
open markets, and protection of minorities. For the first time since the sixties, radical-right populist
leaders are more populist than the radical-left ones.
Panels (c) and (d) compare the trends observed in the European Union and in the rest of the world.
On the one hand, after a sharp decline between the mid-seventies (oil crisis) and the early nineties,
the share of elections with at least one left-wing populist party has steadily increased in all regions of
the world (from 15 to 30 percent), as shown in Panel (c). On the other hand, Panel (d) shows that
the share of elections with at least one right-wing populist party has increased from 5 to more than
50 percent in European Union member states. In the rest of the world, this share right-wing populist
party has increased from 10 to 25 percent over the same period; with a sharp decline during the last
wave of elections. Once more, this evidences an increased “supply” of right-wing populism in Europe
over the last two decades. In Appendix C.3, we show that these changes are even more pronounced
in the core members of the European Union (EU15).

4 Links with Globalization

Previous literature has looked at the determinants of the volume margin of populism and has identified
several important determinants to its recent rise. First, the perception of economic insecurity and
increased inequality is one of the main drivers of the rising demand for populism (Inglehart and Norris,
2016; Guiso et al., 2017; Rodrik, 2018, 2021); economic fears are sometimes linked to automation and
de-industrialization shocks (Frey et al., 2018; Anelli et al., 2018; Gallego et al., 2018), or to severe
economic and financial crises (Funke et al., 2016; De Bromhead et al., 2013; Algan et al., 2017).

21

Electronic copy available at: https://ssrn.com/abstract=4273442

 Figure 5: Stylized facts III – Left-wing and right-wing populism at the aggregate level

2.2
15
Total nb. of populist parties

2
Average populism index
10

1.8
1.6
5

1.4
1.2
0

1960 1970 1980 1990 2000 2010 2020 1960 1970 1980 1990 2000 2010 2020

Total RW LW Centre Populist parties RW LW

(a) Number of populist parties (b) Average score of populist parties

40

50
Election with RW populist party (%)
Election with LW populist party (%)

40
30

30
20

20
10

10
0

1960 1970 1980 1990 2000 2010 2020 1960 1970 1980 1990 2000 2010 2020

Full sample EU28 RoW Full sample EU28 RoW

(c) Elections with a LW populist party (%) (d) Elections with a RW populist party (%)

Notes: Fig. (a) shows the total number of populist parties, dividing between left-wing and right wing.
Fig. (b) presents the average populism score of populist parties, splitting between left-wing and right-wing
parties. Fig. (c) and (d) give the percentage of elections with at least a left-wing and right-wing Populist
party, respectively. Populist parties are defined as those with a score exceeding 1 standard deviation (0.808),
while left-wing and right-wing parties are defined as those that belongs to the first and third tercile of the
left-to-right index. Fig. (a), (b), (c) and (d) show moving averages including 3 years before and 3 years after
each date. The vertical lines indicate shifts in our sample size: inclusion of Greece, Portugal and Spain
around 1975, and inclusion of Latin American and former soviet union countries around 1990. Similar
trends are obtained in the balanced sample (see Fig. B-VII in Appendix B.5).

Second, populism is also associated with the perception that the elites are neglecting the risk of social
conflicts as well as with a perception of lost identity, or cultural dissolution (Norris and Inglehart,
2019; Mukand and Rodrik, 2018; Algan et al., 2018). In addition, the recent rise of populism also

22

Electronic copy available at: https://ssrn.com/abstract=4273442

relates to the expansion of internet and new media (Zhuravskaya et al., 2020; Campante et al., 2018;
Guriev et al., 2019).
While all the above mentioned studies somehow relate to globalization, other studies have focused
explicitly on trade and migration, noting that the associated overall income gains may be distributed
very unevenly. The ”losers from globalization” (i.e., the socially and economically downgraded seg-
ments of the workforce) are then likely to join the ranks increasing the support base of populist parties
(Autor et al., 2013, 2020; Helpman et al., 2017; Colantone and Stanig, 2018; Hays et al., 2019; Colan-
tone et al., 2021). Theory suggests that the distributional consequences of globalization are governed
by the skill structure of immigration and imports, whose roles have been investigated separately and
in very few studies only (Edo et al., 2019; Moriconi et al., 2022, 2019; Autor et al., 2020; Mayda et al.,
2022). From the cultural perspective, rising immigration produced a cultural backlash and a stronger
support for political ideas oriented on the cleavage between “good natives” and “bad foreigners” (Halla
et al., 2017; Moriconi et al., 2022; Shehaj et al., 2019).
This section focuses on the empirical relationship between the margins of populism, and the size
and structure of immigration and imports. Compared to existing works, our empirical analysis brings
four main innovations. First, we conduct a unified analysis of the effect of immigration and import
shocks that accounts for their size and their skill-specific structure. Second, we provide cross-country
evidence on the populist responses to globalization shocks in a long-term panel setting that covers 55
countries, 628 elections, and a 60-year span. Third, we quantify the effect of globalization not only
on the volume margin of populism but also on its mean margin, which captures the average “extent”
of populism that voters are confronted with after the election. Fourth, we distinguish between the
left-wing and right-wing populist responses to the size and structure of globalization shocks.

4.1 Empirical Strategy

Our empirical approach aims to quantify the effect of economic, cultural, communication, and glob-
alization factors on the evolution of the volume of populism (ΠVi,e,t defined in Eq. (3)), as proxied by
the share of votes for populist parties, and on the evolution of the mean margin of populism (ΠM
i,e,t
defined in Eq. (2)), as proxied by the weighted average populism score of all parties having obtained
at least one seat in a given election.15

Baseline specification. – Similarly to Guriev and Papaioannou (2021), we consider the following
specification for both margins of populism:

Πm S S


i,e,t = F Xi,e,t , Migi,e,t , Impi,e,t , (4)

where m = (V, M ) is the margin of populism. MigSi,e,t measures skill-specific inflow of immigrants
expressed as the mean of the percentage of the population in the election year t and in the previous
15
In the Appendix, we decompose the volume margin into its extensive and intensive margins (denoted by
ΠE
i,e,t and ΠIi,e,t , respectively), and analyze their specific determinants.

23

Electronic copy available at: https://ssrn.com/abstract=4273442

year t − 1 (with S = HS for the high-skilled and S = LS for the low-skill); and ImpSi,e,t measures skill-
specific imports expressed as the mean percentage of GDP in years t and t − 1. We also include Xi,e,t ,
a vector of traditional determinants of populism, which includes GDP per capita, human capital, the
employment rate, the size of the population, and the number of parties in an election, all of them in
logarithms. We remain parsimonious in our baseline specification but experiment with richer sets of
covariates in our robustness checks, such as voter turnout,16 skill-specific exports and emigration, or
the electoral system, in Appendix D. All our results, however, are robust to including these additional
controls.
The specification of the F-function differs according to the dependent variable. The mean margin
is a continuous variable that, given our normalization procedure, can take both negative and positive
values. For this reason, our baseline model assumes that ΠM
i,e,t is a linear function of the globalization
variables. On the contrary, the volume margin is a continuous variable that takes non-negative values
only, exhibits a high level of heteroskedasticity, and includes a non negligible share of zeroes (about 60%
in the full sample). We estimate it with the Poisson pseudo maximum likelihood (hereafter PPML)
estimator, which is found to perform better under various heteroskedasticity patterns, large number of
zeroes and rounding errors for the dependent variable (Santos Silva and Tenreyro, 2006, 2010). Hence,
our baseline model assumes that ΠVi,e,t is an exponential function of the logged transformation of the
globalization variables.

Econometric issues. – Three additional issues might lead the OLS/PPML standard models to
generate inconsistent estimates. First, the margins of populism can be influenced by a large number of
observable and unobservable determinants. Second, the relationship between populism and globaliza-
tion is potentially influenced by mismeasurement problems and reverse causality, as populist parties
tend to support anti-globalization policies. Hence, OLS/PPML estimates for the globalization terms
can underestimate the causal impact of globalization on populism, thus calling for an instrumental
approach. Third, the effect of globalization shocks can be amplified under adverse economic condi-
tions, when social media networks are expanding, or when the cultural diversity embedded in foreign
goods/people increases. We address these issues sequentially, within the limits of our cross-country
setting.
We first mitigate unobserved heterogeneity concerns by saturating the model with a full set of coun-
try and year fixed effects, which allow to account for time-invariant unobservable factors and common
trends. Assuming that all drivers of populism act in an additive way, our baseline specifications of

16
Guiso et al. (2017, 2020) show that economic insecurity depresses voting turnout in a selected manner,
and increases the share of (participating) electors voting for a populist party. Leininger and Meijers (2020)
find that the presence of populist parties (both left and right) in an election increases political participation of
citizens. Hence, drivers of turnout potentially influence populist vote shares, and voting turnout could respond
to globalization shocks. As shown in Appendix D.5, our results are robust to the inclusion of turnout as control;
moreover, we show that turnout is not significantly impacted by our measures of globalization.

24

Electronic copy available at: https://ssrn.com/abstract=4273442

Eq. (4) writes as:

S S
ΠM M M
P M P M
i,e,t = α + β Xi,e,t + S γS Migi,e,t + S ζS Impi,e,t




M M M
 +θi + θt + i,e,t ,



(5)

ΠVi,e,t = exp[αV + β V Xi,e,t + S γSV log(MigSi,e,t ) + S ζSV log(ImpSi,e,t )

 P P



 +θV + θV + V ].

i t i,e,t

where β m is a set of coefficients associated with the traditional determinants of populism included
in our Xi,e,t vector already described in equation (4) above, γSm is a pair of coefficients associated
with skill-specific immigration shocks, ζSm is a pair of coefficients associated with skill-specific import
shocks, (θim , θtm ) is a set of country and year fixed effects; and m
i,e,t is the error term. Coefficients of
the mean margin model are simple incidence parameters, whereas coefficients of the volume margin
model must be interpreted as elasticities.
Second, our baseline specification allows to identify an association between globalization shocks
and populism, without necessarily capturing a causal relationship between them. Causation is always
hard to establish with aggregate data. As detailed in Section 4.3, we rely on instrumental variables
and two-stage least squares (2SLS) techniques to mitigate endogeneity concerns. Starting from the
linear OLS specification of the mean-margin model, we can use the standard 2SLS estimator and
instrument all globalization terms jointly. In line with Frankel and Romer (1999), Munshi (2003) or
Autor et al. (2020), our instrumentation strategy relies on a “zero-stage” gravity model that predicts
the bilateral and skill structure of imports and immigration using dyadic and origin-specific factors
(destination-specific factors are excluded). We then aggregate these dyadic predicted flows for each
destination, and use these skill-specific sums (less prone to endogeneity concerns) as instruments for
observed globalization variables. With regard to the volume of populism, implementing a standard
IV approach can induce an additional bias due to the incidental parameter problem. This is due to
the non-linear structure of the PPML model and to the presence of a large number of fixed effects
(Lancaster, 2000). For the volume margin, we follow Angrist and Pischke (2008) and compare our
PPML results with those of a reduced-form IV approach, which consists in replacing actual import
and immigration flows with predicted ones.
Third, in Section 4.4, we conduct a series of robustness checks and analyze whether the base-
line results hold when considering sub-samples of countries and years, alternative lag structures for
measuring globalization shocks, and alternative thresholds used to define populist parties.
Finally, the estimation of Eq. (5) sheds light on the average effect of skill-specific globalization
shocks on populism. In Section 4.5, we supplement Eq. (5) with interaction terms between globaliza-
tion shocks and a subset of potential amplifiers of the magnitude of populist responses to skill-specific

25

Electronic copy available at: https://ssrn.com/abstract=4273442

globalization shocks (denoted by Xi,e,t ). Our extended specifications writes as:

S S
ΠM M M
P M P M
i,e,t = α + β Xi,e,t + S γS0 Migi,e,t + S γS1 Migi,e,t × Xi,e,t



 P M S P M S V V V
 + S ζS0 Impi,e,t + S ζS1 Impi,e,t × Xi,e,t + θi + θt + i,e,t ,



(6)

ΠVi,e,t = exp[αV + β V Xi,e,t + S γS0 log(MigSi,e,t ) + S γS1 log(MigSi,e,t ) × Xi,e,t

 P V P V


 P V S P V S V V V
S ζS1 log(Impi,e,t ) × Xi,e,t + θi + θt + i,e,t ].

 +
S ζS0 log(Impi,e,t ) +

The set of amplifiers Xi,e,t includes a dummy equal to one if the country experienced a year of
negative real growth since the previous election as well as proxies for de-industrialization, and dummies
capturing high levels of diversity in the origin mix of imported goods and of immigrants (proxies for
cultural diversity embedded in goods or in people), and high levels of internet expansion. Additional
interactions are considered in Appendix D.

Data. – Annual trade data are obtained from Feenstra et al. (2005) for the years 1962-2000 and
from the United Nations Comtrade database for the years 2001-2015. We extract the series of annual
imports for each country, and we split them by type of goods using the Standard International Trade
Classification (SITC) described in the Trade and Development Report (2002). Product categories at
the 3-digit level are classified on the basis of their technological complexity, capital and skill intensities.
Five categories are distinguished, namely primary commodities, labor-intensive and resource-based
manufacturing goods, and manufacturing goods with high intensities in low-, medium-, and high-
skilled labor and technology. In our baseline regressions, we only account for the divide between
manufacturing goods that are intensive in low-skilled and high-skilled labor (in short: “low-skilled
goods” and “high-skilled goods”). We experiment with different treatment of labor-intensive and of
medium-skilled manufacturing in the robustness section.
Data on 5-year migration inflows by country of destination for the same period are obtained from
Abel (2018).17 We combine these data with information about the skill level of the stock of migrant
population from each origin in each country for a few census rounds (say, 1990, 2000 and 2010). We
then compute a skill-selection index, proxied by the ratio of college graduates in the dyadic migration
stock to the one in the native (pre-migration) population. We use this ratio in the closest available
year to impute a skill level for the immigration flows.18
In terms of our set of controls, GDP per capita is computed as the GDP at constant 2011 national
prices (in million 2011 USD) divided by the population (in millions), both taken from the Penn World
Table. The human capital and the employment rate is the ratio of employed to the working-age
population, also come from the Penn World Table. The number of years since the last election and
the number of parties in each election are from the MPD. Table 3 provides descriptive statistics of our
main controls and variables of interest.

17
We interpolate the 5-year data to get annual migration flows over the time period.
18
Some aggregate and country-specific stylized facts are provided in Appendix C.2.

26

Electronic copy available at: https://ssrn.com/abstract=4273442

 Table 3: Summary Statistics - 55 Countries, 1960-2018

Variable Mean S.D. Min. Max. Obs. Pc(25) Pc(50) Pc(75)

PANEL A - Populism Vars.
Volume Margin (All) 8.94 16.82 0.00 92.18 592 0.00 0.00 10.08
Volume Margin (RW) 4.81 12.37 0.00 84.73 592 0.00 0.00 0.00
Volume Margin (LW) 2.53 8.07 0.00 87.33 592 0.00 0.00 0.00
Mean Margin (All) -0.07 0.44 -1.15 1.94 592 -0.35 -0.14 0.13
Mean Margin (RW) 0.03 0.28 -1.06 1.51 472 -0.12 -0.00 0.11
Mean Margin (LW) -0.04 0.24 -0.83 1.74 479 -0.16 -0.03 0.04

PANEL B - Globalization Vars.

log Imp (LS) -3.40 0.97 -7.55 -1.28 578 -4.00 -3.23 -2.69
log Imp (HS) -2.40 0.88 -5.71 -0.10 578 -2.86 -2.28 -1.83
log Mig (LS) -4.13 1.34 -11.71 -1.50 584 -4.77 -3.86 -3.22
log Mig (HS) -5.51 1.48 -15.11 -2.79 584 -6.26 -5.42 -4.58
Imp (LS) 0.05 0.04 0.00 0.28 578 0.02 0.04 0.07
Imp (HS) 0.13 0.11 0.00 0.91 578 0.06 0.10 0.16
Mig (LS) 0.03 0.03 0.00 0.22 587 0.01 0.02 0.04
Mig (HS) 0.01 0.01 0.00 0.06 587 0.00 0.00 0.01

PANEL C - Country Control Vars.

log GDP/capita -3.91 0.60 -6.36 -2.47 592 -4.25 -3.86 -3.49
log Pop 16.18 1.52 12.08 19.55 592 15.24 16.05 17.47
log HC 1.05 0.18 0.21 1.32 592 0.97 1.08 1.17
log Emp/Pop -0.42 0.17 -1.22 0.15 592 -0.51 -0.40 -0.30
log Parties 1.72 0.45 0.00 2.89 592 1.39 1.79 2.08

4.2 Baseline Empirical Results

Tables 4 provides estimates of our baseline PPML and OLS models as depicted in Eq. (5), in which all
potential drivers of populism act in an additive way, and skill-specific levels of imports and immigration
are included jointly. The left panel of Table 4 focuses on the volume margin of populism,19 while the
right panel shows the results for the mean margin of populism.
Despite potential collinearity issues, we account for GDP per capita and employment rates as well
as for human capital. We confirm that in general, higher levels of human capital tend to reduce the
volume and mean margins of populism. The coefficient of GDP per capita is usually insignificant,
except for the volume of right-wing populism. This seems to be inconsistent with our stylized facts,
which show that all margins of populism increase in times of crisis. As global crises affect all countries
in our sample in a potentially non-linear way, their role is likely to be captured by the year fixed
effects. Figure 6 plots the year fixed effects estimated for the volume margin (diamonds) and for the
mean margin (circles) of populism, as well as their moving average. We observe a positive trend for
both margins, and even more so during the first half of the seventies, in the first half of the nineties,
and in the years after 2008 (Funke et al., 2016; De Bromhead et al., 2013; Algan et al., 2017). Other
control variables tend to be insignificant.
19
In Appendix D.2, we decompose these effects along the extensive and intensive margins

27

Electronic copy available at: https://ssrn.com/abstract=4273442

 Figure 6: Time fixed effects for the volume and mean margins of populism

.4
.2
0

0
-.2
-2

-.4
-.6
-4

1960 1970 1980 1990 2000 2010 2020

Volume margin Mean margin

Notes: Red diamonds (left scale) and blue circles (right scale) represent the year fixed estimated from Eq. (5).
The red solid line and blue dashed line are the centered moving average computed over 5 years over the times
fixed effects estimated for the volume margin and the mean margin, respectively.

In line with the existing literature, imports of low-skill intensive goods are positively and sig-
nificantly associated with total, right-wing, and left-wing populism. On the contrary, imports of
high-skill intensive goods are associated with lower volumes of populism in general, and with lower
levels of right-wing populism in particular.
With regard to immigration, its association with the overall volume of populism is insignificant.
Our results support, however, a substitution between left-wing and right-wing populism. Low-skilled
immigration is associated with higher volumes of right-wing populism and with smaller volumes of
left-wing populism. Again, high-skilled immigration tends to generate substitution from right-wing to
left-wing populism, although the coefficients are slightly smaller and less significant.
These results are in line with Autor et al. (2020), Edo et al. (2019) or Moriconi et al. (2022, 2019),
however these papers did not consider trade and immigration jointly. We find that the skill structure
of globalization shocks matters for the volume of populism. Both changes to import and immigration
are associated with a more than proportionate change in the volume of right-wing populism, but only
when these changes are ”low-skill”. By contrast, shocks that are intensive in high-skilled labor reduce
the volume of right-wing populism, both for trade and for immigration.
Nonetheless, the analysis of the volume margin fails to capture the effect of globalization shocks
on the actual “extent” of populism which voters are exposed to during an election. The right panel

28

Electronic copy available at: https://ssrn.com/abstract=4273442

 Table 4: Baseline PPML and OLS results – Volume and Mean Margins

Volume (ΠVi,e,t ) Mean margin (ΠM

i,e,t )
All RW LW All RW LW
(1) (2) (3) (4) (5) (6)
log GDP/capit -1.22 -2.46∗∗ 0.70 -0.15 0.03 0.03
(0.95) (1.19) (1.38) (0.21) (0.12) (0.16)
log Popit 1.28 1.00 2.98 0.30 0.48∗ 0.04
(0.96) (1.33) (1.84) (0.23) (0.25) (0.19)
log HCit -4.81∗∗ -9.01∗∗∗ 5.06 -1.74∗∗∗ -1.85∗∗∗ -0.04
(2.09) (3.41) (5.27) (0.54) (0.54) (0.37)
log Empit /Popit -0.98 -0.15 -5.00 -0.21 -0.05 -0.06
(1.46) (1.99) (3.65) (0.23) (0.19) (0.19)
log Partiesit 0.45 0.51 0.83∗ 0.09 -0.05 0.09
(0.29) (0.50) (0.43) (0.06) (0.06) (0.05)

log Impi,t (LS) 0.83∗∗∗ 1.33∗∗ 1.49∗∗

(0.30) (0.56) (0.62)
log Impi,t (HS) -0.71 -1.30∗∗∗ -1.25
(0.44) (0.49) (0.86)
log Migi,t (LS) 0.14 1.52∗∗∗ -1.78∗∗∗
(0.34) (0.55) (0.59)
log Migi,t (HS) -0.28 -1.32∗∗∗ 1.17∗
(0.29) (0.48) (0.64)
Impi,t (LS) 3.78∗∗ 4.28∗∗∗ -0.11
(1.65) (1.47) (0.70)
Impi,t (HS) -0.21 -0.50∗ 0.36
(0.43) (0.28) (0.23)
Migi,t (LS) -0.17 1.73 -1.28
(1.93) (2.45) (1.28)
Migi,t (HS) 1.86 -2.63 3.65
(4.99) (4.74) (3.49)
Observations 575 575 575 578 461 470
Pseudo-R2 0.40 0.37 0.51
R2 0.50 0.41 0.48
Year FE 3 3 3 3 3 3
Country FE 3 3 3 3 3 3
Notes: ***, **, and * denote significance at the 1, 5, and 10 percent levels, respectively; clus-
tered standard errors at the country level are reported in parentheses; coefficients presented
in column (1) to (3) have been estimated with PPML using the Stata command ppmlhdfe,
while coefficients in column (4) to (6) have been estimated with OLS using the Stata com-
mand reghdfe

of Table 4 focuses on the association between globalization and the mean margin of populism.20 The
20
When we distinguish between left- and right-wing populism, the number of observations decreases. This
is because our sample includes some elections without parties belonging to the first or latter tercile of the
left-to-right index distribution.
29

Electronic copy available at: https://ssrn.com/abstract=4273442

mean margin accounts for the level of populism of all parties (classified as populist or non-populist)
running for election at time t, as well as for their vote shares.21 We find that imports of low-skill
labor intensive goods are positively and significantly associated with the mean margin of total and
right-wing populism. The coefficient is around 4, which means that a 1 percentage point change in
the import rate of goods which are intensive in low-skill labor is associated with a 0.04 increase in the
mean margin of populism. On the contrary, imports of goods which are intensive in high-skill labor,
as well as high-skilled immigration, are not significantly correlated with the mean margin of populism.
The combined analysis of the volume and mean margins as well as the supplementary results
provided in Appendix D allow us to better understand the mechanisms at work. In Appendix D.2, we
decompose the volume margin into its extensive (number of populist parties) and intensive (vote share
per populist party) components. In Appendix D.3, we divide our parties into two groups – those who
have never been classified as populist, and those who have been classified at least once as populist
(including potential switchers) – and we estimate the links between globalization shocks and the mean
populism score within these two groups.
Imports of low-skilled goods are associated with an increase in the share of votes for centrist and
right-wing populist parties (volume margin) and in the average post-election level of centrist and right-
wing populism (mean margin). Our decomposition suggests that the volume-margin effect operates
along the intensive margin, and the mean-margin effect is jointly governed by the rising vote share for
populist parties, and by an increase in the populism score of centrist populist parties.
In contrast, low-skill immigration is associated with a transfer of votes from left-wing to right-wing
populist parties, without impacting the total volume of populism or the average “extent” of populism
(mean margin). The decomposition suggests that these changes operate along the extensive margin of
right- and left-wing populism, and are concomitant with a decrease in the mean level of populism of all
types of parties. The most likely hypothesis is that low-skill immigration encourages new right-wing
populist parties with moderate populism scores to run for election, or allows them to gain at least
one seat in the election. Furthermore, it is worth emphasising that low-skilled intensive imports and
immigration never increase the mean populism score of traditional (i.e., never populist) parties.

4.3 Regressions with Instrumental Variables

The correlations presented in the previous section can be driven by unobserved common determinants
of globalization and populism and suffer from reverse causation problems. In particular, we may expect
that a rise in populism translates into greater restrictions on trade and immigration, implying that
the estimates in Tables 4 might underestimate the causal impact of globalization shocks on populism.
To mitigate such endogeneity concerns, we use an instrumental variable approach (IV, hereafter) with
instruments pertaining to the origin country (for both trade and migration flows). Following Autor
21
In MPD data, the cumulative vote share is less than 100% for many election-year pairs. This is because
small parties and most independent candidates running for election failed to obtain a seat and are excluded from
the sample. In the last three columns, we normalize the vote shares of parties represented in the parliament so
that their sum is equal to 100%. In Appendix D.3, we show that our results are robust to this normalization.

30

Electronic copy available at: https://ssrn.com/abstract=4273442

et al. (2013, 2016), the “China shock” has been abundantly exploited in the trade literature as a source
of exogenous variations in imports and exports in the partner countries. Similary, following Munshi
(2003), push factors of origin countries have been frequently used to instrument immigration shocks
in the destination country (Boustan, 2010; Klemans and Magruder, 2018; Monras, 2020).
We generalize this approach by predicting dyadic flows of goods and migrants between countries
relying on origin countries’ time-varying characteristics and time-invariant dyadic factors. We then
aggregate these flows by destination, and use the aggregate predictions as instruments for skill-specific
imports and immigration flows. Hence, our IV strategy relies on a “zero-stage” gravity-model for
dyadic trade and migration (Frankel and Romer, 1999; Feyrer, 2019; Alesina et al., 2016; Docquier
et al., 2020), which writes as:

Yij,t = exp [α + θij ∗ P ost1990 + θj,t + ij,t ] , (7)

where Yij,t is the dyadic skill-specific flow of either imported goods (ImpSi,t ) or immigrants (MigSi,t )
from origin country j to destination country i at year t.22
Our zero-stage regression in Eq. (7) includes a set of fixed effects. We have dyadic fixed effects (θij )
capturing bilateral determinants such as distance, colonial linkages, cultural and linguistic proximity,
as well as time-invariant destination-specific characteristics. Remember that in our second stage, we
control for country fixed effects and identify the effect of globalization shocks using the within-variation
in imports and immigration. Dyadic fixed effects are interacted with a post-1990 dummy (P ost1990 ),
which proxies structural changes due to the fall of the Berlin Wall (including political transformations
in Eastern European countries and greater intra-European labor mobility). We also have origin-year
fixed effects (θj,t ) capturing time-varying shocks in the origin country (e.g., changes in trade policies,
economic shocks, socio-demographic changes, conflicts, natural disasters, etc.). Given the large number
of zeroes in dyadic flows, we estimate Eq. (7) using PPML, which explains the exp transformation of
the right-hand-side term. We estimate Eq. (7) over the global matrix of destination-origin countries.
We predict skill-specific trade and migration flows, Ybij,t using the estimated coefficients from
P b
Eq. (7), then aggregate them using Ybi,t ≡ j Yij,t , and use Ybi,t as an external instrument for Yi,t
in the model for the mean margin. Being estimated from the gravity model without time-varying
destination-country characteristics, the predicted flows should be less prone to reverse causation and
omitted variable biases. When focusing on the volume margin, we use a reduced-form IV approach
and replace the actual flows (Yi,t ) by the predicted ones (Ybi,t ) in the PPML setting, as recommended
by Angrist and Pischke (2008). First-stage regression results are provided in Table D-I, in Appendix
D.1.23

22
The dependent variable for trade is skill-specific, i.e., Yij,t refers either to low- or high-skill import flows. For
migration, we use total flows and we rely on the strategy used in the baseline to derive skill-specific immigration
flows.
23
The predicted levels are nicely correlated with the actual ones, and the coefficients of the instruments are
highly significant close to unity. The adjusted R-squared is usually large despite the fact that our zero-stage
dyadic regressions abstract from destination-time characteristics.

31

Electronic copy available at: https://ssrn.com/abstract=4273442

 Table 5: Reduced-form IV PPML and 2SLS results – Volume and Mean Margins

Volume (ΠVi,e,t ) Mean margin (ΠM

i,e,t )
All RW LW All RW LW
(1) (2) (3) (4) (5) (6)
log Imp
d i,t (LS) 0.91* 1.82** 0.97
(0.50) (0.84) (0.84)
log Imp
d i,t (HS) -1.22* -2.14** -0.72
(0.66) (0.87) (0.83)
log Mig
d i,t (LS) 0.53 1.97*** -1.70*
(0.43) (0.58) (0.92)
log Mig
d i,t (HS) -1.04* -2.02** 0.60
(0.56) (0.89) (1.23)
Impi,t (LS) 4.99** 4.06** 1.29
(2.33) (1.77) (1.42)
Impi,t (HS) -0.22 -0.59 0.45
(0.54) (0.38) (0.37)
Migi,t (LS) 0.52 0.74 -0.75
(3.13) (3.01) (1.53)
Migi,t (HS) 0.99 3.15 3.34
(10.12) (7.90) (4.75)
Observations 575 575 575 578 461 470
Pseudo-R2 0.40 0.36 0.50
R2 0.06 0.09 0.01
K-Paap F-stat 12.05 11.36 9.45
Year & Country FE 3 3 3 3 3 3
Controls 3 3 3 3 3 3
Notes: ***, **, and * denote significance at the 1, 5, and 10 percent levels, respectively; clus-
tered standard errors at the country level are reported in parentheses; coefficients presented
in column (1) to (3) have been estimated with PPML using the Stata command ppmlhdfe and
predicted globalization variables from the model estimated in equation (7), while coefficients
in column (4) to (6) have been estimated with 2SLS using the Stata command ivreghdfe.

The results for the globalization variables are presented in Table 5. The left panel provides reduced-
form IV estimates for the volume margin of populism. These estimates are very much in line with
the results of our baseline PPML regressions. They confirm that the skill structure of globalization
shocks plays a key role. Imports of low-skill goods foster votes for populist parties in general, and
for right-wing parties in particular. By contrast, imports of high-skill goods decrease the votes for
right-wing populist parties. With regard to immigration, the IV results also confirm those of the
baseline regressions. Low-skill immigration leads to a substitution of left-wing populism for right-wing
populism. High-skill immigration reduces the votes for populist parties. Compared to the baseline
regressions, the elasticities are larger by a factor of 1.3, which is in line with the existence of a

32

Electronic copy available at: https://ssrn.com/abstract=4273442

downward-sloping reverse causation link.
The right panel provides the 2SLS estimates for the mean margin of populism. These estimates
are also in line with the OLS results of Table 4. Imports of low-skill intensive goods tend to increase
the mean margin of total and right-wing populism. On the contrary, imports of high-skill goods and
both types of immigration do not lead to such populist responses. The coefficients are in the same
order of magnitude as in the OLS setting. As for the strength of the instrument, the Kleibergen-Paap
F-stat is around 10 across the different specifications, which is a reasonable value given the fact that
we are instrumenting four different endogenous variables simultaneously. Reassuringly, our results are
preserved and Kleibergen-Paap F-stat are much larger when instrumenting one variable at a time or
in pairs (see Appendix D.4).
Instrumental variable techniques are also used in the decomposition of the volume and mean
margins of populism (see Appendix D.2). The IV results tend to reinforce the mechanisms highlighted
in the previous section. With regard to imports of low-skill intensive goods, their effect on the volume
of (centrist and right-wing) populism operates along the intensive margin, whereas their effect on
the mean margin is partly governed by a greater populism score of centrist populist parties. Low-
skill immigration, on the other hand, favors new right-wing populist parties with moderate populism
scores to run for election or to gain a seat without influencing their mean populism score. Finally,
globalization shocks have no effect on the populism score of traditional (i.e., ”never populist”) parties
(see Appendix D.3).

4.4 Robustness Checks

To investigate whether our results are sensitive to specification choices, party classification, or sub-
samples of countries and years, we conduct a battery of robustness checks using the IV estimators.
Detailed results for the volume and mean margins of populism are provided in Appendix D. We
summarize below our main findings, mostly focusing on the populism responses to imports of low-
skilled intensive goods and low-skilled immigration.

Lag structure for globalization shocks (Appendix D.8.1). In our baseline results, the skill-specific
migration and import variables are defined as the sum of import and immigration flows over two years,
namely the election year and the year prior to the election. To assess whether our results are sensitive
to the lag structure of our model, we provide results with skill-specific import and migration defined
as (i ) the flows observed in the election year (t), (ii ) the flows observed in the year before the election
(t − 1), (iii ) the flows observed two years before before the election (t − 2), (iv ) the sum of the flows
between the election year and two years before, and (v ) the sum of the flows between the last two
elections. The number of lags used to compute import and immigration shocks influences the scale
of these variables and the magnitude of the coefficients. Overall, results for immigration are highly
robust to the lag structure. The sign and significance of the result for imports is also preserved, except
when shocks at measured in the year of election (too short a period) or between two elections (too
long a period). In the vein of Rodrik (2018), a left-wing populist response to imports cannot be ruled

33

Electronic copy available at: https://ssrn.com/abstract=4273442

out when the import shock is computed over a longer period, implying some form of persistence.

Classification of populist parties (Appendix D.8.2). In our baseline results, we define populist
parties as those exhibiting a populism score above one standard deviation (η = 1.0). The choice of
this threshold maximizes the partial correlation with most existing classifications, and defines a clear-
cut bundle of parties when using unsupervised clustering algorithms. We provide results obtained
when using less restrictive (η = 0.9) or more restrictive (η = 1.1) thresholds. The significance and
magnitude of the effects are well preserved when using a lax (or more inclusive) classification of populist
parties. A few effects becomes insignificant when using a stricter (or less inclusive) definition. It is
worth emphasizing that many parties usually perceived as populist by political scientists exit the list
when using the stricter definition.24

Proxies for skill-specific immigration shocks (Appendix D.8.3). We analyze the sensitivity of our
results to the imputation of the skill structure of immigration flows, or to the inclusion of interactions
with migrant stock variables. First, we impute the skill structure of migration inflows using the
selection ratios observed in the year 2000 only, rather than relying on the closest year (1990, 2000,
2010). The results remain similar both in terms of magnitude and significance. Second, we explore
whether the populist responses to immigration flows are magnified when the pre-existing stock of
immigrants is large. To capture these non-linear effects, we interact low-skilled immigrant flows with
a dummy variables equal to unity if the ratio of immigrant stock to population in the destination
country belongs to the top or bottom quartile of the distribution in 1960. With a few exceptions,
these interaction terms are insignificant. The magnitude and significance of the direct impacts of
imports and immigration are well preserved.

Proxies for skill-specific import shocks (Appendix D.8.4). We consider alternative ways to charac-
terize the skill and technological content of imported goods. Following the classification of the Trade
and Development Report (2002), we first expand our specification by adding imports of labor-intensive
goods (both high- and low-skill labor intensive goods) to the set of regressors. This does not affect the
effect of our baseline globalization shocks. We find that labor-intensive import is positively associated
with the volume of left-wing populism. Second, we augment our specification with imports that are
medium-skill labor intensive. This variant kills the significance of the volume-margin responses to
imports, which is probably due to collinearity issues, while preserving the mean-margin responses.

Combining skill content with economic development at origin. In Section D.8.5, we consider a
more demanding specification in which our main variables of interests are now split according to the
level of economic development of the country of origin. We create dummies for low-income (LI) and
a high-income (HI) countries using the World Bank country classification (combining those defined
as low-income and lower-middle income in the LI category, whereas those considered as upper-middle
and high-income form the HI category). Replicating the baseline analysis with the above variables,
the findings highlight that the positive and significant populism responses to globalization are mostly
24
Some relevant examples of parties that are not classified as populist with the stricter definition are Syriza
in Greece, Movimento 5 Stelle in Italy, and La France Insoumise in France.

34

Electronic copy available at: https://ssrn.com/abstract=4273442

driven by imports and immigration flows of goods and people originating from low-income countries
on the volume margin. However, globalization shocks involving North-North movements seems more
relevant in explaining the mean-margin positive response.

Robustness by sub-sample. (Appendix D.8.6). Since our analysis covers a long time period and a
wide set of countries, we provide results exploring whether our baseline results are driven by specific
time-periods or subsets of countries. We first investigate whether our results are governed by more
recent years, when the pace of globalization increased. To do so, we include in our analysis interaction
terms between our populism-enhancing globalization variables (i.e., low-skilled imports and immigra-
tion) with a post-1990 dummy. Our results are highly robust to the inclusion of this additional terms,
which tend to attenuate the right-wing populist response to imports along the volume and mean
margins. We then account for the relevant presence in our sample of European countries, which are
characterized by different layers of integration among themselves, depending whether they belong to
the countries of the European Union (EU28). Hence, we include a dummy variable capturing whether
a country belongs to the European Union, and interaction with low-skill intensive globalization shocks.
While the direction of the estimates remain the same, the magnitude and the significance of the coef-
ficients is influenced by the subset of countries under analysis. The effect of imports and immigration
on the volume margin of (total and right-wing) populism is mostly driven by EU28 countries. In addi-
tion, we cannot rule out an effect of imports on the volume and mean margins of left-wing populism in
the EU. As further check, we explore whether the results are driven by the Latin American countries
available in our sample. Excluding them from the sample does not influence our estimates. Finally,
we show that our results are confirmed and are not driven by the unbalanced structure of our dataset.
By excluding from the sample countries the ones that enter in the sample after 1970, our skill-specific
estimates both on the volume and mean margin are confirmed.

Additional robustness checks. We also show that our results are not driven by an effect of global-
ization on voting turnout (see Appendix D.5). Our results also hold when controlling for the electoral
system, although a significant effect of imports of low-skilled intensive goods on left-wing populism is
obtained under the proportional representation system (see Appendix D.6). Furthermore, our results
are robust to the inclusion of skill-specific export and emigration flows (see Appendix D.7). We decided
not to include emigration and exports in our benchmark regression for three reasons. First, the effects
of exports and emigration are less significant and robust. Second, instrumenting eight skill-specific
globalization shocks is a heroic task. Third, we already account for the direct impact of emigration
on the skill structure of the labor force by controlling for human capital.

4.5 Searching for Amplifiers

The results described above can be considered as average populist responses to globalization shocks in
normal times. We now consider the extended specification depicted in Eq. (6), which includes other
potential drivers of populism (direct impact) and their interactions with low-skill intensive globalization
shocks (amplifiers).

35

Electronic copy available at: https://ssrn.com/abstract=4273442

 We create five dummies to capture whether (i) the country experienced a year of negative real
income growth in the last two years before the election (a proxy for economic crises), (ii) the country
experienced a variation in the share of manufacturing value added in GDP in the last two years that
belongs to the bottom quartile of the distribution (a proxy for de-industrialization), (iii) the share of
internet users in population belongs to the top decile of the distribution (a proxy for a high prevalence
of social media), (iv) the level of diversity in the origin mix of imports of low-skill labor intensive goods
belongs to the top decile of the distribution (a proxy for diversification in imports), and (v) the weighted
mean of genetic distance between the origin and destination countries of low-skill immigrants belongs
to the top decile of the distribution (a proxy for a high level of cultural distance between natives and
low-skill immigrants).25 Detailed IV regression results including the linear effect of the dummies are
provided in Tables D-XXXIV to D-XXXVI in Appendix D.9. The linear terms are insignificant for the
crisis and de-industrialization dummies, whose roles are likely to be captured by the year fixed effects.
The internet dummy is positive and significant for the volume margin of populism (Zhuravskaya et al.,
2020; Campante et al., 2018; Guriev et al., 2019), and virtually insignificant for the mean margin.
Finally, a high level of diversity in imports increases the mean margin of populism, while we find
insignificant direct impacts for cultural distance between natives and low-skilled immigrants.
However, our main variables of interest are the interaction terms with globalization shocks, which
reflect potential amplifiers of the populism responses to globalization. Figure 7 provides the estimated
coefficients of the interaction terms and their confidence intervals at the 90% threshold. Each sub-
figure focuses on one potential amplifier, and distinguishes between the volume margin of populism
(left panel) and the mean margin (right panel), separated by a vertical line. Each panel includes two
triplets of estimates, namely the effect of imports of low-skill labor intensive goods on the left, and
the effect of low-skill immigration on the right. Finally, a triplet is made of three estimates for the
effect of the interaction term on total (black squares), right-wing (blue triangle) and left-wing (red
diamond) populism, respectively. We explain below how the inclusion of potential amplifiers affects
the main findings of the previous sections.
Our first main result is that imports of low-skilled intensive goods increase the volume of total and
right-wing populism, without affecting the volume of left-wing populism. The estimates in Figure 7
show that these effects are reinforced in times of de-industrialization (Panel b) and when the internet
coverage is high (Panel c). On the contrary, a high level of diversity in imported (low-skilled labor
intensive) goods reduces the right-wing populism response (Panel d). In addition, it cannot ruled out
that imports increases the volume of left-wing populism in times of negative growth (Panel a).
Our second main result is that imports of low-skilled intensive goods increase the mean margin
of total and right-wing populism, without affecting the mean margin of left-wing populism. Figure 7
evidences that the right-wing populism response is larger when the internet coverage is high (Panel c),
25
The data sources are the Penn World Tables for GDP growth rates, the UN National Accounts for the
share of manufacturing output in GDP, Abel (2018) for dyadic immigration flow data, and the World Bank
WDI for internet coverage (we assume zero coverage before 1990, since the World Wide Web was invented in
1989). Data on genetic distance are taken from Spolaore and Wacziarg (2009). The top decile is derived for
values available from 1990.

36

Electronic copy available at: https://ssrn.com/abstract=4273442

 Figure 7: Interactions with amplifiers for volume and mean margins
Reduced-form IV PPML and 2SLS results

4
1
.5

.5

2
Volume Margin

Volume Margin
Mean Margin

Mean Margin
0

0
-.5

-.5

-2
-1
-1.5

-5

-1

-4
Import LS Migration LS Import LS Migration LS Import LS Migration LS Import LS Migration LS

(a) Economic crisis (b) De-industrialization

10

20
6

10
4

0
Volume Margin

Volume Margin
Mean Margin

Mean Margin
2

0
-1
0

-10
0

-20
-2

-5

-2

Import LS Migration LS Import LS Migration LS Import LS Migration LS Import LS Migration LS

(c) Internet coverage (d) Diversity (Imp) & genetic dist. (Mig)

Notes: Black (square), blue (triangle) and red (diamond) objects correspond to overall, right wing and left
wing dimensions, respectively. Dependent variable is the volume margin on the left panels, while is the
mean margin in the right panels. The estimates represent the coefficients of the interaction term between
migration (LS) and imports (LS) with a dummy equal to one if the country experienced a year of negative
real growth five years prior the election year (Figure (a)), as well as proxies for de-industrialization (Figure
(b)), trade diversity and genetic distance (Figure (c)) and for internet coverage (Figure (d)). 90% confidence
intervals are reported.

and that an effect on the mean margin of left-wing populism materializes during severe crises (Panel
a). In line with the volume-margin analysis, the right-wing response to trade is smaller when imported
goods are more diverse (Panel d).
Our third main result is that low-skill immigration induces a transfer of votes from left-wing to

37

Electronic copy available at: https://ssrn.com/abstract=4273442

right-wing populist parties. The results in Figure 7 show that the decline in left-wing populism is
stronger in times of negative growth (Panel a). Interactions with the de-industrialization and internet
coverage dummies are never significant. With regard to cultural distance, it does not amplify the
right-wing populist response to low-skill immigration. Similar findings are obtained when we replace
our proxy for cultural distance by an augmented diversity index a la Greenberg (1956) computed on
low-skill immigrants, that combines diversity, cultural and economic distance in a single variable.26
If anything, a high level of cultural distance reduces the centrist and left-wing populist responses to
immigration (Panel d).
Finally, our fourth results is that low-skill immigration has no meaningful impact on the mean
margin of populism. This results is fairly robust and unaffected when interactions terms are factored
in.

5 Conclusion
The recent waves of national elections have seen populist and nationalist parties gain ground in many
countries, and in the European Union in particular. Populism remains a multifaceted concept that is
difficult to objectify and quantify. We propose and construct new (or updated) measures of populism
that rely on the two main criteria identified in the literature – namely the anti-establishment and
commitment-to-protection stances of political parties and leaders. Our measures are consistent over
time and allow to characterize the populism scores for almost 4,000 party-election pairs from 55
countries, covering 628 elections and a 60-year time span. Equipped with these measures, we are able
to analyze the long-run trends in the ”volume margin” of populism (the measure most commonly used
in existing empirical studies, as it captures the vote share of all parties defined as populist) as well
as in the ”mean margin” of populism (i.e., the vote-weighted average level of populism of all parties,
which captures the extent of populism that voters are exposed to during an election).
We use these measures to characterize the trends in the levels of total, left-wing, centrist and
right-wing populism. In the descriptive part of our paper, we document that both (the volume and
mean) margins of populism have fluctuated since the 1960s, with peaks after each major economic
crisis. Moreover, we show that right-wing populism has reached an all-time high in the last decade.
The situation is particularly worrisome in the EU, where the recent rise in the volume and mean
margins of right-wing populism is more pronounced than in the rest of the world.
Our second objective is to empirically assess how globalization shocks have shaped populism trends
over the last six decades. We provide a unified analysis of the effect of import and immigration shocks
on populism and disentangle their respective effects according to their skill and cultural contents.
To address causation issues, we implement an instrumentation strategy that predicts changes in the
bilateral and in the skill structure of imports and of immigration using origin-specific factors.

26
Section D.10 shows that, once accounting for both economic and cultural distance, diversity has no am-
plifying effect on populism. If any, economic distance, rather than cultural distance, can enhance the effect of
low-skill immigration on the volume of populism.

38

Electronic copy available at: https://ssrn.com/abstract=4273442

 We find the the skill structure of globalization shocks is key to explaining populist trends. In
general, imports of high-skill labor intensive goods and high-skill immigration tend to reduce the
volume of total and right-wing populism. This is not the case of globalization shocks that are likely
to adversely affect low-skill voters and income inequality. Imports of low-skill intensive goods increase
total and right-wing populism along the volume and mean margins. These effects are greater in times
of de-industrialization and when the internet coverage is high, while smaller when the origin mix of
imported goods is more diverse. In normal times, import shocks have no effect on left-wing populism.
The latter results does not hold in times of severe crisis, when import shocks are persistent, or when
focusing on European countries only. Low-skill immigration induces a transfer of votes from left-wing
to right-wing populist parties, without affecting the total volume or mean margin of populism. The
right-wing populist response is not amplified by the average cultural distance between natives and
low-skill newcomers.
Hence, the effect of globalization on populism varies with the type and measure of populism, and
is strongly influenced by the skill and cultural characteristics of imported goods and people. This
suggests that the economic and cultural determinants of populism are not mutually exclusive. Our
analysis is conducted at the country level but the channels at work are likely to imply complex political
competition responses – as evidenced by the differential in the mean margin responses of never-populist
parties and others – as well as entry and exit changes – as evidenced by our decomposition of the
volume margin into its extensive and intensive margins. An empirical analysis conducted at the party
level could shed light on the re-positioning of traditional and populist parties as well as on the role
and intensity of underlying political competition responses to globalization shocks. We leave this for
further research.

39

Electronic copy available at: https://ssrn.com/abstract=4273442

References
Abel, G. J. (2018). Estimates of global bilateral migration flows by gender between 1960 and 2015.
International Migration Review, 52(3):809–852.

Alesina, A., Harnoss, J., and Rapoport, H. (2016). Birthplace diversity and economic prosperity.
Journal of Economic Growth, 21(2):101–138.

Alesina, A., Miano, A., and Stantcheva, S. (2022). Immigration and Redistribution. The Review of
Economic Studies.

Alesina, A., Murard, E., and Rapoport, H. (2021). Immigration and preferences for redistribution in
europe. Journal of Economic Geography, 21(6):925–954.

Algan, Y., Beasley, E., Cohen, D., and Foucault, M. (2018). The rise of populism and the collapse
of the left-right paradigm: Lessons from the 2017 french presidential election. CEPR Discussion
Papers, 13103.

Algan, Y., Guriev, S., Papaioannou, E., and Passari, E. (2017). The european trust crisis and the rise
of populism. Brooking Papers on Economic Activity, 7-8.

Anelli, M., Colantone, I., and Stanig, P. (2018). We were the robots: Automation and voting behavior
in western europe. IZA Discussion Papers, 12485.

Angrist, J. D. and Pischke, J.-S. (2008). Mostly harmless econometrics – An Empiricist’s Companion.
Princeton university press.

Autor, D., Dorn, D., and Hanson, G. (2013). The china syndrome: Local labor market effects of
import competition in the united states. American Economic Review, 103(6):2121–2168.

Autor, D., Dorn, D., and Hanson, G. (2016). The china shock: Learning from labor market adjustment
to large changes in trade. Annual Review of Economics, 8:205–240.

Autor, D., Dorn, D., Hanson, G., and Majlesi, K. (2020). Importing political polarization? the
electoral consequences of rising trade exposure. American Economic Review, 110(10):3139–3183.

Bakker, R., Edwards, E., Hooghe, L., Jolly, S., Koedam, J., Kostelka, F., Marks, G., Polk, J., Rovny,
J., Schumacher, G., Steenbergen, M., Vachudova, M., and Zilovic, M. (2015). 1999-2014 chapel
hill expert survey trend file. Version 1.13 Available on chesdata.eu. Chapel Hill, NC: University of
North Carolina, Chapel Hill.

Barone, G., D’Ignazio, A., De Blasio, G., and Naticchioni, P. (2016). Mr. rossi, mr. hu and politics.
the role of immigration in shaping natives’ voting behavior. Journal of Public Economics, 136:1–13.

Becker, S. and Fetzer, T. (2016). Does migration cause extreme voting? Center for Competitive
Advantage in the Global Economy and the Economic & Social Research Council, Working Papers,
306.

Becker, S., Fetzer, T., and Novy, D. (2017). Who voted for brexit? a comprehensive district-level
analysis. Economic Policy, 32(92):601–650.

Betz, H. (1994). Radical right-wing populism in western europe. Springer.

40

Electronic copy available at: https://ssrn.com/abstract=4273442

Betz, H. (2002). Conditions favouring the success and failure of radical right-wing populist parties in
contemporary democracies. Democracies and the populist challenge. Palgrave Macmillan, London.,
pages 197–213.

Boustan, L. (2010). Was postwar suburbanization “white flight”? evidence from the black migration.
Quarterly Journal of Economics, 125(1):417—-443.

Brunner, B. and Kuhn, A. (2018). Immigration, cultural distance and natives’ attitudes towards
immigrants: Evidence from swiss voting results. Kyklos, 71(1):28–58.

Budge, I. and Laver, M. (2016). Party policy and government coalitions. Springer.

Cadot, O., Carrere, C., and Strauss-Kahn, V. (2011). Export diversification: what’s behind the hump?
Review of Economics and Statistics, 93(2):590–605.

Campante, F., Durante, R., and Sobbrio, F. (2018). Politics 2.0: The multifaceted effect of broadband
internet on political participation. Journal of the European Economic Association, 16(4):1094–1136.

Card, D., Dustmann, C., and Preston, I. (2012). Immigration, wages, and compositional amenities.
Journal of the European Economic Association, 10(1):78–119.

Caselli, F., Koren, M., Lisicky, M., and Tenreyro, A. (2015). Diversification through trade. CEPR
Discussion paper, 10775.

Colantone, I., Ottaviano, G. I., and Stanig, P. (2021). The backlash of globalization. CESifo Workink
Paper, 9289.

Colantone, I. and Stanig, P. (2018). The trade origins of economic nationalism: Import competition
and voting behavior in western europe. American Journal of Political Science, 62(4):936–953.

Dancygier, R., Dehdari, S. H., Laitin, D., Marbach, M., and Vernby, K. (2022). Emigration and
populism. Manuscript.

De Bromhead, A., Eichengreen, B., and O’Rourke, K. (2013). Political extremism in the 1920s and
1930s: Do german lessons generalize? Journal of Economic History, 73(2):371–406.

De Vries, C. (2018). The cosmopolitan-parochial divide: Changing patterns of party and electoral
competition in the netherlands and beyond. Journal of European Public Policy, 25(11):1541–1565.

Di Giovanni, J. and Levchenko, A. (2009). Trade openness and volatility. Review of Economics and
Statistics, 91(3):558–585.

Dippel, C., Gold, R., and Heblich, S. (2015). Globalization and its (dis)-content: trade shocks and
voting behavior. NBER Working Papers, 21812.

Docquier, F., Turati, R., Valette, J., and Vasilakis, C. (2020). Birthplace diversity and economic
growth: evidence from the us states in the post-world war ii period. Journal of Economic Geography,
20(2):321–354.

Dustmann, C., Vasiljeva, K., and Piil Damm, A. (2019). Refugee migration and electoral outcomes.
Review of Economic Studies, 86(5):2035–2091.

41

Electronic copy available at: https://ssrn.com/abstract=4273442

Edo, A., Giesing, Y., Öztunc, J., and Poutvaara, P. (2019). Immigration and electoral support for the
far left and far right. European Economic Review, 115:99–143.

Eichengreen, B. (2018). The populist temptation: Economic grievance and political reaction in the
modern era. Oxford University Press.

Feenstra, R. C., Lipsey, R. E., Deng, H., M., C., A., and Mo, H. (2005). World trade flows: 1962-2000.
National Bureau of Economic Research, w11040.

Feyrer, J. (2019). Trade and income: Exploiting time series in geography. American Economic Journal:
Applied Economics, 11(4):1–35.

Frankel, J. A. and Romer, D. H. (1999). Does trade cause growth? American economic review,
89(3):379–399.

Frey, C., Berger, T., and Chen, C. (2018). Political machinery: Did robots swing the 2016 us presi-
dential election? Oxford Review of Economic Policy, 34(3):418–442.

Funke, M., Schularick, M., and Trebesch, C. (2016). Going to extremes: Politics after financial crises,
1870-2014. European Economic Review, 88:227–260.

Funke, M., Schularick, M., and Trebesch, C. (2020). Populist leaders and the economy. Technical
report, CEPR Discussion Paper DP15405.

Gallego, A., Kurer, T., and Schöll, N. (2018). Not so disruptive after all: How workplace digitalization
affects political preferences. Barcelona GSE Working Paper Series, 1063.

Greenberg, J. H. (1956). The measurement of linguistic diversity. Language, 32(1):109–115.

Grzymala-Busse, A. and McFaul, M. (2020). Votes for populists. Global Populisms Project, Stanford
University.

Guiso, L., Herrera, H., Morelli, M., and Sonno, T. (2017). Demand and supply of populism. CEPR
Discussion Papers, 11871.

Guiso, L., Herrera, H., Morelli, M., and Sonno, T. (2020). Economic insecurity and the demand of
populism in europe. Einaudi Institute for Economics and Finance Working Paper.

Guriev, S., Melnikov, N., and Zhuravskaya, E. (2019). 3g internet and confidence in government.
Manuscript.

Guriev, S. and Papaioannou, E. (2021). The political economy of populism. Journal of Economic
Literature, forthcoming.

Halla, M., Wagner, A., and Zweimuller, J. (2017). Immigration and voting for the far right. Journal
of the European Economic Association, 15(6):1341–1385.

Harmon, N. (2018). Immigration, ethnic diversity and political outcomes : Evidence from denmark.
Journal of Scandinavian Economics, 120(4):1043–1074.

Hausmann, R., Hwang, J., and Rodrik, D. (2007). What you export matters. Journal of Economic
Growth, 12(1):1–25.

42

Electronic copy available at: https://ssrn.com/abstract=4273442

Hawkins, K. A., Aguilar, R., Silva, B., Jenne, E., Kocijan, B., and Kaltwasser, C. (2019). Measuring
populist discourse: The global populism database. Paper presented at the 2019 EPSA Annual
Conference in Belfast, UK.

Hays, J., Lim, J., and Spoon, J. (2019). The path from trade to right-wing populism in europe.
Electoral Studies, 60(102038).

Helpman, E., Itskhoki, O., Muendler, M., and Redding, S. (2017). Trade and inequality: From theory
to estimation. Review of Economic Studies, 84(1):357–405.

Inglehart, R. and Norris, P. (2016). Trump, brexit, and the rise of populism: Economic have-nots and
cultural backlash. Manuscript.

Klemans, M. and Magruder, J. (2018). Labour market responses to immigration: Evidence from
internal migration driven by weather shocks. Economic Journal, 128(613):2032–2065.

Lancaster, T. (2000). The incidental parameter problem since 1948. Journal of econometrics,
95(2):391–413.

Leininger, A. and Meijers, M. J. (2020). Do populist parties increase voter turnout? evidence from
over 40 years of electoral history in 31 european democracies. Political Studies, 69(3):668–685.

Malgouyres, C. (2017). Trade shocks and far-right voting: evidence from french presidential elections.
Banque de France, Working Papers.

March, L. and Mudde, C. (2005). What’s left of the radical left? the european radical left after 1989:
Decline and mutation. Comparative European Politics, 3(1):23–49.

Mayda, A. M. (2006). Who is against immigration? a cross-country investigation of individual atti-

tudes toward immigrants. The review of Economics and Statistics, 88(3):510–530.

Mayda, A. M., Peri, G., and Steingress, W. (2022). The political impact of immigration: Evidence
from the united states. American Economic Journal: Applied Economics, 14(1):358–89.

Mendez, I. and Cutillias, I. (2014). Has immigration affected spanish presidential elections results?
Journal of Population Economics, 27:135–171.

Moffitt, B. and Tormey, S. (2014). Rethinking populism: Politics, mediatisation and political style.
Political studies, 62(2):381–397.

Monras, J. (2020). Immigration and wage dynamics: Evidence from the mexican peso crisis. Journal
of Political Economy, 128(8):3017–3089.

Morelli, M., Nicolo, A., and Roberti, P. (2021). A commitment theory of populism. CESifo Working
Paper, 9473.

Moriconi, S., Peri, G., and Turati, R. (2019). Immigration and voting for redistribution: Evidence
from european elections. Labour Economics, 61(101765).

Moriconi, S., Peri, G., and Turati, R. (2022). Skill of the immigrants and vote of the natives: Immi-
gration and nationalism in european elections 2007–2016. European Economic Review, 141:103986.

Mudde, C. (2004). The populist zeitgeist. Government and Opposition, 39(7):541–563.

43

Electronic copy available at: https://ssrn.com/abstract=4273442

Mukand, S. and Rodrik, D. (2018). The political economy of ideas: On ideas versus interests in
policymaking. NBER Working Paper, 24467.

Munshi, K. (2003). Networks in the modern economy: Mexican migrants in the u.s. labor market.
Quarterly Journal of Economics, 118(2):549–599.

Müller, J. (2016). What is populism? Manuscript: University of Pennsylvania Press.

Naik, G. and Leuthold, R. M. (1986). A note on qualitative forecast evaluation. American Journal of
Agricultural Economics, 68(3):721–726.

Norris, P. and Inglehart, R. (2019). Cultural backlash: Trump, brexit, and authoritarian populism.
Cambridge University Press.

Otto, A. and Steinhardt, M. (2014). Immigration and election outcomes—evidence from city districts
in hamburg. Regional Science and Urban Economics, 45:67–79.

Preacher, K. and MacCallum, R. (2003). Repairing tom swift’s electric factor analysis machine.
Understanding statistics: Statistical issues in psychology, education, and the social sciences, 2(1):13–
43.

Reynolds, A., Reilly, B., and Ellis, A. (2008). Electoral system design: The new international IDEA
handbook. International Institute for Democracy and Electoral Assistance.

Rodrik, D. (1997). Has globalization gone too far? Washington DC: Institute for International
Economics, Working Paper.

Rodrik, D. (2018). Populism and the economics of globalization. Journal of International Business
Policy, 1(1/2):12–33.

Rodrik, D. (2021). Why does globalization fuel populism? economics, culture and the rise of right-wing
populism. Annual Review of Economics, 13(forthcoming).

Rooduijn, M., De Lange, S., and Van der Brug, W. (2014). A populist zeitgeist? programmatic
contagion by populist parties in western europe. Party politics, 20(4):563–575.

Rooduijn, M., Van Kessel, S., Froio, C., Pirro, A., De Lange, S., Halikiopoulou, D., Lewis, P., Mudde,
C., and Taggart, P. (2019). The populist: An overview of populist, far right, far left and eurosceptic
parties in europe. Database.

Santos Silva, J. and Tenreyro, S. (2006). The log of gravity. Review of Economics and Statistics,
88(4):641–658.

Santos Silva, J. and Tenreyro, S. (2010). On the existence of the maximum likelihood estimates for
poisson regression. Economic Letters, 107(2):310–312.

Schneider-Strawczynski, S. (2021). When is contact effective? evidence on refugee-hosting and far-right

support in france. Technical report, Mimeo.

Shea, J. (1997). Instrument relevance in multivariate linear models: A simple measure. Review of
Economics and Statistics, 79(2):348–352.

44

Electronic copy available at: https://ssrn.com/abstract=4273442

Shehaj, A., Shin, A., and Inglehart, R. (2019). Immigration and right-wing populism: An origin story.
Party Politics, 1354068819849888.

Shils, E. (1956). The torment of secrecy: The background and consequences of american security
policies. London: Heinemann.

Sikk, A. (2009). Parties and populism. CEPSI Working Papers, 2009-02.

Spolaore, E. and Wacziarg, R. (2009). The diffusion of development. Quarterly Journal of Economics,
124:469–592.

Stanley, B. (2008). The thin ideology of populism. Journal of Political Ideologies, 13(1):95–110.

Steinmayr, A. (2021). Contact versus exposure: Refugee presence and voting for the far right. Review
of Economics and Statistics, 103(2):310–327.

Swank, D. (2018). Comparative political parties dataset: Electoral, legislative, and government
strength of political parties by ideological group in 21 capitalist democracies, 1950-2015. Electronic
Database, Department of Political Science, Marquette University.

Swank, D.and Betz, H. (2003). Globalization, the welfare state and right-wing populism in western
europe. Socio-Economic Review, 1:215–245.

Taggart, P. (2000). Populism: Concepts in the social sciences. Philadelphia: Open.

Van Kessel, S. (2015). Populist parties in europe: Agents of discontent? Springer.

Vannoorenberghe, G. (2012). Firm-level volatility and exports. Journal of International Economics,

86(1):57–67.

Weyland, K. (2001). Clarifying a contested concept: Populism in the study of latin american politics.
Comparative politics, pages 1–22.

Wiles, P. (1969). A syndrome, not a doctrine: Some elementary theses on populism. Populism: Its
meaning and national characteristics, pages 166–179.

Zhuravskaya, E., Petrova, M., and Enikolopov, R. (2020). Political effects of the internet and social
media. Annual Review of Economics, 12(1):415–438.

45

Electronic copy available at: https://ssrn.com/abstract=4273442

 Online Appendix
A List of countries included in MPD
Figure A-I illustrates the set of countries available in our data set. We cover both economically
developing and developed countries, not all of them being available from the beginning of our period
of analysis.

Figure A-I: Countries available in MPD data

Note: The figure plots the countries that have at least one electoral and the different colors show the year of
the first election available in the sample.
Source: Authors’ elaboration on MPD.

Table A-I provides the list of countries, the year of the first (third column) and last (fourth column)
electoral event available, the number of elections (fifth column) and the total number of unique parties
that won at least one seat in an electoral event (sixth column). The total number of observations in
our dataset (hence party-election) is 3,860. As expected, several former Soviet Union countries enter
the sample after the fall of the Berlin Wall (1989). At this stage, if a party A changes its name to
become party B between elections, we count them as two different parties. MPD provides the 1990
results for the German Democratic Republic (East Germany), but we remove them from the sample.
Data for the 2014 Uruguayan elections is also available, but we drop the country due to the lack of
time variation.

46

Electronic copy available at: https://ssrn.com/abstract=4273442

 Table A-I: Manifesto Project Database – Sample

# 1st E. Last E. N# E. N# P.P. # 1st E. Last E. N# E. N# P.P.

Denmark 1 1960 2015 21 20 Bulgaria 30 1990 2017 10 28
Japan 2 1960 2014 19 28 Croatia 31 1990 2016 9 36
New Zealand 3 1960 2017 20 14 Czech Rep. 32 1990 2017 9 25
Sweden 4 1960 2018 18 16 Georgia 33 1990 2016 9 42
USA 5 1960 2016 15 2 Hungary 34 1990 2014 7 15
Australia 6 1961 2016 22 13 Montenegro 35 1990 2016 10 23
Belgium 7 1961 2014 17 34 Nth Mac. 36 1990 2016 9 29
Germany 8 1961 2017 16 11 Romania 37 1990 2016 8 31
Ireland 9 1961 2016 16 16 Serbia 38 1990 2016 11 38
Israel 10 1961 2015 16 59 Slovakia 39 1990 2016 9 27
Mexico 11 1961 2015 19 26 Slovenia 40 1990 2014 8 22
Norway 12 1961 2017 15 13 Albania 41 1991 2001 5 12
Turkey 13 1961 2018 16 22 Poland 42 1991 2011 7 30
Austria 14 1962 2017 17 13 Estonia 43 1992 2015 7 24
Canada 15 1962 2015 18 9 Lithuania 44 1992 2016 7 27
Finland 16 1962 2015 15 19 Sth Korea 45 1992 2016 7 16
France 17 1962 2017 14 35 Latvia 46 1993 2018 9 36
Iceland 18 1963 2017 17 19 Russia 47 1993 2011 6 25
Italy 19 1963 2018 15 57 Moldova 48 1994 2014 7 16
Netherlands 20 1963 2017 17 28 Sth Africa 49 1994 2014 5 6
Switzerland 21 1963 2015 14 25 Ukraine 50 1994 2007 5 29
Luxembourg 22 1964 2013 11 11 Armenia 51 1995 2012 5 16
UK 23 1964 2017 15 13 Azerbaijan 52 1995 2000 2 6
Greece 24 1974 2015 17 18 Cyprus 53 1996 2016 5 11
Portugal 25 1975 2015 15 19 Malta 54 1996 1998 2 2
Spain 26 1977 2016 13 38 Bolivia 55 2009 2014 2 8
Argentina 27 1989 2013 6 14
Chile 28 1989 2017 6 15
Bosnia-Herz. 29 1990 2014 8 19 Total 628 1206
Note: Countries are sorted by the year of the first election available and alphabetically when having the same
first year in the data.
Source: Authors’ elaboration on MPD.

47

Electronic copy available at: https://ssrn.com/abstract=4273442

B Construction of New Populism Score
B.1 Definitions and Correlation with MPD Components
Table B-I presents the name, description and source of the variables used in the construction of the
Populism score. Panel A presents the two proxies used to capture parties’ anti-establishment stance,
while Panel B shows the four proxies selected to capture the commitment-to-protection stance by
focusing on external/foreign threats. When both positive and negative stances towards a specific issue
are reported in the Manifesto Project Database (e.g. Internationalism), we constructed a measure of
net favorable position, which is the difference between favorable and negative references. Concerning
the proxy on EU institutions (CT P3 ), for parties outside the EU, and so less interested on the topic,
we replace the value of that variable equal to zero.
Table B-II provides the level, direction and significance of the correlation between the above
mentioned political preferences within each domain. Even though the pairwise correlations are small,
going from a value of 0.04 to 0.162 in absolute terms, they are highly statistically significant. Moreover,
the direction of the correlations supports our previous set of intuitions. Parties that are particularly
against political corruption are also more prone to claim themselves better than the others, as the
positive correlation in Col. (1) suggests. Cols. (2) to (4) show that internationalization is positively
related with positive statements towards the European Union, while these aspects are negatively
correlated with positive views towards protectionism and nationalization.
Table B-III describes the results related to the Polychoric Principal Component Analysis used
to construct synthetic indexes for parties’ anti-establishment and commitment-to-protection stances.
For both set of variables, only the first component has an eigenvalue above one, hence following the
Kaiser-Guttman criterion we retain only the first components as our synthetic indexes. Looking at the
coefficients/loadings associated to the anti-establishment stance, we can see that the first component
gives positive and equal weights to both variables, AES1 and AES2 , indicating that parties against
political corruption and pluralism will have an higher first component. We then define this first com-
ponent as our index of anti-establishment stance (IAES ). With regard to commitment to protection,
the first component give high weights to all the analyzed variables, and provides negative weights on
parties’ positive stance towards protectionism (CT P1 ) and nationalization (CT P4 ), positive weights
on support for internationalism (CT P2 ) and EU institutions (CT O3 ). Hence, parties with a more
political openness agenda will score high on the first component. To facilitate the interpretation,
we multiply the first component by minus one, and we define such flipped first component as our
commitment-to-protection index (ICT P ).

48

Electronic copy available at: https://ssrn.com/abstract=4273442

 Table B-I: Selection of Political Dimensions in MPD

Variables Description MPD Label

Panel A:
Anti-establishment Stance
Pol. corruption (AES1 ) Need to eliminate political corruption and associated abuses per304
of political and/or bureaucratic power. Need to abolish
clientelist structures and practices.

Anti-pluralism (AES2 ) References to the manifesto party’s competence to govern per305

and/or other party’s lack of such competence. Also includes
favourable mentions of the desirability of a strong and/or
stable government in general.
Panel B:
Commitment-to-protection stance
Protectionism (CTP1 ) Net favorable position. (per406) Favourable mentions of ex- per406-per407
tending or maintaining the protection of internal markets.
Measures may include: tariffs, quota restrictions and export
subsidies. (per407) Support for the concept of free trade
and open markets. Call for abolishing all means of market
protection.

Internationalism (CTP2 ) Net favorable position. (per107) Need for international co- per107-per109
operation, including co-operation with specific countries.
May also include references to: the need for aid to develop-
ing countries; need for world planning of resources; support
for global governance; need for international courts; support
for UN and international organisations. (per109) Negative
references to international co-operation. Favourable men-
tions of national independence and sovereignty with regard
to the manifesto country’s foreign policy, isolation and/or
unilateralism as opposed to internationalism.

EU Institutions (CTP3 ) Net favorable position. (per108) Favourable mentions of per108-per110

European Community/Union in general. May include the:
desirability of the manifesto country joining (or remaining
a member); desirability of expanding the European Com-
munity/Union; desirability of increasing the ECs/EUs com-
petences; desirability of expanding the competences of the
European Parliament. (per110) Negative references to the
European Community/Union. May include: opposition to
specific European policies which are preferred by European
authorities; opposition to the net-contribution of the mani-
festo country to the EU budget.

Nationalization (CTP4 ) Favourable mentions of government ownership of industries, per413

either partial or complete; calls for keeping nationalised in-
dustries in state hand or nationalising currently private in-
dustries. May also include favourable mentions of govern-
ment ownership of land.

49

Electronic copy available at: https://ssrn.com/abstract=4273442

 Table B-II: Correlations across political dimensions

AES2 CTP2 CTP3 CTP4

Panel A
AES1 .070†

Panel B
CTP1 -.041∗ -.095‡ .081‡
CTP2 .104‡ -.069†
CTP3 -.162‡
Notes: The table shows the pairwise correlation and the precision associated
to the political preferences related to: anti-establishment stance (Panel A) and
commitment-to-protection stance (Panel B). Level of significance: * p<0.05, **
p<0.01, *** p<0.001, † p<0.0001, ‡ p<0.00001.
Source: Authors’ elaboration on MPD.

Table B-III: PPCA - Anti-Establishment & Commitment-to-Protection stances

Commitment to Protection (CTP)

Anti-Establishment (AES) (1) (2) (3)
Comp. Eigenv. Explained Cumulative
(1) (2) (3)
Comp. Eigenv. Explained Cumulative Comp. 1 1.287 0.322 0.322
Comp. 2 0.960 0.240 0.562
Comp. 1 1.070 0.535 0.535
Comp. 3 0.921 0.230 0.792
Comp. 2 0.930 0.465 1
Comp. 4 0.832 0.207 1

Scoring Coefficients/Loadings
Scoring Coefficients/Loadings
Variable Comp 1 Comp 2 Comp 3
Variable Comp 1 Comp 2
CT P1 -0.412 0.668 -0.613
AES1 0.707 0.707 CT P2 0.409 0.736 0.499
AES2 0.707 -0.707 CT P3 0.597 0.043 -0.247
CT P4 -0.552 0.094 0.550

50

Electronic copy available at: https://ssrn.com/abstract=4273442

B.2 Correlation between our populism score and preferences for
immigration, cultural identity and interventionism
p
In Table B-IV, we compute partial correlations between our populism score Si,e,t and four MPD proxies
capturing preferences for immigration and multiculturalism. We control for country and year fixed
effects. We use four variables available in the MPD database: (i) immigration is negative for country’s
national way of life, (ii) immigration is positive for country’s national way of life, (iii) immigration
positively contributes to multiculturalism, and (iv) immigrant should assimilate to the country culture.
Note that these variables are not available for the years prior to 2006. In line with intuition, we find
that the populism score of centrist and right-wing parties is negatively and significantly correlated with
positive attitudes towards immigration and multiculturalism. This is not the case among left-wing
parties.
In Table B-V, we compute pairwise correlations between our populism score and proxies for (i)
cultural conservatism, (ii) welfare state expansion, and (iii) preferences for government intervention
and economic planning. We find that the populism score of centrist and right-wing parties is positively
and significantly correlated with cultural conservatism; this is not the case among left-wing parties.
Interventionism and populism are positively and significantly correlated on both sides of the left-to-
right spectrum.

Table B-IV: Populism Score and Migration-Related Political Preferences

All Parties No Left-Wing Parties Left-Wing Parties

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12)
Immi. Immi. Immi. Immi. Immi. Immi.
Immi. Immi. (+) Assimi- Immi. Immi. (+) Assimi- Immi. Immi. (+) Assimi-
(–) (+) Multicul. lation (–) (+) Multicul. lation (–) (+) Multicul. lation

Populism 0.194 -0.052 -0.066∗∗ 0.095 0.249 -0.060∗∗ -0.085∗∗ 0.156 0.003 0.048 -0.044 –0.043∗∗
Score (0.232) (0.041) (0.032) (0.060) (0.371) (0.029) (0.035) (0.094) (0.026) (0.159) (0.080) (0.018)

R2 0.183 0.275 0.260 0.287 0.285 0.524 0.276 0.436 0.304 0.358 0.392 0.309
Obs. 572 572 572 572 334 334 334 334 229 229 229 229
Cntry FE X X X X X X X X X X X X
Year FE X X X X X X X X X X X X
Notes: ***, **, and * denote significance at the 1, 5, and 10 percent levels, respectively; clustered standard errors at
the country level are reported in parentheses. Analysis available from 2006 on, given the availability of the measures
only from that election-year.

51

Electronic copy available at: https://ssrn.com/abstract=4273442

 Table B-V: Populism Score and preferences for culture and interventionism

All Parties No Left-Wing Parties Left-Wing Parties

(1) (2) (3) (4) (5) (6) (7) (8) (9)
Govern- Govern- Govern-
Cultural Welfare ment Inter. Cultural Welfare ment Inter. Cultural Welfare ment Inter.
Conser- State & Econ. Conser- State & Econ. Conser- State & Econ.
vatism Expansion Planning vatism Expansion Planning vatism Expansion Planning

Populism 0.149∗∗∗ -0.037 0.148∗∗∗ 0.190∗∗∗ -0.060 0.068∗∗ -0.131 0.039 0.357∗∗∗
Score (0.043) (0.043) (0.033) (0.054) (0.046) (0.028) (0.093) (0.059) (0.066)

R2 0.152 0.233 0.190 0.204 0.215 0.216 0.332 0.310 0.268

Obs. 3860 3860 3860 2573 2573 2573 1258 1285 1285
Country FE X X X X X X X X X
Year FE X X X X X X X X X
Notes: ***, **, and * denote significance at the 1, 5, and 10 percent levels, respectively; clustered standard
errors at the country level are reported in parentheses.

52

Electronic copy available at: https://ssrn.com/abstract=4273442

B.3 Selection of the Threshold Used to Define Populist Parties
Most of existing studies provide a dichotomous classification of populist parties. Based on our contin-
uous and centered (i.e., zero-mean) score of populism, we classify a party as populist (1(SD)) when its
score exceeds a certain threshold, which can be expressed as a multiplying factor SD of the standard
deviation. In the core of the text, Figure 1 shows that SD = 1 is a relevant threshold, maximizing
the partial correlation with three existing classifications. Figure B-I below shows that SD = 1 also
maximizes the rate of accurate forecasts for the overall set of parties and for populist parties only,
whatever the classification used as a reference (even the GPop 1 classification).

Figure B-I: Threshold definition - Share of correct predictions (cont’d)

.35

.08
Van Kessel - Goodness of the fit (pop.)
.3

Swank - Goodness of the fit (pop.)

.06
.25
.2

.04
.15

.02
.1
.05

0
0

0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0 0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0
Populism ID (Threshold, SD) Populism ID (Threshold, SD)

(e) Van Kessel - Populist Parties (f) Swank - Populist Parties

.1
.2
PopuList - Goodness of the fit (pop.)

GPop 1 - Goodness of the fit (pop.)

.08
.15

.06
.1

.04
.05

.02
0
0

0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0 0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0
Populism ID (Threshold, SD) Populism ID (Threshold, SD)

(g) PopuList - Populist Parties (h) GPop 1 - Populist Parties

Notes: The figure shows the proportion of good matches among populist parties after predicting a populist
party identifier based on the estimated models presented in Figure 1 and comparing it with the following
populist identifier based on: Van Kessel (2015) (Panel e), Swank (2018) (Panel f), Rooduijn et al. (2019)
(Panel g) and Grzymala-Busse and McFaul (2020) (Panel h). A party is classified as populist if the predicted
probability to be populist is above 0.5.
Source: Authors’ elaboration on MPD.

53

Electronic copy available at: https://ssrn.com/abstract=4273442

 Figure B-I: Threshold definition - Share of correct predictions
.86

.94
.93
.84
Van Kessel - Goodness of the fit

Swank - Goodness of the fit

.92
.82

.91
.8

.9
.78

.89
.76

.88
0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0 0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0
Populism ID (Threshold, SD) Populism ID (Threshold, SD)

(a) Van Kessel - All Parties (b) Swank - All Parties

.9

.9
.89

.895
PopuList - Goodness of the fit

GPop 1 - Goodness of the fit

.88

.89
.87

.885
.86

.88
.85

.875
.84

.87
.83

.865
.82

0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0 0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0
Populism ID (Threshold, SD) Populism ID (Threshold, SD)

(c) PopuList - All Parties (d) GPop 1 - All Parties

Notes: The figure shows the proportion of good matches among all parties after predicting a populist party
identifier based on the estimated models presented in Figure 1 and comparing it with the following populist
identifier based on: Van Kessel (2015) (Panel a), Swank (2018) (Panel b), Rooduijn et al. (2019) (Panel
c) and Grzymala-Busse and McFaul (2020) (Panel d). A party is classified as populist if the predicted
probability to be populist is above 0.5.
Source: Authors’ elaboration on MPD.

54

Electronic copy available at: https://ssrn.com/abstract=4273442

B.4 Stylized Facts: Robustness to Threshold Selection
Figures B-II, B-III and B-IV illustrate the robustness of the stylized fact described in Section 2 to the
selection of the threshold used to classify parties. All stylized facts are preserved when using a lax or
restrictive classification of populist parties.
Figure B-II: Populist parties - different threshold
10

60
Elections with populist party (%)
Total number of populist parties
8

50
6

40
4

30
2

20
0

1960 1970 1980 1990 2000 2010 2020 1960 1970 1980 1990 2000 2010 2020

Standard Less strict Strict Standard Less strict Strict

(a) Number of Populist Parties (b) Elections with a populist party (%)

2015
2010

2000

1990

1980

1970

1963

-.5 0 .5 1 1.5 2
Populism index

Not pop. party (std.) Not pop. party (less) Not pop. party (strict) [Panel 1]

Pop. party (std) Pop. party (less) Pop. party (strict) [Panel 2]

(c) Score of populist vs. non populist

Notes: Fig. (a) shows the total number of populist parties. Fig. (b) gives the percentage of elections with
at least a Populist party. Fig. (c) presents the average populism score of populist and non populist parties.
Populist parties are defined as those with a score exceeding 1 standard deviation (standard), exceeding 0.9
standard deviation (lax) or exceeding 1.1 standard deviation (strict). Figures (a), (b) and (c) show moving
averages including 3 years before and 3 years after each date.
Source: Authors’ elaboration on MPD.

55

Electronic copy available at: https://ssrn.com/abstract=4273442

 20 Figure B-III: Populist parties and the left-right wing divide – different threshold

15
Total nb. of populist parties

Total nb. of populist parties

15

10
10

5
5
0

0
1960 1970 1980 1990 2000 2010 2020 1960 1970 1980 1990 2000 2010 2020

Total RW LW Centre Total RW LW Centre

(a) Number of Populist Parties - Lax (b) Number of Populist Parties - Strict
2.2

2.2
2
Average populism index

Average populism index

2
1.8

1.8
1.6

1.6
1.4

1.4
1.2

1960 1970 1980 1990 2000 2010 2020 1960 1970 1980 1990 2000 2010 2020

Populist parties RW LW Populist parties RW LW

(c) Average Score Populist Parties - Lax (d) Average Score Populist Parties - Strict

Note: Fig. (a)-(b) shows the total number of populist parties, dividing between left-wing and right wing.
Fig. (c)-(d) presents the average populism score of populist parties, splitting between left-wing and right-
wing parties. Populist parties are defined as those with a score exceeding 0.9 standard deviation (Fig.
(a)-(c)) and 1.1 standard deviation (Fig. (b)-(d)), while left-wing and right-wing parties are defined as
those that belongs to the first and third tercile of the left-to-right index. Figures (a), (b), (c) and (d) show
moving averages including 3 years before and 3 years after each date.
Source: Authors’ elaboration on MPD.

56

Electronic copy available at: https://ssrn.com/abstract=4273442

 Figure B-IV: Populist parties and the left-right wing divide - different threshold (cont’d)
40

40
Election with LW populist party (%)

Election with LW populist party (%)

30

30
20

20
10

10
0

0
1960 1970 1980 1990 2000 2010 2020 1960 1970 1980 1990 2000 2010 2020

Full sample EU28 RoW Full sample EU28 RoW

(a) Elections with LW populist party (%) - Lax (b) Elections with LW populist party (%) - Strict

50
50
Election with RW populist party (%)

Election with RW populist party (%)

40
40

30
30

20
20

10
10
0

1960 1970 1980 1990 2000 2010 2020 1960 1970 1980 1990 2000 2010 2020

Full sample EU28 RoW Full sample EU28 RoW

(c) Elections with RW populist party (%) - Lax (d) Elections with RW populist party (%) - Strict

Note: Fig. (a)-(b) shows the percentage of elections with a left-wing party. Fig. (c) - (d) presents the
percentage of elections with a right-wing party. Populist parties are defined as those with a score exceeding
0.9 standard deviation (Fig. (a)-(c)) and 1.1 standard deviation (Fig. (b)-(d)), while left-wing and right-
wing parties are defined as those that belongs to the first and third tercile of the left-to-right index. Figures
(a), (b), (c) and (d) show moving averages including 3 years before and 3 years after each date.
Source: Authors’ elaboration on MPD.

57

Electronic copy available at: https://ssrn.com/abstract=4273442

B.5 Stylized Facts - Robustness to Balanced Sample
Figures B-V, B-VI and B-VII illustrate the robustness of the stylized facts described in Section 2 to
the composition of the sample. In this section, the stylized facts are presented considering the set
of countries that appear in the MPD database starting from the first decade of 1960s. The balanced
sample exclude Greece, Portugal, Spain as well as Latin American and former soviet union countries.

Figure B-V: Stylized facts I – Distribution of populism scores and mean margin of populism in
the balanced Sample (1960-2018)
.8

.04
.6

.03
Density

Theil index
.4

.02
.2

.01
0

-6 -4 -2 0 2 4 6
0

Populism index 1960 1970 1980 1990 2000 2010 2020

1960-69 1970-79 1980-89 1990-99 2000-09 2010-18 Theil all Theil within Theil between

(a) Density of populism score by period (b) Inequality across parties (Theil)
.4
.4

Adj. average populism index (weighted)

.2
Average populism index
.2

0
0

-.2
-.2

-.4
-.4

1960 1970 1980 1990 2000 2010 2020 1960 1970 1980 1990 2000 2010 2020

Full sample EU28 RoW Full sample EU28 RoW

(c) Mean populism score of all parties (unweighted) (d) Mean margin of populism (ΠM
i,e,t )

Notes: Fig. (a) shows the kernel-density of the populism score by decade. Fig. (b) depicts the Theil index
of inequality in populism across parties, and gives its between-countries component and the within-countries
components (Cadot et al., 2011). Fig. (c) plots the average populism score of all parties running for election
in a given year. Fig. (d) plots the mean margin of populism, a weighted average of the populism scores with
weights equal to the party’s share in votes. Fig. (c) and (d) show moving averages including 3 years before
and 3 years after each date. The balanced sample excludes Greece, Portugal and Spain, Latin American
and former soviet union countries.

58

Electronic copy available at: https://ssrn.com/abstract=4273442

Figure B-VI: Stylized facts II – Presence, electoral success and score of populist parties in the
balanced Sample (1960-2018)

70
1.4

60
Election with populist party (%)
1.2
# Populist Parties/# Elections

50
1

40
.8

30
.6

20
.4

10
.2
0

0
1960 1970 1980 1990 2000 2010 2020 1960 1970 1980 1990 2000 2010 2020

Full sample EU28 RoW Full sample EU28 RoW

(a) Number of populist parties per election (b) Elections with a populist party (%)

2020
25
Votes gained by populist parties (%)

2010
20

2000
15

1990
10

1980

1970
5

1960

-.5 0 .5 1 1.5 2
0

1960 1970 1980 1990 2000 2010 2020 Populism index

Full sample EU28 RoW Not populist party Populist party

(c) Volume margin of populism (ΠVi,e,t ) (d) Mean score of populist vs. non populist

Notes: Fig. (a) shows the total number of populist parties. Fig. (b) gives the percentage of elections
with at least a Populist party. Fig. (c) depicts the average share of votes for populist parties (the volume
margin). Fig. (d) presents the average populism score of populist and non populist parties. Populist parties
are defined as those with a score exceeding 1 standard deviation (0.81). Fig. (a), (d), (e) and (f) show
moving averages including 3 years before and 3 years after each date. The balanced sample excludes Greece,
Portugal and Spain, Latin American and former soviet union countries.

59

Electronic copy available at: https://ssrn.com/abstract=4273442

Figure B-VII: Stylized facts III – Left-wing and right-wing populism at the aggregate level in
the balanced Sample (1960-2018)

3
.8
# Populist parties / # Elections

2.5
Average populism index
.6

2
.4

1.5
.2

1
0

1960 1970 1980 1990 2000 2010 2020 1960 1970 1980 1990 2000 2010 2020

Total RW LW Centre Populist parties RW LW

(a) Number of populist parties (b) Average score of populist parties

40

Election with RW populist party (%)

Election with LW populist party (%)

50
30

40
30
20

20
10

10
0

1960 1970 1980 1990 2000 2010 2020 1960 1970 1980 1990 2000 2010 2020

Full sample EU28 RoW Full sample EU28 RoW

(c) Elections with a LW populist party (%) (d) Elections with a RW populist party (%)

Notes: Fig. (a) shows the total number of populist parties, dividing between left-wing and right wing.
Fig. (b) presents the average populism score of populist parties, splitting between left-wing and right-wing
parties. Fig. (c) and (d) give the percentage of elections with at least a left-wing and right-wing Populist
party, respectively. Populist parties are defined as those with a score exceeding 1 standard deviation (0.808),
while left-wing and right-wing parties are defined as those that belongs to the first and third tercile of the
left-to-right index. Fig. (a), (b), (c) and (d) show moving averages including 3 years before and 3 years
after each date. The balanced sample excludes Greece, Portugal and Spain, Latin American and former
soviet union countries.

60

Electronic copy available at: https://ssrn.com/abstract=4273442

B.6 Does Populism Require a More Extensive Definition?
We consider two extended populism scores that exploit additional potential characteristics of pop-
ulist parties, and check whether these extended scores better correlate with existing measures. Our
first extended score accounts for the fact that populist parties are sometimes characterized by their
shortsighted and opportunistic research agenda, which guides their political strategy (Guiso et al.,
2017). Populists rely on narrow/thin ideological references, which can cohabit with other ideological
framework, like the usual left-right divide (Mudde, 2004; Rooduijn et al., 2014). However, a frequent
denominator is that their main objective is to increase parties political support and consensus in
the short-run (Weyland, 2001; Betz, 2002), without addressing the long-run challenges faced by the
society. Populist parties tend to focus on more actual and immediately salient issues, implying the
concealment of long run costs and issues. Building on our Standard Populism Score, we construct an
extended index that includes a third component. We refer to it as the 3C Populism Score, which
accounts for shortsighted opportunistic strategy (OPP). To do so, we combine two additional MPD
variables covering aspects which are primarily influenced by policies with a long-term perspective, i.e.,
the salience of and position towards (i ) education expansion, which involves mentions towards expan-
sion of educational provision and the reduction of educational fees, and (ii ) environmental protection,
capturing parties’ favorable positions towards green economy and the need for fighting climate change.
Our second extended score accounts for the whole set of information available in MPD. We con-
struct synthetic indices of political preferences using the remaining set of 44 variables available from
the MPD. We only consider variables that are available for all political parties included in our sample
over the whole period. In line with our PPCA approach, we first perform a PPCA over the variables
belonging to the different domains covered by MPD and then retain components with an eigenvalue
above one, in line with Kaiser’s criterion. We end up with 12 synthetic indices capturing new political
dimensions. We then combine them with the three dimensions of populism used to construct the 3C
Populism Score (i.e., AES, CTP and OPP).27 We use the same dimensionality reduction technique
(PPCA) as in the previous section to construct our populism score, referred to as the 15C Populism
Score.
Table B-VI provides the eigenvectors associated with the variables within each component. The
first component, which explains the majority of the variance in the data, is positively correlated with
our three highlighted indices. In addition, the size of their coefficient is intuitive, suggesting that the
three indices play a relevant role in the definition of the first component. We then define this first
component as our 15C. Such an index not only has at its core the main features which characterized
the 3C Populism Score (positive correlation with parties’ stance towards anti-establishment issues,

27
These new dimensions are: (1) promotion of peaceful external relationship; (2) support towards freedom,
democracy and constitution; (3) support for political decentralization and public administration efficiency;
(4) support for free markets and incentives; (5) economic growth and investments as main tool for country
development; (6) support for government intervention in the economy and economic planning; (7) welfare state
expansion and support for equality; (8) support for cultural activities likes museums; (9) support for cultural
conservatism; (10) support for tradition-based national cohesion rather then public enforcement; (11) focus on
non-economic groups of the society; (12) focus on economic groups of the society.

61

Electronic copy available at: https://ssrn.com/abstract=4273442

commitment to protection, and concealment of long term issues), but it also account for parties’
position towards the whole spectrum of political issues.

Table B-VI: PPCA - Eigenvectors

Cpt 1 Cpt 2 Cpt 3 Cpt 4 Cpt 5 Cpt 6

Index
Anti-establishment .3158497 .1073219 .1535753 -.4652453 -.060134 .1672175
Protectionism .248782 -.1051573 .4849515 .0699972 .0612197 .2616881
LT costs .391317 .2846425 -.0258601 -.0218205 .2953913 .1147275
Peaceful ext. relations .3080404 -.2912411 -.1521264 .0211118 .2251427 .2179866
Freedom & democracy .2698057 -.118912 -.4486792 -.1652234 .0360432 -.2501172
Political decentralization -.2338466 .1994102 -.1471295 -.5093002 .2693444 .1828562
Free market -.1443971 .500238 -.0213432 .0493518 .0840578 -.2568531
Economic growth -.326282 -.0733257 .3307502 .0351375 .3905409 -.1060229
Economic planning .2513203 .0518399 .2256998 .3248763 .4520339 -.0572067
Welfare state expansion -.1061088 -.4944631 .269303 -.0236585 -.0187879 -.2319607
Cultural conservatism .0731171 .2733431 .2519378 .1362854 -.5866738 .2836778
Tradition-based cohesion .2641652 -.002604 -.2346418 .3876198 -.1576766 -.3252288
Non-econ. groups focus -.1079801 -.1032262 -.346962 .3349732 .1290002 .6277441
Econ. groups focus -.0954114 .3998768 .0256558 .2829833 .1735995 -.0327582
Support cultural activities -.4064793 -.0979823 -.1468994 .1419717 -.027272 .1766541

Table B-VII presents the correlation between the standard populism index and the six components
from our last PPCA. The first component (the one we defined as Extended Populism Index) has the
highest positive correlation with the standard Populism Index. Second, it is also able to explain the
highest amount of variance of the standard populism index, as it is reported by the R2 value. Hence,
the first component looks a suitable candidate as alternative and extended populism index.
Finally, we analyze whether the extended scores better identify populist parties and better correlate
with existing measures. First, for illustrative purpose, we rely on the same unsupervised machine-
learning algorithm to cluster political parties. Figure B-VIII provides the result of the cluster analysis.
The top panel shows the results obtained when accounting for 3 dimensions of populism (3C). The
left panel considers all election-party pairs and identifies three clusters of parties colored in gray, blue
and red. The top-right panel isolates election-party pairs with standard populism scores above the
one standard deviation threshold. It shows that populist parties tend to cluster in a specific upper
part of the artificial space, just as in Figure 2 when we account for two dimensions only. The bottom
panel shows the results obtained with fifteen dimensions of populism (15C). The bottom-right panel
shows that populist parties tend to cluster in a specific upper part of the artificial space, even if this
pattern is less clear-cut as with 2 or 3 dimensions.
Second, we apply a second-stage PPCA of the 3 or 15 indices computed from the MPD and retain
components with eigenvalues above one. We then define this first component of these PPCA as our
3C vs. 15C Populism Scores. These alternative scores not only have at their core the main features

62

Electronic copy available at: https://ssrn.com/abstract=4273442

 Table B-VII: Correlations between standard populism index and political dimensions

(1) (2) (3) (4) (5) (6) (7)

OLS OLS OLS OLS OLS OLS OLS
Dependent variable: standard populism index
Cpt 1 .30∗∗∗ .30∗∗∗
Populismext (.01) (.01)
Cpt 2 -.00 -.00
(.01) (.01)
Cpt 3 .35∗∗∗ .35∗∗∗
(.01) (.01)
Cpt 4 -.20∗∗∗ -.20∗∗∗
(.01) (.01)
Cpt 5 .00 .00
(.01) (.01)
Cpt 6 .23∗∗∗ .23∗∗∗
(.01) (.01)
Obs. 3860 3860 3860 3860 3860 3860 3860
R2 0.28 0.00 0.26 0.07 0.00 0.09 0.70

of the standard populism score (positively correlated with AES, CTP), but they also account for the
OPP component (3C) or for political parties’ position towards the whole spectrum of political issues
covered in MPD (15C).
Although the extended populism scores account for a larger number of political characteristics,
they do not provide better proxies for populism. Adding more information to the populism score can
create additional noise. In Table B-VIII, we compare the partial correlations between the standard
and extended populism scores and the alternative classifications and measures available in existing
literature. These partial correlations are the outcomes of Probit regressions when the dependent is
a dichotomous classification variables, and of OLS regressions when the dependent is a continuous
variable. In both case, the regression includes country and year fixed effects.
Whatever the alternative source, our standard populism score exhibits a greater correlation with
existing measures and experts’ views than the 3C and 15C extended scores. Adding the OPP compo-
nent usually reduces the partial correlation estimates, while roughly preserving the ratio of accurate
forecasts and pseudo-R2 . It is worth noticing that parties considered as populist by many experts
(such as the Movimento 5 Stelle in Italy, the Front National in France, or Podemos in Spain) exit the
list when OPP is included.28 Moreover, adding the whole set of information available in MPD strongly
deteriorates the correlation with existing classifications our measures. These regressions suggest that

28
This is driven by the fact that in more recent years several parties took a strong pro-environment stance,
which generates a lower 3C score to parties like the Movimento 5 Stelle.

63

Electronic copy available at: https://ssrn.com/abstract=4273442

Figure B-VIII: Unsupervised clustering analysis on three and fifteen selected dimensions of
populism

UKIP,2001 UKIP,2001

FPÖ,1994 FPÖ,1994

DF,2015 DF,2015
UKIP,2015 UKIP,2015
DF,1998 AfD,2017 DF,1998 AfD,2017
DF,2001
FN,2002FN,2007 DF,2001
FN,2002FN,2007
FN,1997
FN,2017
AfD,2013 FPÖ,2013 FPÖ,2013
DF,2011
FN,2012
FN,1993
FN,1988
DF,2005
DF,2007 FPÖ,2017 UKIP,2017FPÖ,1990
FPÖ,2008 FPÖ,1986
FN,1986
FPÖ,1971
FPÖ,2006 FPÖ,1975
FPÖ,1983
FPÖ,1999
FPÖ,2002
FPÖ,1979 FPÖ,1962

FPÖ,1966

FPÖ,1970

(a) All parties – 3 dimensions (b) Populist parties only – 3 dimensions

DF,2001
FN,1997 DF,2015
AfD,2017
FPÖ,1986 FPÖ,1994 UKIP,2001
AfD,2013
FPÖ,2013 DF,2001
FN,1997 DF,2015 FPÖ,1994 UKIP,2001
AfD,2013
FPÖ,2013
AfD,2017
FPÖ,1990
FPÖ,1971
FN,1993
FN,1988 DF,2005
DF,2011
DF,2007
UKIP,2017
FN,1986 UKIP,2015
FN,2012 FPÖ,1979FPÖ,1966 FPÖ,1962 FPÖ,1962
FPÖ,2006
DF,1998 FN,2007 FPÖ,1970
FPÖ,2017
FPÖ,2002FPÖ,1975
FN,2017 FPÖ,2008
FPÖ,1983
FN,2002
FPÖ,1999

(c) All parties – 15 dimensions (d) Populist parties only – 15 dimensions

Note: We perform a clustering analysis using the fifteen political indicators built from the MPD. The left
panel presents the space including all parties, while the right panel shows the space once we focus on populist
parties only (populism index above one standard deviation). Source: Authors’ elaboration on MPD.

our standard populism score is a relevant – and perhaps better – proxy for populism, and that there
is not need to exploit the whole amount of information available in MPD for approximating populism.

64

Electronic copy available at: https://ssrn.com/abstract=4273442

 Table B-VIII: Standard versus extended populism scores – Correlation

I. Van Kessel (2000-2013) II. Swank (1960-2015) III. PopuList (1989-2018)

Populist party (PRB) RW Populist party (PRB) Populist party (PRB)
(1) (2) (3) (4) (5) (6) (7) (8) (9)
Standard 0.695∗∗∗ 0.462∗∗∗ 0.550∗∗∗
(0.162) (0.113) (0.094)
3C 0.456∗∗∗ 0.409∗∗∗ 0.412∗∗∗
(0.102) (0.091) (0.075)
15C 0.266∗∗∗ 0.229∗∗∗ 0.256∗∗∗
(0.069) (0.082) (0.064)
Obs. 641 641 641 1657 1657 1657 1635 1635 1635
Countries 25 25 25 16 16 16 28 28 28
Country FE 3 3 3 3 3 3 3 3 3
Year FE 3 3 3 3 3 3 3 3 3
Pseudo-R2 0.17 0.16 0.11 0.17 0.19 0.13 0.17 0.17 0.13
RAF (%) 81.75 81.44 80.81 91.25 91.31 91.43 86.18 85.75 86.24

IV. GPop 1 (1960-2018) V. GPop 2 VI. CHES (1998-2018)

Average Populism
Populist party (PRB) People vs. Elite (OLS)
Speeches (OLS)
(10) (11) (12) (13) (14) (15) (16) (17) (18)
Standard 0.379∗∗∗ 0.121∗∗ 1.262∗∗∗
(0.082) (0.051) (0.210)
3C 0.291∗∗∗ 0.106∗∗∗ 0.602∗∗∗
(0.057) (0.030) (0.166)
15C 0.190∗∗∗ 0.039 0.646∗∗∗
(0.049) (0.029) (0.140)
Obs. 2850 2850 2850 101 101 101 176 176 176
Countries 36 36 36 31 31 31 28 28 28
Country FE 3 3 3 7 7 7 3 3 3
Year FE 3 3 3 3 3 3 3 3 3
Pseudo-R2 0.16 0.16 0.14
RAF (%) 88.74 88.46 88.56
R2 0.22 0.25 0.19 0.37 0.23 0.27
Note: In Cols. (1) to (12), we provide partial correlations between parties’ political induces and the probability of be-
ing coded as populist party or right wing populist party following the definition of Van Kessel (2015), Swank (2018),
Rooduijn et al. (2019) and Grzymala-Busse and McFaul (2020) and adopting a probit model. Each regression con-
trols for country and year fixed-effects. We also provides the ratio of accurate forecasts (RAF) between our estimated
model and actual data, using a predicted probability of 0.5 as threshold to define a party as populist. In Cols. (13) to
(15), we provide partial correlations between political indices and party leader’s speeches (Hawkins et al., 2019) after
controlling for year fixed-effects. In Cols. (16) to (18), we provide partial correlations between political indices and
expert evaluations of parties degree of populism (Bakker et al., 2015). Standard errors are clustered at country level.
Level of significance: * p<0.1, ** p<0.05, *** p<0.01. Source: Authors’ elaboration on data from XX.

65

Electronic copy available at: https://ssrn.com/abstract=4273442

C Stylized Facts by Country Group
C.1 Volume and Mean Margins of Populism
These aggregate trends mask significant disparities across countries. In Figure C-I, we distinguish
five types of countries, namely Western European countries (France, Germany and the UK), Euro-
pean Union countries characterized by rising votes for radical parties (Austria, Greece and Italy),
Eastern European countries (Czech Republic, Hungary and Poland), traditional settlement countries
(Australia, Canada and the U.S.), and Latin American countries (Argentina, Chile and Mexico). For
each group of countries, we plot the evolution of the volume of populism, the extensive and intensive
margins of populism in the left, middle and right panels, respectively.
The left panel shows large ups and downs in the volume of populism across elections in virtually all
countries. This is due to the fact that some populist parties appear and disappear, either because they
enter and exit our sample (remember that our sample only includes countries with at least one seat
in the Parliament), or because they moderate their anti-establishment and anti-corruption discourses
once they come to power or reach a certain level of popularity. This means that some parties classified
as populist in an election can be classified as non populist in a different election. Using a time-
invariant definition or score of populism would avoid such fluctuations, but it would also prevent us
from exploiting variations in populism attitudes over a long time span.
The mean margin does not rely on a dichotomous classification of parties and use the continuous
populism score. The right panel of Figure C-I shows that the evolution of the mean margin is smoother,
but large variations are observed in many countries.

66

Electronic copy available at: https://ssrn.com/abstract=4273442

Figure C-I: Stylized facts III – Volume and margins or populism for selected countries

60

1
Vote-weighted average of populism scores
Votes for populist parties (%)

.5
40

0
20

-.5
-1
0

1960 1970 1980 1990 2000 2010 2015 1960 1970 1980 1990 2000 2010 2015

France Germany United Kingdom France Germany United Kingdom

(a) Volume marg. in Western Europe (b) Mean marg. in Western Europe
80

1
Vote-weighted average of populism scores
Votes for populist parties (%)
60

.5
40

0
-.5
20

-1
0

1960 1970 1980 1990 2000 2010 2015 1960 1970 1980 1990 2000 2010 2015

Austria Greece Italy Austria Greece Italy

(c) Volume marg. in populist Europe (d) Mean marg. in populist Europe
80

1
Vote-weighted average of populism scores
Votes for populist parties (%)
60

.5
40

0
-.5
20

-1
0

1960 1970 1980 1990 2000 2010 2015 1960 1970 1980 1990 2000 2010 2015

Czech Republic Hungary Poland Czech Republic Hungary Poland

(e) Volume marg. in Eastern Europe (f) Mean marg. in Eastern Europe

67

Electronic copy available at: https://ssrn.com/abstract=4273442

Figure C-I: Stylized facts III – Volume and margins or populism for selected countries (cont’d)
80

Vote-weighted average of populism scores

1
Votes for populist parties (%)
60

.5
40

0
20

-.5
-1
0

1960 1970 1980 1990 2000 2010 2015 1960 1970 1980 1990 2000 2010 2015

Australia Canada United States Australia Canada United States

(g) Volume marg. in settlement countries (h) Mean marg. in settlement countries
40

.5
Vote-weighted average of populism scores
Votes for populist parties (%)
30

0
20
10

-.5
0

1960 1970 1980 1990 2000 2010 2015 1960 1970 1980 1990 2000 2010 2015

Argentina Chile Mexico Argentina Chile Mexico

(i) Volume marg. in Latin America (j) Mean marg. in Latin America

Note: The figures present the two margins (volume and mean) for a subset of countries from the rest
of the world. Source: Authors’ elaboration on MPD.

68

Electronic copy available at: https://ssrn.com/abstract=4273442

C.2 Long-run Trends in Globalization
Figure C-II describes globalization trends at the aggregate level. The top panel compares European
countries with the rest of the world. Both immigration and import trends are very similar across
regions, although their intensity varies. Panel (a) shows that the share of immigrants has gradually
increased since the mid-seventies, slightly decreased in the first half of the nineties, before increasing
again until the financial crisis of 2008. Post-1990 changes are more pronounced in the Europe as a
result of the enlargement of the European Union to Eastern Europe. With regard to imports, their
share in GDP remained stable from 1960 to 1990. A slight decrease is observed after the second oil
crisis. Trade growth has been more pronounced since the mid-nineties. Technological changes and
policy reforms (multilateral and bilateral negotiations at the WTO) have given the first impetus,
followed by the entry of China in WTO after 2000. Due to the financial economic crisis, this pace
has slowed down in recent years. Again, the recent increase in trade is more pronounced in European
Union countries. In the bottom panel, we split immigration and import flows by education level or
by level of development of the origin countries. Panel (c) evidences a gradual increase in low-skill
immigration between the early seventies and the financial crisis. The enlargement of the European
Union also materializes in rising immigration rates from middle-income countries to Europe after the
nineties. Panel (d) evidences a marked rise in imports of medium- and high-skill labor intensive goods
after the mid-nineties. To a lesser extent, imports of low-skill labor intensive goods have almost
doubled as well over the same period.
As low-skilled immigration and imports of low-skill labor intensive goods are shown to translate
into populist pressures. In Figure C-III, we focus on these two indicators and compare the trends
observed in the five groups of countries defined in Figure C-I, i.e., Western Europe, European countries
characterized by rising votes for radical parties, Eastern Europe, traditional settlement countries,
and Latin America. With regard to low-skill immigration, it has gradually increased in virtually all
countries since the early eighties. The highest levels are observed in settlement countries (Australia,
Canada and the U.S.), in the UK, Germany, Austria, Italy and Chile. The Czech Republic shows a
peak between 1995 and the financial crisis. The evolution of imports of low-skill labor intensive goods
follows even more homogeneous patterns. The share of imports in GDP has increased in all countries
since the early nineties. The most pronounced changes are observed in Eastern European countries,
Latin America, Austria, Canada and Australia. Our panel data analysis takes advantage of these huge
variations to identify the effect of globalization shocks on the margins of populism.

69

Electronic copy available at: https://ssrn.com/abstract=4273442

 Figure C-II: Stylized facts IV – Trade and immigration trends at the aggregate level

.5
.022

.4
.02
Migration (over pop.)

Imports (over GDP)

.018

.3
.016

.2
.014

.1
.012

0
1963 1970 1980 1990 2000 2010 2015 1963 1970 1980 1990 2000 2010 2015

Full sample EU28 RoW Full sample EU28 RoW

(a) Immigration by broad destination (b) Imports by broad destination

.015

.12
.1
Migration (over pop.)

Imports (over GDP)

.08
.01

.06
.005

.04
.02
0

1963 1970 1980 1990 2000 2010 2015 1963 1970 1980 1990 2000 2010 2015

Low-skill High-skill Low-skill High-skill

(c) Immigration by skill level (d) Imports by skill level

Note: Figures (a), (b), (c) and (d) show moving averages including 3 years before and 3 years after
each date. Source: Authors’ calculations on Abel (2018), Feenstra et al. (2005) and UN Comtrade.

70

Electronic copy available at: https://ssrn.com/abstract=4273442

Figure C-III: Stylized facts IV – Low-skill immigration and imports in selected countries

.04
.03
Low-skill migration (over pop.)

Low-skill imports (over GDP)

.025

.03
.02

.02
.015

.01
.01
.005

0
1960 1970 1980 1990 2000 2010 2015 1960 1970 1980 1990 2000 2010 2015

France Germany United Kingdom France Germany United Kingdom

(a) Immigration in Western Europe (b) Imports in Western Europe

.04

.06
Low-skill migration (over pop.)

Low-skill imports (over GDP)

.03

.04
.02

.02
.01
0

1960 1970 1980 1990 2000 2010 2015 1960 1970 1980 1990 2000 2010 2015

Austria Greece Italy Austria Greece Italy

(c) Immigration in populist Europe (d) Imports in populist Europe

.08
.02
Low-skill migration (over pop.)

Low-skill imports (over GDP)

.06
.015

.04
.01
.005

.02
0

1960 1970 1980 1990 2000 2010 2015 1960 1970 1980 1990 2000 2010 2015

Czech Republic Hungary Poland Czech Republic Hungary Poland

(e) Immigration in Eastern Europe (f) Imports in Eastern Europe

Note: Figures (a), (b), (c), (d), (e) and (f) show moving averages including 3 years before and 3
years after each date. Source: Authors’ calculations on Abel (2018), Feenstra et al. (2005) and UN
Comtrade.

71

Electronic copy available at: https://ssrn.com/abstract=4273442

Figure C-III: Stylized facts IV – Low-skill immigration and imports in selected countries (cont’d)

.03
.05
Low-skill migration (over pop.)

Low-skill imports (over GDP)

.04

.02
.03

.01
.02
.01

0
1960 1970 1980 1990 2000 2010 2015 1960 1970 1980 1990 2000 2010 2015

Australia Canada United States Australia Canada United States

(g) Immigration in settlement countries (h) Imports in settlement countries

.015

.02
Low-skill migration (over pop.)

Low-skill imports (over GDP)

.015
.01

.01
.005

.005
0

1960 1970 1980 1990 2000 2010 2015 1960 1970 1980 1990 2000 2010 2015

Argentina Chile Mexico Argentina Chile Mexico

(i) Immigration in Latin America (j) Imports in Latin America

Note: Figures (g), (h), (i) and (j) show moving averages including 3 years before and 3 years after
each date. Source: Authors’ calculations on Abel (2018), Feenstra et al. (2005) and UN Comtrade.

72

Electronic copy available at: https://ssrn.com/abstract=4273442

C.3 Right- and Left-Wing Populism Across Broad Regions
Compared with the core of the text, we plot the evolution of the margins of populism and number
of election with populist parties in the EU15 countries and in non-European countries. The EU15
countries are the member states of the European Union prior to the accession of ten candidate coun-
tries on 1 May 2004: Austria, Belgium, Denmark, Finland, France, Germany, Greece, Ireland, Italy,
Luxembourg, Netherlands, Portugal, Spain, Sweden, United Kingdom.

Figure C-IV: Evolution of Populism: EU15 vs. RoW

.4

20
Adj. average populism index (weighted)

Votes gained by populist parties (%)

.2

15
10
0
-.2

5
-.4

1960 1970 1980 1990 2000 2010 2020 1960 1970 1980 1990 2000 2010 2020

Full sample EU15 RoW Full sample EU15 RoW

(a) Mean margin of populism (ΠM

i,e,t ) (b) Volume margin of populism (ΠVi,e,t )
60
40

Election with RW populist party (%)

Election with LW populist party (%)

50
30

40
30
20

20
10

10
0

1960 1970 1980 1990 2000 2010 2020 1960 1970 1980 1990 2000 2010 2020

Full sample EU15 RoW Full sample EU15 RoW

(c) Elections with a LW populist party (%) (d) Elections with a RW populist party (%)

Fig. (a) plots the mean margin of populism, a weighted average of the populism scores with weights
equal to the party’s share in votes. Fig. (b) depicts the average share of votes for populist parties
(the volume margin). Fig. (c) and (d) give the percentage of elections with at least a left-wing and
right-wing Populist party, respectively. Populist parties are defined as those with a score exceeding 1
standard deviation (0.808), while left-wing and right-wing parties are defined as those that belongs
to the first and third tercile of the left-to-right index. Fig. (a), (b), (c) and (d) show moving averages
including 3 years before and 3 years after each date.

73

Electronic copy available at: https://ssrn.com/abstract=4273442

Figure C-V: Evolution of Populism: RW and LW Populist Parties across broad regions
.3
Adj. average populism index (weighted)

Votes gained by populist parties (%)

9
.2
.1

6
0

3
-.1
-.2

0
1960 1970 1980 1990 2000 2010 2020 1960 1970 1980 1990 2000 2010 2018

EU15-RW EU15-LW RoW-RW RoW-LW EU15-RW EU15-LW RoW-RW RoW-LW

(a) Mean margin of populism (ΠM

i,e,t ) (b) Volume margin of populism (ΠVi,e,t )

Fig. (a) plots the mean margin of populism, a weighted average of the populism scores with weights
equal to the party’s share in votes. Fig. (b) depicts the average share of votes for populist parties (the
volume margin). Populist parties are defined as those with a score exceeding 1 standard deviation
(0.808), while left-wing and right-wing parties are defined as those that belongs to the first and third
tercile of the left-to-right index. Fig. (a) and (b) show moving averages including 3 years before and
3 years after each date.

74

Electronic copy available at: https://ssrn.com/abstract=4273442

D Supplementary Empirical Results
D.1 Reduce-Form IV Regression: First-Stage Results
Table D-I shows the results of the related first stage. Observed import and immigration flows by skill
group are regressed on their predicted levels obtained after combining dyadic predictions from Eq. (7),
as well as on the control variables and fixed effects used in the second-stage Eq. (5). The predicted
levels are nicely correlated with the actual ones, and the coefficients of the instruments are highly
significant close to unity. The adjusted R-squared is usually large despite the fact that our zero-stage
dyadic regressions abstract from destination-time characteristics.

Table D-I: Actual and predicted flows of imports and immigrants

(1) (2) (3) (4)

ImpHS
i,e,t ImpLS
i,e,t MigHS
i,e,t MigLS
i,e,t

d HS
Imp 1.100***
i,e,t
(0.100)
d LS
Imp i,e,t 1.139***
(0.112)

d HS
Migi,e,t 1.235***
(0.113)
d LS
Mig 1.137***
i,e,t
(0.083)
Observations 575 575 575 575
Countries 52 52 52 52
Adj. R2 0.94 0.93 0.86 0.86
Year & country FE 3 3 3 3
Controls 3 3 3 3
Notes: ***, **, and * denote significance at the 1, 5, and 10 percent levels, respec-
tively; clustered standard errors at the country level are reported in parentheses; all
regressions have been estimated with OLS using the Stata command reghdfe.

75

Electronic copy available at: https://ssrn.com/abstract=4273442

D.2 Volume of Populism: Extensive and Intensive Margins
In this section, we focus on the volume of populism, as measured by the share of votes for populist
parties, and on its intensive and extensive margins. In Table D-II reports the PPML results.
Imports of low-skill labor intensive goods are positively and significantly associated with right-
and left-wing populism. The link with right-wing populism materializes through the intensive margin
(share of votes for existing populist parties), while the effect on left-wing populism is less significant
and linked to the extensive margin (number of populist parties). By contrast, imports of high-skill
labor intensive goods are associated with lower volumes of populism in general, and with lower levels
of right-wing populism in particular. The elasticity of the intensive margin of populism to imports of
low-skill labor intensive goods is usually greater than unity.
With regard to immigration, its association with the overall volume of populism is insignificant.
Our results support, however, a substitution between left-wing and right-wing populism. low-skill
immigration is associated with highest volumes of right-wing populism and with smallest volumes
for left-wing populism. This substitution operates along both extensive and intensive margins. By
contrast, high-skill immigration tends to generate opposite substitution from right-wing to left-wing
populism, although the effects are slightly smaller and less significant.
Table D-III presents the reduced-form IV estimates for the volume margin of populism and of its
two components. Focusing first on the volume of populism, the IV estimates are pretty much in line
with the results of our baseline PPML regressions. They confirm that the skill structure of global-
ization shocks plays a key role. Imports of low-skill labor intensive goods foster votes for right-wing
populist parties, and the effect mostly materializes through the intensive margin. By contrast, imports
of high-skill labor intensive goods decrease the votes for right-wing populist parties. With regard to
immigration, the IV results also confirm those of the baseline regressions. low-skill immigration leads
to a substitution of left-wing populism for right-wing populism. This effect mostly materializes along
the extensive margin (while it also affects both margins in baseline PPML regressions). High-skill im-
migration reduces the votes for (and number of) populist parties. Compared with baseline regressions,
the elasticities are larger by a factor of 1.3, which is in line with the existence of a reverse causation
link: the rise in populism could lead to greater trade and immigration restrictions.

76

Electronic copy available at: https://ssrn.com/abstract=4273442

 Table D-II: Baseline PPML results – Volume of populist votes and its margins

V )
Volume (Pi,e,t E )
Ext. margin (Pi,e,t I )
Int. margin (Pi,e,t
All RW LW All RW LW All RW LW
(1) (2) (3) (4) (5) (6) (7) (8) (9)
log GDP/capit -1.22 -2.46∗∗ 0.70 -0.93 -2.35∗∗∗ 0.94 -0.85 -1.82∗ -0.40
(0.95) (1.19) (1.38) (0.63) (0.88) (0.85) (0.79) (1.00) (1.54)
log Popit 1.28 1.00 2.98 0.04 -0.46 1.70 1.16 1.16 2.20
(0.96) (1.33) (1.84) (0.75) (1.20) (1.16) (0.86) (1.35) (1.39)
log HCit -4.81∗∗ -9.01∗∗∗ 5.06 -0.82 -7.21∗∗∗ 5.95∗∗ -6.01∗∗∗ -7.75∗∗ 3.04
(2.09) (3.41) (5.27) (1.73) (2.26) (3.03) (2.21) (3.19) (4.88)
log Empit /Popit -0.98 -0.15 -5.00 1.43 2.30 -3.73 -1.12 -0.90 -2.61
(1.46) (1.99) (3.65) (1.05) (1.83) (2.34) (1.43) (1.98) (3.11)
log Partiesit 0.45 0.51 0.83∗ 1.36∗∗∗ 1.29∗∗∗ 1.46∗∗∗ -0.05 0.05 0.41
(0.29) (0.50) (0.43) (0.24) (0.41) (0.38) (0.28) (0.53) (0.48)

log Impi,t (LS) 0.83∗∗∗ 1.33∗∗ 1.49∗∗ 0.36 0.66 0.86∗ 1.05∗∗∗ 1.60∗∗∗ 1.02
(0.30) (0.56) (0.62) (0.26) (0.46) (0.45) (0.35) (0.56) (0.78)
log Impi,t (HS) -0.71 -1.30∗∗∗ -1.25 -0.19 -0.45 -0.99 -0.94∗∗ -1.65∗∗∗ -0.46
(0.44) (0.49) (0.86) (0.37) (0.46) (0.69) (0.43) (0.52) (1.03)

log Migi,t (LS) 0.14 1.52∗∗∗ -1.78∗∗∗ -0.16 1.01∗∗ -1.14∗∗∗ 0.21 1.19∗∗ -1.55∗∗∗
(0.34) (0.55) (0.59) (0.29) (0.48) (0.42) (0.34) (0.52) (0.58)
log Migi,t (HS) -0.28 -1.32∗∗∗ 1.17∗ -0.12 -1.05∗∗ 0.71∗ -0.20 -1.09∗∗ 1.20∗
(0.29) (0.48) (0.64) (0.25) (0.41) (0.39) (0.34) (0.48) (0.65)
Observations 575 575 575 575 575 575 575 575 575
Pseudo-R2 0.40 0.37 0.51 0.30 0.27 0.31 0.34 0.33 0.44
Year FE 3 3 3 3 3 3 3 3 3
Country FE 3 3 3 3 3 3 3 3 3
Notes: ***, **, and * denote significance at the 1, 5, and 10 percent levels, respectively; clustered standard
errors at the country level are reported in parentheses; all regressions have been estimated with PPML
using the Stata command ppmlhdfe.

77

Electronic copy available at: https://ssrn.com/abstract=4273442

 Table D-III: IV – Volume of populist votes and its margins

V )
Volume (Pi,e,t E )
Ext. margin (Pi,e,t I )
Int. margin (Pi,e,t
All RW LW All RW LW All RW LW
(1) (2) (3) (4) (5) (6) (7) (8) (9)
log Imp
d i,t (LS) 0.91∗ 1.82∗∗ 0.97 0.62∗ 0.92 0.94 1.40∗∗∗ 2.10∗∗ 1.40
(0.50) (0.84) (0.84) (0.38) (0.67) (0.76) (0.51) (0.84) (0.89)
log Imp
d i,t (HS) -1.22∗ -2.14∗∗ -0.72 -0.96∗∗ -1.20 -1.12 -1.17∗∗ -2.16∗∗ -0.62
(0.66) (0.87) (0.83) (0.46) (0.80) (0.82) (0.58) (0.93) (0.91)

log Mig
d i,t (LS) 0.53 1.97∗∗∗ -1.70∗ 0.15 1.55∗∗∗ -1.33∗∗ 0.19 1.22∗ -1.35
(0.43) (0.58) (0.92) (0.35) (0.53) (0.66) (0.48) (0.72) (0.89)
log Mig
d i,t (HS) -1.04∗ -2.02∗∗ 0.60 -1.05∗∗ -2.44∗∗∗ 0.34 0.14 -0.86 0.93
(0.56) (0.89) (1.23) (0.43) (0.79) (0.75) (0.64) (0.97) (1.20)
Observations 575 575 575 575 575 575 575 575 575
Pseudo-R2 0.40 0.36 0.50 0.31 0.28 0.32 0.33 0.32 0.43
Year & country FE 3 3 3 3 3 3 3 3 3
Controls 3 3 3 3 3 3 3 3 3
Notes: ***, **, and * denote significance at the 1, 5, and 10 percent levels, respectively; clustered stan-
dard errors at the country level are reported in parentheses; all regressions have been estimated with
PPML using the Stata command ppmlhdfe.

78

Electronic copy available at: https://ssrn.com/abstract=4273442

D.3 Additional Results: Mean Margin of Populism
Table D-IV focuses on the association between globalization shocks and three alternative measures
of mean margin of populism. Interestingly, when we provide the split between left and right wing
populism, we have a lower number of observations, driven by country specific elections where no
party has such a strong ideological stance. The first three columns shows the association between
imports/immigration and the unweighted average level of populism of parties included in our sample.
In Cols. (4-6), we focus on the weighted average level of populism, using parties’ vote shares as weights.
However, since our data set includes parties that won at least one seat in the parliament, it excludes
small parties and most independent candidates running for election. Hence the cumulative vote share
is less than 100% for many election-year pairs. In the last three columns, we normalize the vote shares
of parties represented in the parliament so that their sum is equal to 100%.
Whatever the definition of the dependent variable, we find that imports of low-skill labor inten-
sive goods are positively and significantly associated with the mean margin of total and right-wing
populism. The elasticity is large, ranging from 3.5 to 7.5. These results point out that import shocks
positively influence the mean level of populism (i.e., the average supply of populism in a society), both
in raw terms and when we account for parties political relevance. By contrast, imports of high-skill
labor intensive goods and immigration rates are not significantly correlated with populism. In Panel
B of Table D-IV, we produce IV results using the same instruments as in the previous section, and
rely on a standard 2SLS approach. Panel B is in line with the OLS results.
In Table D-V, we investigate separately the effects of globalization shocks on the (vote-weighted)
mean populism score of parties that have never been classified as populist, and parties that have been
classified as populist in at least one election.

79

Electronic copy available at: https://ssrn.com/abstract=4273442

 M
Table D-IV: Mean margin of populism with alternative measures of Pi,e,t (OLS and 2SLS)

Parties Parliament Parliament (adj.)

All RW LW All RW LW All RW LW
(1) (2) (3) (4) (5) (6) (7) (8) (9)
Panel A: OLS

Impi,t (LS) 3.94∗∗ 7.68∗∗ 2.28 4.00∗∗ 7.62∗∗∗ 0.84 3.78∗∗ 4.28∗∗∗ -0.11
(1.95) (2.87) (2.40) (1.71) (2.73) (1.81) (1.65) (1.47) (0.70)
Impi,t (HS) -0.28 -0.53 0.32 -0.27 -0.72 0.20 -0.21 -0.50∗ 0.36
(0.40) (0.58) (0.60) (0.47) (0.61) (0.55) (0.43) (0.28) (0.23)

Migi,t (LS) -1.80 0.80 -6.56∗ -0.31 3.05 -6.24∗ -0.17 1.73 -1.28
(1.83) (4.58) (3.76) (2.07) (4.61) (3.41) (1.93) (2.45) (1.28)
Migi,t (HS) 0.03 -3.93 10.96 2.21 -7.35 12.71 1.86 -2.63 3.65
(6.35) (12.07) (11.07) (5.31) (11.96) (9.97) (4.99) (4.74) (3.49)

Observations 578 461 470 578 461 470 578 461 470
R2 0.55 0.47 0.53 0.50 0.46 0.52 0.50 0.41 0.48
Panel B: 2SLS

Imp
d i,t (LS) 5.77** 7.37* 7.35** 5.27** 4.13 6.03 4.99** 4.06** 1.29
(2.39) (4.08) (3.19) (2.48) (4.14) (3.86) (2.33) (1.77) (1.42)
Imp
d i,t (HS) -0.57 -1.12 0.23 -0.28 -0.70 0.34 -0.22 -0.59 0.45
(0.54) (0.87) (0.79) (0.59) (0.82) (0.85) (0.54) (0.38) (0.37)

Mig
d i,t (LS) -0.86 -0.90 -7.26* 0.42 -0.42 -6.05 0.52 0.74 -0.75
(2.89) (6.19) (4.32) (3.39) (5.74) (4.31) (3.12) (3.01) (1.53)
d i,t (HS)
Mig -1.27 -0.90 17.23 1.57 1.10 18.43 0.99 3.15 3.34
(10.84) (19.00) (12.84) (11.04) (19.03) (11.65) (10.12) (7.89) (4.75)

Observations 578 460 469 578 460 469 578 460 469
K-Paap F-stat 12.07 11.39 9.47 12.07 11.39 9.47 12.07 11.39 9.47
Year FE 3 3 3 3 3 3 3 3 3
Country FE 3 3 3 3 3 3 3 3 3
Notes: ***, **, and * denote significance at the 1, 5, and 10 percent levels, respectively; clustered
standard errors at the country level are reported in parentheses; all regressions have been estimated
with OLS and 2SLS using the Stata command reghdfe and ivreghdfe, in Panel A and B, respec-
tively.

80

Electronic copy available at: https://ssrn.com/abstract=4273442

Table D-V: Mean margin of populist and non-populist parties (OLS and 2SLS)

Never populist Populist at least once

All RW LW All RW LW
(1) (2) (3) (4) (5) (6)
Panel A: OLS

Impi,t (LS) 0.22 1.34 -1.29 7.36∗∗ 5.42 6.33

(0.91) (1.96) (1.12) (2.99) (5.57) (4.35)
Impi,t (HS) 0.12 -0.08 0.19 0.69 -0.86 1.28
(0.37) (0.38) (0.45) (0.93) (1.81) (1.70)
Migi,t (LS) -2.71∗ -3.13 -3.74∗ -3.81 -3.48 -11.74∗∗
(1.41) (3.76) (1.92) (5.10) (7.68) (5.33)
Migi,t (HS) 7.19 6.30 3.30 11.20 3.28 40.01∗∗
(4.87) (9.15) (10.57) (9.81) (17.62) (14.66)

Observations 527 325 364 470 293 294

R2 0.50 0.49 0.51 0.34 0.39 0.47
Panel B: 2SLS

Impi,t (LS) 0.77 1.57 -2.70 9.84** -7.45 19.63**

(1.73) (3.09) (2.26) (4.33) (11.15) (8.08)
Impi,t (HS) 0.15 -0.05 0.57 0.11 -3.10 0.39
(0.46) (0.55) (0.67) (1.21) (3.30) (1.71)
Migi,t (LS) -2.11 -4.87 -3.14 -3.02 -9.59 -10.49
(1.86) (5.50) (2.30) (6.74) (8.24) (7.48)
Migi,t (HS) 1.70 -1.39 -2.86 2.81 5.67 52.53**
(6.77) (18.51) (10.76) (18.24) (23.95) (23.60)

Observations 527 325 364 470 293 294

R2 0.02 -0.02 0.10 0.07 -0.03 0.00
K-Paap F-stat 8.82 3.54 9.20 23.94 8.33 22.76
Year & Country FE 3 3 3 3 3 3
Controls 3 3 3 3 3 3
Notes: ***, **, and * denote significance at the 1, 5, and 10 percent levels, respectively; clus-
tered standard errors at the country level are reported in parentheses; coefficients presented
in columns (1) to (6) have been estimated with 2SLS using the Stata command ivreghdfe.
The mean margin is computed over the sample of parties that is never classified as populist in
columns (1) to (3), while is computed over the sample of parties that is classified at least once
as populist in columns (4) to (6).

81

Electronic copy available at: https://ssrn.com/abstract=4273442

D.4 Treating Endogenous Variables Separately
Tables D-VI and D-VIII provide the results on the volume and mean margin once the skill-specific
import and migration flows are treated as endogenous variables separately and not simultaneously.
Although such assumption is rather counter intuitive, since there are no specific evidence that justify
an exclusive exogeneity of some skill-specific globalization shocks compared to the others, a consistency
in the estimated results would minimize concerns driven by the highly demanding econometric speci-
fication while instrumenting four endogenous variables simultaneously. The variable instrumented is:
low-skill import (cols. 1-3), high-skill import (cols. 4-6), low-skill immigration (cols. 7-9) and high-
skill immigration (cols. 10-12). The last three columns report the estimates once the four variables
are treated as endogenous simultaneously for a comparison purpose.
The direction of the correlations between skill-specific globalization shocks and the volume margin
is confirmed across specifications. However, the significance of the correlation of a skill-specific flow
is affected if only one skill-specific component is treated as endogenous. For instance, the positive
correlation of low-skill migration on right-wing populism is not statistically significant once only low-
skill immigration (cols. 7-9) or only high skill immigration is treated as endogenous (cols. 10-12).
Hence, treating the entire flows (either migration or import) as endogenous appears as an important
empirical choice, given the degree of correlation among trade and migration flows presented in Table
D-IX. Table D-VII confirms this intuition: once either imports or migration flows are treated as
endogenous, the estimates are consistent with our benchmark results.
Concerning the mean margin, Table D-VIII shows that the estimates are rather consistent disre-
garding the selection of endogenous variables. The F-stat reported in columns (1) to (12) suggest that
each instrument is strong enough for its corresponding endogenous variable. Moreover, columns (13)
to (15) report, as an alternative proxy of the strength of the instrumental variables, the Shea Partial
R2 (Shea, 1997) associated to each instrument once the other instrumental variables are partial out.
The values of the partial R2 fluctuates around 0.5, providing evidence of our instrumental variables
relevance.

82

Electronic copy available at: https://ssrn.com/abstract=4273442

 Table D-VI: Reduced-form IV PPML results – Volume (one endogenous variable)

All RW LW All RW LW All RW LW All RW LW All RW LW

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15)
Predicted Var. Impi,t (LS) Impi,t (HS) Migi,t (LS) Migi,t (HS) All

log Impi,t (LS) 0.85∗∗ 0.95∗ 1.29∗∗ 0.79∗∗∗ 1.23∗∗ 1.11∗ 0.75∗∗∗ 1.09∗∗ 1.39∗∗∗
(0.37) (0.50) (0.55) (0.29) (0.53) (0.59) (0.27) (0.52) (0.54)
log Impi,t (HS) 0.07 -0.27 0.01 -0.68 -1.25∗∗ -0.85 -0.64 -1.04∗∗ -1.18
(0.40) (0.41) (0.73) (0.42) (0.50) (0.78) (0.40) (0.48) (0.80)
log Migi,t (LS) 0.09 1.42∗∗∗ -1.67∗∗∗ 0.15 1.40∗∗∗ -1.62∗∗∗ 0.01 0.33 -0.58∗∗
(0.34) (0.55) (0.60) (0.33) (0.53) (0.59) (0.13) (0.25) (0.29)
log Migi,t (HS) -0.19 -1.17∗∗ 1.14 -0.28 -1.18∗∗∗ 1.05 -0.10 -0.11 0.01
(0.29) (0.48) (0.70) (0.27) (0.44) (0.64) (0.12) (0.22) (0.31)
log Imp
d
i,t (LS) 0.08 0.60 0.46 0.86∗ 1.74∗∗ 1.03
(0.47) (0.74) (0.92) (0.52) (0.86) (0.88)
log Imp
d
i,t (HS) -1.32∗ -1.30∗ -1.61 -1.19∗ -2.07∗∗ -0.79
(0.70) (0.72) (1.05) (0.67) (0.88) (0.84)
log Mig
d
i,t (LS) -0.17 0.62 -1.30∗∗ 0.52 1.95∗∗∗ -1.70∗
(0.34) (0.44) (0.59) (0.43) (0.58) (0.92)
log Mig
d
i,t (HS) -0.54 -0.47 -0.27 -1.04∗ -2.01∗∗ 0.60
(0.39) (0.52) (0.62) (0.56) (0.88) (1.23)
Observations 575 575 575 575 575 575 575 575 575 575 575 575 575 575 575
Pseudo-R2 0.39 0.35 0.50 0.40 0.36 0.52 0.40 0.35 0.51 0.40 0.35 0.50 0.40 0.36 0.50
Year & Country FE 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3
Controls 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3
Notes: ***, **, and * denote significance at the 1, 5, and 10 percent levels, respectively; clustered standard er-
rors at the country level are reported in parentheses; coefficients have been estimated with PPML using the Stata
command ppmlhdfe and predicted globalization variables from the model estimated in equation (7).

83

Electronic copy available at: https://ssrn.com/abstract=4273442

Table D-VII: Reduced-form IV PPML results – Volume (two endogenous variables)

All RW LW All RW LW
(1) (2) (3) (4) (5) (6)
Predicted Var. Impi,t (LS)(HS) Migi,t (LS)(HS)

log Imp
d i,t (LS) 0.87∗ 1.60∗ 1.57∗
(0.50) (0.86) (0.94)
log Imp
d i,t (HS) -1.19∗ -1.83∗∗ -1.68∗
(0.71) (0.82) (1.02)
log Migi,t (LS) 0.16 1.50∗∗∗ -1.52∗∗
(0.33) (0.55) (0.62)
log Migi,t (HS) -0.27 -1.25∗∗∗ 0.99
(0.28) (0.48) (0.72)
log Impi,t (LS) 0.78∗∗∗ 1.33∗∗ 1.16∗∗
(0.27) (0.54) (0.52)
log Impi,t (HS) -0.66∗ -1.36∗∗∗ -0.86
(0.40) (0.52) (0.78)
log Mig
d i,t (LS) 0.43 1.90∗∗∗ -1.89∗∗
(0.42) (0.59) (0.96)
log Mig
d i,t (HS) -0.97∗ -1.99∗∗ 0.85
(0.51) (0.81) (1.19)
Observations 575 575 575 575 575 575
Pseudo-R2 0.39 0.36 0.50 0.40 0.37 0.51
Year & Country FE 3 3 3 3 3 3
Controls 3 3 3 3 3 3
Notes: ***, **, and * denote significance at the 1, 5, and 10 percent levels, respec-
tively; clustered standard errors at the country level are reported in parentheses;
coefficients have been estimated with PPML using the Stata command ppmlhdfe
and predicted globalization variables from the model estimated in equation (7).

84

Electronic copy available at: https://ssrn.com/abstract=4273442

 Table D-VIII: IV results – Mean Margin

All RW LW All RW LW All RW LW All RW LW All RW LW

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15)
Predicted Var. Impi,t (LS) Impi,t (HS) Migi,t (LS) Migi,t (HS) All

Impi,t (LS) 5.56∗∗ 3.54∗ 2.37 3.49∗∗ 4.60∗∗∗ -0.61 3.79∗∗ 4.33∗∗∗ -0.11 3.77∗∗ 4.05∗∗ -0.12 5.04∗∗ 3.92∗∗ 1.43
(2.52) (1.93) (1.66) (1.64) (1.52) (0.73) (1.66) (1.48) (0.70) (1.67) (1.52) (0.71) (2.34) (1.78) (1.45)
Impi,t (HS) -0.47 -0.38 0.02 -0.01 -0.68∗ 0.70∗ -0.21 -0.51∗ 0.36 -0.21 -0.45 0.36 -0.23 -0.58 0.45
(0.51) (0.35) (0.28) (0.52) (0.38) (0.39) (0.43) (0.27) (0.23) (0.43) (0.30) (0.23) (0.54) (0.38) (0.38)
Migi,t (LS) -0.13 1.55 -1.25 -0.25 1.74 -1.30 0.52 2.46 -0.76 -0.39 -0.77 -1.48 0.53 0.71 -0.74
(1.94) (2.51) (1.27) (1.93) (2.47) (1.31) (3.94) (3.36) (1.94) (3.65) (3.22) (1.69) (3.13) (3.02) (1.54)
Migi,t (HS) 1.61 -1.91 3.28 2.15 -2.56 3.68 0.08 -4.31 2.16 2.77 6.59 4.52 0.99 3.19 3.35
(5.06) (4.98) (3.70) (5.00) (4.80) (3.52) (10.31) (7.82) (5.34) (14.06) (9.15) (6.36) (10.12) (7.90) (4.77)
Observations 578 461 470 578 461 470 578 461 470 578 461 470 578 461 470
R2 0.06 0.09 -0.01 0.07 0.09 0.02 0.07 0.09 0.02 0.07 0.08 0.02 0.06 0.09 0.00
K-Paap F-stat 83.78 38.85 52.68 293.00 269.67 324.80 61.84 24.65 53.64 33.64 20.34 90.46 12.00 12.76 9.66
Shea Partial R2 ImpLS 0.49 0.41 0.48
Shea Partial R2 ImpHS 0.74 0.70 0.76
Shea Partial R2 M igLS 0.64 0.57 0.69
Shea Partial R2 M igHS 0.51 0.50 0.62
Year & Country FE 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3
Controls 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3
Notes: ***, **, and * denote significance at the 1, 5, and 10 percent levels, respectively; clustered standard er-
rors at the country level are reported in parentheses; coefficients have been estimated with 2SLS using the Stata
command ivreghdfe.

Table D-IX: Correlations globalization flows

Actual Flows (logs) Predicted Flows (logs)

(1) (2) (3) (1) (2) (3)
Imp (LS) Imp (HS) Mig (LS) Imp (LS) Imp (HS) Mig (LS)
Imp (HS) 0.8590**** Imp (HS) 0.8314***
Mig (LS) 0.2572*** 0.3097*** Mig (LS) 0.2304*** 0.3263***
Mig (HS) 0.1485**** 0.2135*** 0.9265*** Mig (HS) 0.1169*** 0.2226*** 0.9314***
Actual Flows Predicted Flows
(1) (2) (3) (1) (2) (3)
Imp (LS) Imp (HS) Mig (LS) Imp (LS) Imp (HS) Mig (LS)
Imp (HS) 0.7201**** Imp (HS) 0.7435***
Mig (LS) 0.2562*** 0.2930*** Mig (LS) 0.2110*** 0.2871***
Mig (HS) 0.0554 0.1211*** 0.7106*** Mig (HS) 0.0003 0.0884** 0.6734***

85

Electronic copy available at: https://ssrn.com/abstract=4273442

D.5 Additional Results: Globalization and Turnout
Table D-X explores the potential implication of globalization shocks on electoral participation. Relying
on the Voting Turnout Database of the International Institute for Democracy and Electoral Assistance
(IDEA), which documents electoral participation in parliamentary and presidential elections from 1945,
we compare the skill-specific effect of immigration and imports estimated in our full sample of countries
(Cols. 1-2) and in the sample of countries where voting is not compulsory (Cols. 3-4). Moreover, we
use two complementary proxies for electoral participation: the total number of votes divided by the
total number of names in the voters’ register (Cols. 1 and 3), and the total number of votes divided
by the population in age of voting (Cols. 2 and 4). While the first dependent variable relies on the
standard definition of voting turnout, the second one accounts (labeled as VAP Turnout) for the fact
that voters’ registration is not always reliable or that some individuals face unexpected problems when
enrolling in electoral register. Nonetheless, the two variables are highly correlated (0.833).
Whatever the definition or the sample, we find that imports are not significantly correlated with
turnout. Concerning immigration, the results are sensible to the sample and the definition. Immi-
gration of low-skill workers is positively and significantly correlated with voting turnout in the overall
sample, however the correlation is not statistically different from zero in the other specifications.
Similarly, inflows of highly educated immigrants is negatively correlated with electoral participation,
however it is statistically different from zero only among countries with a not compulsory voting system
and on the standard definition of voting turnout. Overall, these results suggest that the implication
of globalization shocks on voting turnout are not driving our results.
Alternatively, Table D-XI includes the standard measure of voting turnout as additional control in
our benchmark specification. Although being a “bad control” due to the simultaneous determination of
the populism variables and voting turnout, the skill-specific globalization estimates are not influenced
by the inclusion of electoral participation as a potential confounding factor. Moreover, turnout is not
significantly correlated with any margin of populism.

86

Electronic copy available at: https://ssrn.com/abstract=4273442

 Table D-X: Turnout and Globalization (2SLS)

All Countries Not Compulsory Voting

Turnout VAP Turnout Turnout VAP Turnout
(1) (2) (3) (4)
log GDP/capitait 0.00 0.03 0.01 0.08∗
(0.03) (0.04) (0.04) (0.04)
log Popit 0.16∗∗ 0.16∗∗∗ 0.03 0.06
(0.07) (0.06) (0.07) (0.06)
log HCit -0.17 -0.05 -0.49∗∗∗ -0.42∗∗∗
(0.17) (0.16) (0.14) (0.12)
log Empit /Popit 0.00 -0.06 0.11 0.06
(0.10) (0.08) (0.09) (0.08)
log Partiesit -0.03∗∗ -0.04∗∗∗ -0.03 -0.05∗∗∗
(0.01) (0.01) (0.02) (0.02)

Impi,t (LS) -0.18 -0.16 -0.63∗ -0.19

(0.29) (0.31) (0.37) (0.40)
Impi,t (HS) 0.07 0.07 -0.01 -0.00
(0.09) (0.09) (0.12) (0.13)
Migi,t (LS) 1.08∗∗ 0.48 0.85 0.31
(0.49) (0.43) (0.59) (0.54)
Migi,t (HS) -1.93 -0.61 -2.96∗∗ -0.62
(1.55) (1.23) (1.37) (1.21)

Observations 558 557 441 441

R2 0.09 0.08 0.17 0.12
K-Paap F-stat 8.66 8.62 38.72 39.73
Year & Country FE 3 3 3 3
Controls 3 3 3 3
Notes: ***, **, and * denote significance at the 1, 5, and 10 percent levels, respectively;
clustered standard errors at the country level are reported in parentheses; coefficients
have been estimated with 2SLS using the Stata command ivreghdfe. The dependent
variables is: the total number of votes divided by the number of names in voters’ regis-
ter (col. (1) and (3)) and the total number of votes divided by the population in age of
voting (col. (2) and (4)). The sample includes: all available countries in columns (1) and
(2), while only countries where voting is not compulsory in columns (3) and (4).

87

Electronic copy available at: https://ssrn.com/abstract=4273442

Table D-XI: Reduced-form IV PPML and 2SLS results – Controlling for Turnout

Volume (ΠVi,e,t ) Mean margin (ΠM

i,e,t )
All RW LW All RW LW
(1) (2) (3) (4) (5) (6)
Turnout -0.21 1.29 -0.30 0.28 0.34 0.17
(1.54) (1.93) (2.13) (0.28) (0.23) (0.24)
log Imp
d i,t (LS) 1.01∗ 1.70∗ 0.78
(0.55) (0.87) (0.91)
log Imp
d i,t (HS) -1.41∗∗ -1.69∗ -1.34
(0.69) (0.90) (0.96)
log Mig
d i,t (LS) 0.36 1.68∗∗∗ -2.16∗∗
(0.44) (0.62) (0.96)
log Mig
d i,t (HS) -0.95 -1.94∗∗ 1.19
(0.58) (0.89) (1.18)
Impi,t (LS) 5.06∗∗ 4.34∗∗ 1.34
(2.21) (1.66) (1.44)
Impi,t (HS) -0.30 -0.59 0.39
(0.56) (0.37) (0.38)
Migi,t (LS) -0.23 -0.76 -0.76
(3.11) (3.12) (1.60)
Migi,t (HS) 1.95 4.38 3.51
(9.70) (7.89) (4.70)

Observations 555 555 555 558 443 459

Pseudo-R2 0.39 0.36 0.51
R2 0.07 0.09 0.01
K-Paap F-stat 8.52 18.32 8.47
Year & Country FE 3 3 3 3 3 3
Controls 3 3 3 3 3 3
Notes: ***, **, and * denote significance at the 1, 5, and 10 percent levels, respectively; clus-
tered standard errors at the country level are reported in parentheses; coefficients presented
in column (1) to (3) have been estimated with PPML using the Stata command ppmlhdfe and
predicted globalization variables from the model estimated in equation (7), while coefficients
in column (4) to (6) have been estimated with 2SLS using the Stata command ivreghdfe.

88

Electronic copy available at: https://ssrn.com/abstract=4273442

D.6 Additional Results: Role of Electoral System
Tables D-XII and D-XIII explore the potential implications driven by the country-specific institutional
setting defining the electoral rules. Relying on the Electoral System Design database developed by
the International Institute for Democracy and Electoral Assistance (IDEA) (Reynolds et al., 2008),
we collect information on countries’ electoral system from 1990 to recent years, and we construct a
dummy variable that takes a value of one if the electoral system is characterized by a proportional
representation (P R).
Proportional representation implies a direct translation of the votes for a party into a corresponding
proportion of seats in the parliament. It might be argued that new and small populist parties benefit
from such type of electoral system. Due to the lack of information on the pre-1990 period, we impute
the electoral system of each country over such period based on their electoral system in the first
available election year. Table D-XII shows that controlling for having a proportional system do not
influence the skill-specific effect of migration and imports on the volume and mean margins of populism.
Additionally, Table D-XIII includes interaction terms with low-skill specific globalization shocks.
Interestingly, the results show that imports have a strong and positive effect on the left-wing volume
margin in countries with a proportional representation, while there is no specific effect on right-wing
margins. This result suggests that left-wing populist parties, in presence of skill-specific import shocks,
are particularly able to exploit the institutional setting to enhance their electoral gains.

89

Electronic copy available at: https://ssrn.com/abstract=4273442

Table D-XII: Reduced-form IV PPML and 2SLS results – Controlling for PR

Volume (ΠVi,e,t ) Mean margin (ΠM

i,e,t )
All RW LW All RW LW
(1) (2) (3) (4) (5) (6)
PR -1.11∗ -1.72∗ 0.27 -0.14 -0.07 0.02
(0.64) (0.95) (0.86) (0.09) (0.08) (0.08)
log Imp
d i,t (LS) 0.83 1.81∗∗ 1.01
(0.52) (0.82) (0.92)
log Imp
d i,t (HS) -1.07 -2.08∗∗ -0.81
(0.67) (0.90) (0.87)
log Mig
d i,t (LS) 0.44 1.88∗∗∗ -1.68∗
(0.44) (0.60) (0.92)
log Mig
d i,t (HS) -1.09∗∗ -2.20∗∗∗ 0.61
(0.54) (0.80) (1.23)
Impi,t (LS) 5.02∗∗ 3.87∗∗ 1.43
(2.34) (1.78) (1.45)
Impi,t (HS) -0.24 -0.58 0.45
(0.54) (0.38) (0.38)
Migi,t (LS) 0.30 0.56 -0.70
(3.17) (3.06) (1.56)
Migi,t (HS) 1.36 3.36 3.26
(10.12) (7.93) (4.84)
Observations 575 575 575 578 461 470
Pseudo-R2 0.41 0.37 0.50
R2 0.07 0.09 0.00
K-Paap F-stat 11.84 13.17 9.48
Year & Country FE 3 3 3 3 3 3
Controls 3 3 3 3 3 3
Notes: ***, **, and * denote significance at the 1, 5, and 10 percent levels, respectively; clus-
tered standard errors at the country level are reported in parentheses; coefficients presented in
column (1) to (3) have been estimated with PPML using the Stata command ppmlhdfe and
predicted globalization variables from the model estimated in equation (7), while coefficients
in column (4) to (6) have been estimated with 2SLS using the Stata command ivreghdfe.

90

Electronic copy available at: https://ssrn.com/abstract=4273442

Table D-XIII: Reduced-form IV PPML and 2SLS results – Interactions with PR

Volume (ΠVi,e,t ) Mean margin (ΠM

i,e,t )
All RW LW All RW LW
(1) (2) (3) (4) (5) (6)
log Imp
d i,t (LS) 0.69 2.09∗∗ 0.34
(0.60) (0.93) (1.11)
log Imp
d i,t (HS) -0.91 -2.11∗∗ -1.06
(0.65) (0.95) (0.96)
log Mig
d i,t (LS) -0.02 1.51∗∗ -2.52∗∗∗
(0.43) (0.67) (0.90)
log Mig
d i,t (HS) -1.18∗∗ -2.24∗∗∗ 1.47
(0.57) (0.79) (1.10)
d i,t (LS) × PR
log Imp 0.29 -0.32 2.76∗∗∗
(0.49) (0.49) (0.83)
d i,t (LS) × PR
log Mig 0.76∗ 0.48 0.65
(0.44) (0.57) (0.71)
Impi,t (LS) 6.71∗ 5.40∗ -0.32
(3.40) (2.96) (2.50)
Impi,t (HS) -0.18 -0.54 0.45
(0.53) (0.39) (0.37)
Migi,t (LS) 4.53 2.86 -0.63
(3.28) (3.21) (2.01)
Migi,t (HS) 0.07 3.16 3.18
(10.35) (8.05) (4.32)
Impi,t (LS) × PR -1.74 -1.70 1.68
(2.44) (2.30) (1.79)
Migi,t (LS) × PR -4.35∗ -2.76 -0.03
(2.45) (2.45) (2.29)
Observations 575 575 575 578 461 470
Pseudo-R2 0.41 0.38 0.53
R2 0.06 0.09 0.01
K-Paap F-stat 7.66 7.94 6.04
Year & Country FE 3 3 3 3 3 3
Controls 3 3 3 3 3 3
Notes: ***, **, and * denote significance at the 1, 5, and 10 percent levels, respectively; clustered
standard errors at the country level are reported in parentheses; coefficients presented in column
(1) to (3) have been estimated with PPML using the Stata command ppmlhdfe and predicted
globalization variables from the model estimated in equation (7), while coefficients in column (4)
to (6) have been estimated with 2SLS using the Stata command ivreghdfe.

91

Electronic copy available at: https://ssrn.com/abstract=4273442

D.7 Additional Results: Adding Emigration and Exports
We complement our analysis of the effect of trade and migration on the dynamics of populism by
including in our set of explanatory variables skill-specific emigration and export flows. Given the
bilateral dimension of our skill-specific migration and trade data, the construction of the variables
as outflows rather than inflows is simply determined by aggregating the dyadic levels of trade and
migration from the origin-country perspective, rather from the destination-country perspective. The
objective of this extension is to investigate whether the inclusion of emigration and export influences
our skill-specific results driven by immigration and imports. We treat emigration and export shocks
as exogenous, as endogenizing eight variables simultaneously would be heroic.
We first explore in Table D-XIV the skill-specific effect of outflows on the volume and mean margin
with a standard PPML/OLS framework, since endogeneity driven by reverse-causation is likely to be
less salient in this context. Note that (Dancygier et al., 2022) find a relationship between populism
and emigration, but causation is hard to establish and we control for an important mechanism of
transmission of emigration shocks, namely the level of human capital. Our estimates show a positive
and statistically significant relationship between the volume of left-wing populism and exports of high-
skill intensive goods or low-skill emigration. We do not find significant correlation for the volume of
overall or right-wing populism, nor for the mean margin. These results suggests that emigration and
exports are correlated with the left-wing dimension of populism, which can potentially be due to the
influence of unobserved factors.
Going one step further, Table D-XV includes simultaneously the skill-specific inflows and outflows
of trade and migration in a standard PPML/OLS framework. Importantly, the baseline effects of low-
skill immigration and imports are confirmed for both volume and mean margins of populism. Moreover,
the positive relationship between the volume of left-wing populism and exports (both low and high-
skill intensive) or low-skill emigration is also confirmed. Right-wing populism is less responsive to
outflows of goods and people. Table D-XVI shows that those findings are also confirmed – although
being less precisely estimated – once we instrument skill-specific immigration and import shocks.

92

Electronic copy available at: https://ssrn.com/abstract=4273442

 Table D-XIV: PPML and OLS results – Export and Emigration

Volume (ΠVi,e,t ) Mean margin (ΠM

i,e,t )
All RW LW All RW LW
(1) (2) (3) (4) (5) (6)
log Expi,t (LS) 0.08 0.19 0.51
(0.24) (0.33) (0.34)
log Expi,t (HS) 0.02 -0.64∗ 0.79∗∗∗
(0.25) (0.36) (0.30)
log Emigi,t (LS) 0.36 -0.20 1.37∗∗
(0.41) (0.49) (0.70)
log Emigi,t (HS) -0.02 0.51 -0.76
(0.41) (0.53) (0.67)
Expi,t (LS) -0.18 1.22 -0.12
(0.98) (1.04) (0.48)
Expi,t (HS) -0.11 -0.04 0.12
(0.28) (0.26) (0.11)
Emigi,t (LS) 2.17 -2.75 2.39
(2.60) (2.33) (1.56)
Emigi,t (HS) -9.71 7.43 -3.68
(10.89) (8.60) (5.51)
Observations 570 570 570 578 461 470
Pseudo-R2 0.41 0.35 0.54
R2 0.49 0.38 0.49
Year & Country FE 3 3 3 3 3 3
Controls 3 3 3 3 3 3
Notes: ***, **, and * denote significance at the 1, 5, and 10 percent levels, respectively;
clustered standard errors at the country level are reported in parentheses; coefficients pre-
sented in column (1) to (3) have been estimated with PPML using the Stata command
ppmlhdfe and predicted globalization variables from the model estimated in equation (7),
while coefficients in column (4) to (6) have been estimated with 2SLS using the Stata com-
mand ivreghdfe.

93

Electronic copy available at: https://ssrn.com/abstract=4273442

Table D-XV: PPML and OLS results – Import, Immigration, Export and Emigration

Volume (ΠVi,e,t ) Mean margin (ΠM

i,e,t )
All RW LW All RW LW
(1) (2) (3) (4) (5) (6)
log Impi,t (LS) 1.17∗∗∗ 1.37∗∗ 1.99∗∗∗
(0.39) (0.54) (0.67)
log Impi,t (HS) -0.92∗ -0.83 -1.90∗∗
(0.52) (0.61) (0.81)
log Expi,t (LS) 0.00 -0.05 0.71∗∗
(0.24) (0.33) (0.33)
log Expi,t (HS) 0.11 -0.66∗∗ 0.78∗∗∗
(0.24) (0.33) (0.27)
log Emigi,t (LS) 0.42 -0.36 1.35∗∗∗
(0.41) (0.52) (0.50)
log Emigi,t (HS) -0.02 0.70 -0.73
(0.43) (0.56) (0.50)
log Migi,t (LS) 0.02 1.48∗∗∗ -2.09∗∗∗
(0.30) (0.53) (0.62)
log Migi,t (HS) -0.02 -1.06∗∗ 1.59∗∗
(0.26) (0.49) (0.63)
Impi,t (LS) 4.75∗∗ 4.16∗∗ -0.04
(2.06) (1.68) (0.94)
Impi,t (HS) -0.24 -0.42 0.34
(0.55) (0.46) (0.31)
Expi,t (LS) -1.90 -0.07 -0.05
(1.20) (1.10) (0.63)
Expi,t (HS) -0.09 -0.05 0.01
(0.27) (0.39) (0.12)
Migi,t (LS) 0.09 1.71 -1.47
(1.76) (2.49) (1.25)
Migi,t (HS) 2.28 -2.37 4.16
(4.51) (5.16) (3.46)
Emigi,t (LS) 2.68 -1.84 2.33
(2.46) (2.03) (1.43)
Emigi,t (HS) -11.75 3.76 -2.76
(10.27) (7.62) (5.06)
Observations 567 567 567 578 461 470
Pseudo-R2 0.43 0.39 0.59
R2 0.51 0.41 0.49
Year & Country FE 3 3 3 3 3 3
Controls 3 3 3 3 3 3
Notes: ***, **, and * denote significance at the 1, 5, and 10 percent levels, respectively; clus-
tered standard errors at the country level are reported in parentheses; coefficients presented in
column (1) to (3) have been estimated with PPML using the Stata command ppmlhdfe and
predicted globalization variables from the model estimated in equation (7), while coefficients
in column (4) to (6) have been estimated with 2SLS using the Stata command ivreghdfe.

94

Electronic copy available at: https://ssrn.com/abstract=4273442

Table D-XVI: Reduced-form IV PPML and 2SLS results – Import, Immigration (endogenous)
Export and Emigration (exogenous)

Volume (ΠVi,e,t ) Mean margin (ΠM

i,e,t )
All RW LW All RW LW
(1) (2) (3) (4) (5) (6)
log Imp
d i,t (LS) 1.53∗∗∗ 1.70∗ 1.90∗∗
(0.56) (0.89) (0.92)
log Imp
d i,t (HS) -1.79∗∗ -1.46 -2.40∗∗
(0.79) (1.17) (1.20)
log Expi,t (LS) 0.07 0.01 0.70∗∗
(0.24) (0.31) (0.33)
log Expi,t (HS) 0.20 -0.62 1.21∗∗∗
(0.25) (0.40) (0.37)
log Emigi,t (LS) 0.33 -0.31 1.43∗∗∗
(0.35) (0.49) (0.50)
log Emigi,t (HS) 0.04 0.59 -0.82∗
(0.38) (0.55) (0.47)
d i,t (LS)
log Mig 0.32 1.94∗∗∗ -2.30∗∗∗
(0.38) (0.64) (0.86)
log Mig
d i,t (HS) -0.59 -1.77∗ 1.24
(0.54) (1.01) (0.96)
Impi,t (LS) 7.20∗∗ 4.38∗∗ 1.97
(2.97) (2.11) (2.09)
Impi,t (HS) -0.45 -0.92 0.41
(0.79) (0.74) (0.66)
Migi,t (LS) 0.19 0.81 -1.32
(2.81) (3.08) (1.53)
Migi,t (HS) 1.25 4.02 3.68
(9.40) (8.24) (4.72)
Expi,t (LS) -2.63∗∗ -0.31 -0.65
(1.19) (1.12) (0.85)
Expi,t (HS) -0.03 0.30 -0.03
(0.26) (0.51) (0.17)
Emigi,t (LS) 2.69 -1.56 2.42
(2.43) (2.06) (1.49)
Emigi,t (HS) -11.83 2.84 -3.08
(10.00) (8.11) (5.39)
Observations 567 567 567 572 461 464
Pseudo-R2 0.43 0.38 0.58
R2 0.07 0.09 0.01
K-Paap F-stat 16.89 29.23 8.83
Year & Country FE 3 3 3 3 3 3
Controls 3 3 3 3 3 3
Notes: ***, **, and * denote significance at the 1, 5, and 10 percent levels, respectively; clus-
tered standard errors at the country level are reported in parentheses; coefficients presented in
column (1) to (3) have been estimated with PPML using the Stata command ppmlhdfe and
predicted globalization variables from the model estimated in equation (7), while coefficients
in column (4) to (6) have been estimated with 2SLS using the Stata command ivreghdfe.
95

Electronic copy available at: https://ssrn.com/abstract=4273442

D.8 Additional Results: Detailed Robustness Checks
In the subsections below, we conduct a robustness analysis and produce results with alternative lag
structure for computing globalization shocks, alternative party classifications, alternative measures of
migration shocks (including interactions between migration inflows and stocks) and import shocks,
alternative classification of low-skill intensive shocks, interactions with period and region dummies.

D.8.1 Alternative Lag Structures

Table D-XVII: IV results with globalization shocks at time t

Volume (ΠVi,e,t ) Mean margin (ΠM

i,e,t )
All RW LW All RW LW
(1) (2) (3) (4) (5) (6)
log Imp
d it (LS) 0.62 1.02 0.66
(0.45) (0.67) (0.67)
log Imp
d it (HS) -0.98 -0.89 -1.16
(0.83) (0.85) (0.96)
log Mig
d it (LS) 0.41 1.83*** -1.86**
(0.44) (0.52) (0.86)
d it (HS)
log Mig -1.02* -1.85** 0.55
(0.53) (0.78) (1.06)
Impit (LS) 8.80* 7.05* 1.33
(4.85) (3.61) (2.76)
Impit (HS) -0.31 -0.77 0.91
(1.07) (0.63) (0.74)
Migit (LS) 1.04 0.39 -1.83
(6.68) (6.32) (3.48)
Migit (HS) -1.78 5.81 4.54
(24.19) (16.74) (10.03)
Observations 586 586 586 586 472 473
Pseudo-R2 0.40 0.35 0.51
R2 0.05 0.08 0.02
K-Paap F-stat 11.16 10.48 9.11
Year & Country FE 3 3 3 3 3 3
Controls 3 3 3 3 3 3
Notes: ***, **, and * denote significance at the 1, 5, and 10 percent levels, respectively; clustered
standard errors at the country level are reported in parentheses; coefficients presented in column
(1) to (3) have been estimated with PPML using the Stata command ppmlhdfe and predicted
globalization variables from the model estimated in equation (7), while coefficients in column (4)
to (6) have been estimated with 2SLS using the Stata command ivreghdfe.

96

Electronic copy available at: https://ssrn.com/abstract=4273442

 Table D-XVIII: IV results with globalization shocks at time t − 1

Volume (ΠVi,e,t ) Mean margin (ΠM

i,e,t )
All RW LW All RW LW
(1) (2) (3) (4) (5) (6)
log Imp
d i,t−1 (LS) 1.06** 1.99** 0.71
(0.45) (0.79) (0.79)
log Imp
d i,t−1 (HS) -1.52*** -2.27*** -1.34
(0.47) (0.66) (1.02)
log Mig
d i,t−1 (LS) 0.62 2.24*** -1.95**
(0.42) (0.65) (0.94)
log Mig
d i,t−1 (HS) -1.23** -2.30** 0.62
(0.56) (0.95) (1.17)
Impi,t−1 (LS) 7.87* 7.82* 1.62
(4.16) (4.02) (2.40)
Impi,t−1 (HS) -0.21 -1.09 0.86
(1.05) (0.79) (0.67)
Migi,t−1 (LS) -0.40 0.70 -1.71
(6.20) (6.02) (3.13)
Migi,t−1 (HS) 3.75 6.46 5.99
(19.10) (16.42) (9.12)
Observations 572 572 572 572 461 464
Pseudo-R2 0.41 0.37 0.51
R2 0.06 0.06 0.01
K-Paap F-stat 12.51 13.81 9.60
Year & Country FE 3 3 3 3 3 3
Controls 3 3 3 3 3 3
Notes: ***, **, and * denote significance at the 1, 5, and 10 percent levels, respectively; clustered
standard errors at the country level are reported in parentheses; coefficients presented in column (1)
to (3) have been estimated with PPML using the Stata command ppmlhdfe and predicted global-
ization variables from the model estimated in equation (7), while coefficients in column (4) to (6)
have been estimated with 2SLS using the Stata command ivreghdfe.

97

Electronic copy available at: https://ssrn.com/abstract=4273442

 Table D-XIX: IV results with globalization shocks at time t − 2

Volume (ΠVi,e,t ) Mean margin (ΠM

i,e,t )
All RW LW All RW LW
(1) (2) (3) (4) (5) (6)
log Imp
d i,t−2 (LS) 0.84* 1.73** 1.04
(0.50) (0.87) (0.87)
log Imp
d i,t−2 (HS) -1.15** -1.85** -1.99**
(0.51) (0.80) (1.01)
log Mig
d i,t−2 (LS) 0.71* 2.23*** -1.68*
(0.37) (0.62) (1.02)
log Mig
d i,t−2 (HS) -1.34*** -2.31** 0.42
(0.51) (0.99) (1.26)
Impi,t−2 (LS) 9.19* 9.98** 1.78
(4.89) (4.71) (2.64)
Impi,t−2 (HS) -0.70 -1.35 0.85
(1.19) (0.88) (0.76)
Migi,t−2 (LS) -0.50 2.23 -2.12
(6.15) (6.17) (3.21)
Migi,t−2 (HS) 4.06 1.20 5.68
(17.92) (17.90) (9.00)
Observations 564 564 564 564 456 458
Pseudo-R2 0.41 0.38 0.52
R2 0.03 0.01 0.00
K-Paap F-stat 10.86 19.43 12.63
Year & Country FE 3 3 3 3 3 3
Controls 3 3 3 3 3 3
Notes: ***, **, and * denote significance at the 1, 5, and 10 percent levels, respectively; clustered
standard errors at the country level are reported in parentheses; coefficients presented in column
(1) to (3) have been estimated with PPML using the Stata command ppmlhdfe and predicted glob-
alization variables from the model estimated in equation (7), while coefficients in column (4) to (6)
have been estimated with 2SLS using the Stata command ivreghdfe.

98

Electronic copy available at: https://ssrn.com/abstract=4273442

 Table D-XX: IV results with globalization shocks between t − 2 and t

Volume (ΠVi,e,t ) Mean margin (ΠM

i,e,t )
All RW LW All RW LW
(1) (2) (3) (4) (5) (6)
log Imp
d i,t−2→t (LS) 0.75 1.61* 0.82
(0.55) (0.95) (0.90)
log Imp
d i,t−2→t (HS) -1.04 -1.86** -1.04
(0.65) (0.93) (0.87)
log Mig
d i,t−2→t (LS) 0.62 2.24*** -1.68*
(0.42) (0.64) (0.96)
log Mig
d i,t−2→t (HS) -1.30** -2.51** 0.33
(0.54) (1.00) (1.25)
Impi,t−2→t (LS) 3.28** 2.98** 0.79
(1.61) (1.30) (0.95)
Impi,t−2→t (HS) -0.17 -0.40 0.28
(0.37) (0.27) (0.25)
Migi,t−2→t (LS) 0.17 0.77 -0.61
(2.07) (2.06) (1.09)
Migi,t−2→t (HS) 0.39 0.65 2.14
(6.86) (5.95) (3.27)
Observations 564 564 564 564 456 458
Pseudo-R2 0.40 0.37 0.51
R2 0.06 0.08 0.00
K-Paap F-stat 13.26 10.40 8.67
Year & Country FE 3 3 3 3 3 3
Controls 3 3 3 3 3 3
Notes: ***, **, and * denote significance at the 1, 5, and 10 percent levels, respectively; clus-
tered standard errors at the country level are reported in parentheses; coefficients presented in
column (1) to (3) have been estimated with PPML using the Stata command ppmlhdfe and
predicted globalization variables from the model estimated in equation (7), while coefficients in
column (4) to (6) have been estimated with 2SLS using the Stata command ivreghdfe.

99

Electronic copy available at: https://ssrn.com/abstract=4273442

 Table D-XXI: IV results with globalization shocks between two elections

Volume (ΠVi,e,t ) Mean margin (ΠM

i,e,t )
All RW LW All RW LW
(1) (2) (3) (4) (5) (6)
log Imp
d i,t−e→t (LS) 0.87 1.24 1.98**
(0.53) (0.91) (0.86)
log Imp
d i,t−e→t (HS) -0.67 -1.38 -1.53*
(0.58) (0.88) (0.88)
log Mig
d i,t−e→t (LS) 0.36 1.91*** -1.93*
(0.44) (0.64) (1.02)
log Mig
d i,t−e→t (HS) -0.85* -2.12*** 0.98
(0.49) (0.77) (1.36)
Impi,t−e→t (LS) 1.27* 1.34** 0.08
(0.65) (0.63) (0.41)
Impi,t−e→t (HS) -0.14 -0.24 0.08
(0.18) (0.18) (0.10)
Migi,t−e→t (LS) -1.25 -0.62 -0.75
(1.26) (0.98) (0.61)
Migi,t−e→t (HS) -2.12 -0.10 -0.19
(3.39) (2.24) (1.92)
Observations 574 574 574 574 460 468
Pseudo-R2 0.40 0.36 0.52
R2 0.04 0.06 0.01
K-Paap F-stat 6.32 8.84 5.27
Year & Country FE 3 3 3 3 3 3
Controls 3 3 3 3 3 3
Notes: ***, **, and * denote significance at the 1, 5, and 10 percent levels, respectively; clus-
tered standard errors at the country level are reported in parentheses; coefficients presented in
column (1) to (3) have been estimated with PPML using the Stata command ppmlhdfe and
predicted globalization variables from the model estimated in equation (7), while coefficients
in column (4) to (6) have been estimated with 2SLS using the Stata command ivreghdfe.

100

Electronic copy available at: https://ssrn.com/abstract=4273442

D.8.2 Alternative Party Classifications and Populism Score Measures

Table D-XXII: IV results with lax and strict definitions of populist parties

Lax Definition (>0.9 SD) Strict Definition (>1.1 SD)

All RW LW All RW LW
(1) (2) (3) (4) (5) (6)
log Imp
d i,t (LS) 1.00** 1.70** 1.43 0.83 1.49* 0.52
(0.50) (0.82) (0.88) (0.51) (0.78) (1.26)
log Imp
d i,t (HS) -1.30* -2.11** -1.25 -1.19 -1.89** -0.36
(0.68) (0.91) (0.84) (0.75) (0.94) (1.17)
log Mig
d i,t (LS) 0.42 1.89*** -1.61* 0.55 2.20*** -1.51
(0.45) (0.58) (0.87) (0.44) (0.64) (1.03)
log Mig
d i,t (HS) -0.83 -1.97** 0.65 -1.21** -2.66*** 0.52
(0.58) (0.92) (1.17) (0.60) (0.90) (1.36)
Observations 575 575 575 575 575 575
Pseudo-R2 0.40 0.37 0.51 0.39 0.35 0.48
Year & Country FE 3 3 3 3 3 3
Controls 3 3 3 3 3 3
Notes: ***, **, and * denote significance at the 1, 5, and 10 percent levels, respectively; clustered
standard errors at the country level are reported in parentheses; coefficients presented in columns
(1) to (6) have been estimated with PPML using the Stata command ppmlhdfe and predicted
globalization variables from the model estimated in equation (7).

101

Electronic copy available at: https://ssrn.com/abstract=4273442

 Table D-XXIII: IV results using the 3C Populism Score

Volume (ΠVi,e,t ) Mean margin (ΠM

i,e,t )
All RW LW All RW LW
(1) (2) (3) (4) (5) (6)
log Imp
d i,t (LS) 0.35 1.02 1.84
(0.54) (0.65) (1.27)
log Imp
d i,t (HS) -0.99 -1.49* -0.75
(0.70) (0.79) (1.25)
log Mig
d i,t (LS) 0.99** 2.18*** -0.87
(0.39) (0.58) (0.70)
log Mig
d i,t (HS) -1.71*** -2.47*** -0.92
(0.49) (0.84) (0.88)
Impi,t (LS) 8.08** 7.19*** 2.90
(3.92) (2.56) (2.33)
Impi,t (HS) 0.05 -0.86 0.98*
(0.79) (0.71) (0.49)
Migi,t (LS) 4.31 6.08 -1.54
(3.99) (3.72) (3.06)
Migi,t (HS) -9.37 -8.56 7.40
(14.27) (9.76) (9.80)
Observations 575 575 575 578 461 470
Pseudo-R2 0.46 0.38 0.62
R2 0.04 0.07 -0.00
K-Paap F-stat 12.05 11.36 9.45
Year & Country FE 3 3 3 3 3 3
Controls 3 3 3 3 3 3
Notes: ***, **, and * denote significance at the 1, 5, and 10 percent levels, respectively; clustered
standard errors at the country level are reported in parentheses; coefficients presented in column
(1) to (3) have been estimated with PPML using the Stata command ppmlhdfe and predicted glob-
alization variables from the model estimated in equation (7), while coefficients in column (4) to (6)
have been estimated with 2SLS using the Stata command ivreghdfe.

102

Electronic copy available at: https://ssrn.com/abstract=4273442

 Table D-XXIV: IV results using the 15C Populism Score

Volume (ΠVi,e,t ) Mean margin (ΠM

i,e,t )
All RW LW All RW LW
(1) (2) (3) (4) (5) (6)
log Imp
d i,t (LS) 0.86* 1.41** 2.14**
(0.49) (0.60) (1.07)
log Imp
d i,t (HS) -1.04* -2.21*** -1.45
(0.58) (0.74) (1.10)
log Mig
d i,t (LS) 0.59* 1.50*** -0.03
(0.35) (0.52) (0.55)
log Mig
d i,t (HS) -0.72 -1.66** -0.52
(0.48) (0.65) (0.63)
Impi,t (LS) 7.47 4.99 4.44
(5.65) (3.02) (3.04)
Impi,t (HS) 0.05 -0.62 0.66
(1.05) (0.64) (0.46)
Migi,t (LS) 7.78 3.97 1.41
(4.83) (3.67) (3.13)
Migi,t (HS) 13.85 7.72 8.51
(21.27) (12.26) (10.96)
Observations 575 575 575 578 461 470
Pseudo-R2 0.54 0.46 0.59
R2 0.06 0.03 -0.00
K-Paap F-stat 12.05 11.36 9.45
Year & Country FE 3 3 3 3 3 3
Controls 3 3 3 3 3 3
Notes: ***, **, and * denote significance at the 1, 5, and 10 percent levels, respectively; clustered
standard errors at the country level are reported in parentheses; coefficients presented in column
(1) to (3) have been estimated with PPML using the Stata command ppmlhdfe and predicted
globalization variables from the model estimated in equation (7), while coefficients in column (4)
to (6) have been estimated with 2SLS using the Stata command ivreghdfe.

103

Electronic copy available at: https://ssrn.com/abstract=4273442

D.8.3 Alternative Measures of Migration Shocks

Table D-XXV: IV results with skill-selection imputed using data for the year 2000

Volume (ΠVi,e,t ) Mean margin (ΠM

i,e,t )
All RW LW All RW LW
(1) (2) (3) (4) (5) (6)
log Imp
d i,t (LS) 0.90* 1.66* 0.96
(0.52) (0.86) (0.89)
log Imp
d i,t (HS) -1.19* -1.91** -0.76
(0.69) (0.86) (0.82)
log Mig
d i,t (LS) 0.49 1.43** -1.26
(0.51) (0.70) (0.92)
log Mig
d i,t (HS) -0.93 -1.19 -0.01
(0.62) (0.96) (1.17)
Impi,t (LS) 5.02** 4.21** 1.31
(2.32) (1.73) (1.42)
Impi,t (HS) -0.20 -0.62* 0.45
(0.54) (0.37) (0.38)
Migi,t (LS) 3.66 1.94 -0.02
(3.52) (3.20) (1.75)
Migi,t (HS) -7.91 -0.59 -0.02
(7.41) (5.27) (3.03)
Observations 569 569 569 572 461 464
Pseudo-R2 0.40 0.35 0.50
R2 0.07 0.09 0.00
K-Paap F-stat 11.14 32.40 11.39
Year & Country FE 3 3 3 3 3 3
Controls 3 3 3 3 3 3
Notes: ***, **, and * denote significance at the 1, 5, and 10 percent levels, respectively; clus-
tered standard errors at the country level are reported in parentheses; coefficients presented in
column (1) to (3) have been estimated with PPML using the Stata command ppmlhdfe and
predicted globalization variables from the model estimated in equation (7), while coefficients
in column (4) to (6) have been estimated with 2SLS using the Stata command ivreghdfe.

104

Electronic copy available at: https://ssrn.com/abstract=4273442

Table D-XXVI: IV results using interactions with 1960 immigrants’ share in total population

Volume (ΠVi,e,t ) Mean margin (ΠM

i,e,t )
All RW LW All RW LW
(1) (2) (3) (4) (5) (6)
log Imp
d i,t (LS) 0.91* 1.76** 0.80
(0.52) (0.86) (0.77)
log Imp
d i,t (HS) -1.22* -2.07** -0.93
(0.67) (0.93) (0.85)
log Mig
d i,t (LS) 0.23 1.77*** -2.04**
(0.50) (0.62) (0.95)
log Mig
d i,t (HS) -1.03* -1.97** 0.16
(0.55) (0.89) (1.17)
d i,t (LS) × dSH B
log Mig 0.64 0.29 1.50**
1960
(0.52) (1.02) (0.72)
d i,t (LS) × dSH T
log Mig 1.32* 0.97 3.78***
1960
(0.68) (1.00) (1.47)
Impi,t (LS) 4.93** 4.16** 0.95
(2.32) (1.84) (1.46)
Impi,t (HS) -0.17 -0.63 0.49
(0.58) (0.38) (0.39)
Migi,t (LS) -0.53 0.45 -1.13
(3.91) (3.21) (1.89)
Migi,t (HS) 2.18 7.13 2.06
(10.55) (8.40) (5.01)
B
Migi,t (LS) × dSH1960 3.93 4.16 -1.30
(4.45) (4.39) (1.85)
T
Migi,t (LS) × dSH1960 0.26 -2.99 1.41
(4.48) (3.19) (2.73)
Observations 575 575 575 578 461 470
Pseudo-R2 0.41 0.37 0.52
R2 0.07 0.11 0.02
K-Paap F-stat 26.76 12.35 7.12
Year & Country FE 3 3 3 3 3 3
Controls 3 3 3 3 3 3
Notes: ***, **, and * denote significance at the 1, 5, and 10 percent levels, respectively; clus-
tered standard errors at the country level are reported in parentheses; coefficients presented in
column (1) to (3) have been estimated with PPML using the Stata command ppmlhdfe and
predicted globalization variables from the model estimated in equation (7), while coefficients
in column (4) to (6) have been estimated with 2SLS using the Stata command ivreghdfe.
dShareB T
1960 and dShare1960 are dummies equal to one if the country belong the bottom or top
quartile in terms o immigration share in the 1960, respectively.

105

Electronic copy available at: https://ssrn.com/abstract=4273442

D.8.4 Alternative Measures of Import Shocks

Table D-XXVII: IV results with labor-intensive imports

Volume (ΠVi,e,t ) Mean margin (ΠM

i,e,t )
All RW LW All RW LW
(1) (2) (3) (4) (5) (6)
log Imp
d i,t (LS) 1.18* 1.99** -1.10
(0.67) (0.97) (1.09)
log Imp
d i,t (LAB) -0.43 -0.18 3.95**
(0.60) (0.83) (1.54)
log Imp
d i,t (HS) -0.98 -2.01** -2.07
(0.70) (0.98) (1.50)
log Mig
d i,t (LS) 0.55 2.04*** -1.84*
(0.43) (0.58) (0.96)
log Mig
d i,t (HS) -1.07* -2.12** 0.66
(0.57) (0.88) (1.23)
Impi,t (LS) 5.73** 6.25*** 1.56
(2.36) (1.85) (1.43)
Impi,t (LAB) -1.42 -2.57** -0.91
(1.22) (1.25) (0.61)
Impi,t (HS) -0.00 -0.37 0.61
(0.66) (0.45) (0.41)
Migi,t (LS) -0.49 0.00 -1.43
(3.47) (3.17) (1.52)
Migi,t (HS) 2.68 3.29 4.50
(10.04) (7.33) (4.45)
Observations 572 572 572 572 461 464
Pseudo-R2 0.40 0.36 0.52
R2 0.06 0.07 0.02
K-Paap F-stat 8.94 11.01 7.22
Year & Country FE 3 3 3 3 3 3
Controls 3 3 3 3 3 3
Notes: ***, **, and * denote significance at the 1, 5, and 10 percent levels, respectively; clus-
tered standard errors at the country level are reported in parentheses; coefficients presented in
column (1) to (3) have been estimated with PPML using the Stata command ppmlhdfe and
predicted globalization variables from the model estimated in equation (7), while coefficients in
column (4) to (6) have been estimated with 2SLS using the Stata command ivreghdfe.

106

Electronic copy available at: https://ssrn.com/abstract=4273442

 Table D-XXVIII: IV results with imports of medium-skilled intensive goods

Volume (ΠVi,e,t ) Mean margin (ΠM

i,e,t )
All RW LW All RW LW
(1) (2) (3) (4) (5) (6)
log Imp
d i,t (LS) 1.29 1.39 1.95**
(0.81) (1.08) (0.99)
log Imp
d i,t (MS) -0.82 1.04 -2.51***
(1.27) (1.91) (0.87)
log Imp
d i,t (HS) -0.76 -2.57* 0.45
(0.76) (1.33) (0.81)
log Mig
d i,t (LS) 0.54 2.07*** -1.87**
(0.42) (0.59) (0.93)
log Mig
d i,t (HS) -1.00* -2.21** 1.05
(0.53) (0.94) (1.21)
Impi,t (LS) 7.40*** 5.06* 2.79*
(2.51) (2.55) (1.39)
Impi,t (MS) -1.82 -0.68 -1.29**
(1.20) (1.17) (0.58)
Impi,t (HS) 0.29 -0.41 0.78*
(0.67) (0.47) (0.42)
Migi,t (LS) -0.72 0.64 -1.74
(3.27) (3.07) (1.50)
Migi,t (HS) 4.47 4.01 5.89
(10.16) (7.88) (5.05)
Observations 572 572 572 572 461 464
Pseudo-R2 0.40 0.36 0.51
R2 0.07 0.08 0.02
K-Paap F-stat 14.12 15.24 8.04
Year & Country FE 3 3 3 3 3 3
Controls 3 3 3 3 3 3
Notes: ***, **, and * denote significance at the 1, 5, and 10 percent levels, respectively; clustered
standard errors at the country level are reported in parentheses; coefficients presented in column
(1) to (3) have been estimated with PPML using the Stata command ppmlhdfe and predicted
globalization variables from the model estimated in equation (7), while coefficients in column (4)
to (6) have been estimated with 2SLS using the Stata command ivreghdfe.

107

Electronic copy available at: https://ssrn.com/abstract=4273442

D.8.5 Origin-specific Measures of Migration and Imports shocks
Table D-XXIX: IV results with skill-origin specific flows

Volume (ΠVi,e,t ) Mean margin (ΠM

i,e,t )
All RW LW All RW LW
(1) (2) (3) (4) (5) (6)
log Imp
d i,t (LS-LI) 0.86∗∗∗ 1.45∗∗∗ 0.42
(0.15) (0.29) (0.32)
log Imp
d i,t (LS-HI) -0.25 -0.12 0.42
(0.41) (0.86) (1.04)
log Imp
d i,t (HS-LI) -0.07 -0.71 0.71∗∗
(0.23) (0.47) (0.32)
log Imp
d i,t (HS-HI) -1.09∗ -1.22 -1.72∗
(0.66) (0.98) (1.04)
log Mig
d i,t (LS-LI) 0.89∗∗ 1.81∗∗∗ -1.73∗
(0.37) (0.49) (1.00)
log Mig
d i,t (LS-HI) 0.22 -0.20 0.52
(0.43) (0.63) (0.51)
log Mig
d i,t (HS-LI) -1.31∗∗∗ -1.94∗∗∗ 0.85
(0.45) (0.51) (1.07)
log Mig
d i,t (HS-HI) -0.40 -0.15 -1.13∗
(0.51) (0.77) (0.58)
Impi,t (LS-LI) 11.35 7.39 3.19
(7.11) (5.79) (5.90)
Impi,t (LS-HI) 4.31∗∗ 2.71 1.34
(1.84) (1.77) (1.23)
Impi,t (HS-LI) 1.52 -1.12 4.83∗
(3.58) (2.79) (2.68)
Impi,t (HS-HI) -0.45 -0.39 0.14
(0.45) (0.41) (0.27)
Migi,t (LS-LI) -1.38 3.76 -3.56∗∗
(2.64) (2.32) (1.54)
Migi,t (LS-HI) 7.81∗∗ 2.24 3.56∗
(3.50) (4.61) (1.96)
Migi,t (HS-LI) 8.95 0.53 9.86∗∗
(8.26) (6.83) (4.79)
Migi,t (HS-HI) -30.54∗∗ -19.04 -7.09
(14.00) (15.21) (7.39)
Observations 575 575 575 578 461 470
Pseudo-R2 0.45 0.42 0.53
R2 0.04 0.06 0.01
K-Paap F-stat 14.87 8.73 14.75
Year & Country FE 3 3 3 3 3 3
Controls 3 3 3 3 3 3
Notes: ***, **, and * denote significance at the 1, 5, and 10 percent levels, respectively;
clustered standard errors at the country level are reported in parentheses; coefficients pre-
sented in column (1) to (3) have been estimated with PPML using the Stata command
ppmlhdfe and predicted globalization variables from the model estimated in equation (7),
while coefficients in column (4) to (6) have been estimated with 2SLS using the Stata com-
mand ivreghdfe.

108

Electronic copy available at: https://ssrn.com/abstract=4273442

D.8.6 Analysis by Sub-sample

Table D-XXX: IV results using interactions with post-1990 dummy

Volume (ΠVi,e,t ) Mean margin (ΠM

i,e,t )
All RW LW All RW LW
(1) (2) (3) (4) (5) (6)
log Imp
d i,t (LS) 1.04* 2.20** 0.64
(0.53) (0.86) (0.91)
log Imp
d i,t (HS) -1.04 -1.71 -0.70
(0.74) (1.12) (0.96)
log Mig
d i,t (LS) 0.49 1.97*** -1.82*
(0.43) (0.57) (1.00)
d i,t (HS)
log Mig -1.09 -2.30*** 0.76
(0.68) (0.85) (1.26)
d i,t (LS) × dpost1990
log Imp -0.51 -1.49*** 0.31
(0.38) (0.55) (0.32)
d i,t (LS) × dpost1990
log Mig 0.49 1.48* -0.48
(0.44) (0.85) (0.33)
Impi,t (LS) 5.87** 5.25*** 0.39
(2.37) (1.67) (1.56)
Impi,t (HS) 0.01 -0.11 0.22
(0.57) (0.54) (0.36)
Migi,t (LS) 0.01 1.39 -1.18
(2.86) (3.02) (1.71)
Migi,t (HS) -5.13 -2.51 5.53
(10.42) (8.87) (5.11)
Impi,t (LS) × dpost1990 -2.88 -4.30** 2.25
(2.02) (1.84) (1.39)
Migi,t (LS) × dpost1990 4.50** 3.70 -1.10
(2.01) (2.61) (1.33)
Observations 575 575 575 578 461 470
Pseudo-R2 0.41 0.40 0.51
R2 0.06 0.08 0.01
K-Paap F-stat 6.52 7.09 11.84
Year & Country FE 3 3 3 3 3 3
Controls 3 3 3 3 3 3
Notes: ***, **, and * denote significance at the 1, 5, and 10 percent levels, respectively; clus-
tered standard errors at the country level are reported in parentheses; coefficients presented in
column (1) to (3) have been estimated with PPML using the Stata command ppmlhdfe and
predicted globalization variables from the model estimated in equation (7), while coefficients
in column (4) to (6) have been estimated with 2SLS using the Stata command ivreghdfe.

109

Electronic copy available at: https://ssrn.com/abstract=4273442

 Table D-XXXI: IV results using interactions with EU28 dummy

Volume (ΠVi,e,t ) Mean margin (ΠM

i,e,t )
All RW LW All RW LW
(1) (2) (3) (4) (5) (6)
log Imp
d i,t (LS) 0.53 1.52* 0.06
(0.50) (0.84) (0.96)
log Imp
d i,t (HS) -1.12 -2.14** -0.72
(0.69) (0.98) (0.85)
log Mig
d i,t (LS) -0.37 1.07 -2.60***
(0.39) (0.75) (0.99)
log Mig
d i,t (HS) -0.84 -1.76* 0.95
(0.57) (0.91) (1.13)
d i,t (LS) × dEU 28
log Imp 0.99** 0.60 2.29***
(0.43) (0.39) (0.70)
d i,t (LS) × dEU 28
log Mig 1.28*** 1.14* 1.12*
(0.35) (0.65) (0.67)
Impi,t (LS) 5.21** 6.35*** 1.02
(2.59) (1.65) (1.55)
Impi,t (HS) -0.42 -0.62 0.37
(0.63) (0.40) (0.39)
Migi,t (LS) -5.86 -2.78 -2.94**
(3.60) (2.76) (1.43)
Migi,t (HS) 11.34 10.30 7.31
(11.53) (7.88) (5.19)
Impi,t (LS) × dEU 28 -0.49 -2.00** 0.07
(1.37) (0.89) (0.63)
Migi,t (LS) × dEU 28 9.08*** 4.64 2.98*
(3.25) (2.86) (1.58)
Observations 575 575 575 578 461 470
Pseudo-R2 0.43 0.38 0.54
R2 0.04 0.10 -0.01
K-Paap F-stat 3.26 3.20 3.08
Year & Country FE 3 3 3 3 3 3
Controls 3 3 3 3 3 3
Notes: ***, **, and * denote significance at the 1, 5, and 10 percent levels, respectively; clustered
standard errors at the country level are reported in parentheses; coefficients presented in column
(1) to (3) have been estimated with PPML using the Stata command ppmlhdfe and predicted
globalization variables from the model estimated in equation (7), while coefficients in column (4)
to (6) have been estimated with 2SLS using the Stata command ivreghdfe.

110

Electronic copy available at: https://ssrn.com/abstract=4273442

 Table D-XXXII: IV results excluding Latin American Countries

Volume (ΠVi,e,t ) Mean margin (ΠM

i,e,t )
All RW LW All RW LW
(1) (2) (3) (4) (5) (6)
log Imp
d i,t (LS) 0.85∗ 1.78∗∗ 0.30
(0.51) (0.88) (0.97)
log Imp
d i,t (HS) -1.53∗∗ -2.38∗∗∗ -0.39
(0.68) (0.89) (0.96)
log Mig
d i,t (LS) 0.64 2.02∗∗∗ -1.85∗∗
(0.44) (0.56) (0.92)
log Mig
d i,t (HS) -1.36∗∗ -2.24∗∗∗ 0.36
(0.57) (0.86) (1.26)
Impi,t (LS) 5.08∗∗ 3.71∗∗ 1.20
(2.45) (1.83) (1.53)
Impi,t (HS) -0.22 -0.56 0.47
(0.53) (0.38) (0.36)
Migi,t (LS) 0.37 0.10 -0.60
(3.13) (3.09) (1.44)
Migi,t (HS) 1.29 4.38 3.00
(10.27) (8.04) (4.69)
Observations 545 545 545 548 449 445
Pseudo-R2 0.41 0.36 0.52
R2 0.06 0.09 0.01
K-Paap F-stat 11.60 17.78 10.53
Year & Country FE 3 3 3 3 3 3
Controls 3 3 3 3 3 3
Notes: ***, **, and * denote significance at the 1, 5, and 10 percent levels, respec-
tively; clustered standard errors at the country level are reported in parentheses; coeffi-
cients presented in column (1) to (3) have been estimated with PPML using the Stata
command ppmlhdfe and predicted globalization variables from the model estimated in
equation (7), while coefficients in column (4) to (6) have been estimated with 2SLS us-
ing the Stata command ivreghdfe. The sample of countries exclude Argentina, Chile
and Mexico.

111

Electronic copy available at: https://ssrn.com/abstract=4273442

 Table D-XXXIII: IV results over Balanced Sample of Countries

Volume (ΠVi,e,t ) Mean margin (ΠM

i,e,t )
All RW LW All RW LW
(1) (2) (3) (4) (5) (6)
log Imp
d i,t (LS) 1.58∗ 3.56∗∗∗ 1.81∗
(0.86) (1.22) (1.07)
log Imp
d i,t (HS) -2.55∗∗∗ -3.06∗ -4.28∗∗∗
(0.93) (1.62) (1.33)
log Mig
d i,t (LS) 0.32 1.82∗ -2.21∗∗
(0.54) (1.06) (1.09)
log Mig
d i,t (HS) -0.95 -3.29∗∗ 2.70∗∗
(0.72) (1.57) (1.28)
Impi,t (LS) 6.34∗∗ 7.58∗∗∗ 0.80
(2.77) (1.70) (1.53)
Impi,t (HS) 0.06 -0.79∗∗ 0.57
(0.67) (0.36) (0.37)
Migi,t (LS) -2.02 -6.32∗∗ 1.67
(4.23) (2.98) (1.94)
Migi,t (HS) 4.76 16.50∗∗ -3.11
(13.24) (7.91) (7.29)
Observations 363 363 363 363 289 325
Pseudo-R2 0.51 0.51 0.61
R2 0.09 0.11 0.04
K-Paap F-stat 8.75 14.03 7.48
Year & Country FE 3 3 3 3 3 3
Controls 3 3 3 3 3 3
Notes: ***, **, and * denote significance at the 1, 5, and 10 percent levels, respectively;
clustered standard errors at the country level are reported in parentheses; coefficients pre-
sented in column (1) to (3) have been estimated with PPML using the Stata command
ppmlhdfe and predicted globalization variables from the model estimated in equation (7),
while coefficients in column (4) to (6) have been estimated with 2SLS using the Stata
command ivreghdfe. The sample of countries includes countries which have their first
election before 1970.

112

Electronic copy available at: https://ssrn.com/abstract=4273442

D.9 IV Results with Interaction Terms
In Tables D-XXXIV to D-XXXVI, we start from a parsimonious version of Eq. (5) and Table 4 –
including imports of low-skill labor intensive goods ImpLS LS



i,e,t and low-skill immigration Migi,e,t – and
supplement it with interactions between globalization shocks and other potential drivers of populism.
The new specification is given by Eq. (6).
We create four dummies to capture whether (i) the country experienced a year of negative real
income growth in the last two years before the election (a proxy for an economic crisis), (ii) the
country experienced a variation in the share of manufacturing value added in GDP in the last two
years that belongs to the bottom quartile of the distribution (a proxy for de-industrialization), (iii)
the level of diversity in the origin mix of imports and genetic distance of the migration inflows belongs
to the top decile of the distribution (a proxy for the underlying cultural diversity involved in imported
goods or brought by immigrants), and (iv) the share of internet users belongs to the top decile of the
population (a proxy for the prevalence of social media).

113

Electronic copy available at: https://ssrn.com/abstract=4273442

Table D-XXXIV: Reduced-form IV PPML and 2SLS results – Volume and Mean Margins
Interaction with economic crisis (dGi,t )

Volume (ΠVi,e,t ) Mean margin (ΠM

i,e,t )
All RW LW All RW LW
(1) (2) (3) (4) (5) (6)
dGi,t -0.03 -1.79 -1.35 -0.08 -0.03 -0.09*
(0.75) (1.71) (1.55) (0.08) (0.07) (0.05)
log Imp
d i,t (LS) 0.88* 1.91** 0.58
(0.47) (0.84) (0.89)
log Imp
d i,t (HS) -1.35** -2.35** -0.74
(0.68) (0.93) (0.82)
log Mig
d i,t (LS) 0.53 1.91*** -1.60*
(0.43) (0.63) (0.95)
log Mig
d i,t (HS) -1.11* -2.08** 0.61
(0.57) (0.81) (1.23)
d i,t (LS) × dGi,t
log Imp 0.01 -0.40 0.62**
(0.17) (0.38) (0.31)
d i,t (LS) × dGi,t
log Mig 0.10 0.07 -0.77**
(0.14) (0.29) (0.35)
Impi,t (LS) 4.67** 4.08** 0.48
(2.20) (1.73) (1.19)
Impi,t (HS) -0.26 -0.55 0.44
(0.53) (0.38) (0.30)
Migi,t (LS) 1.36 0.60 -0.79
(3.62) (3.09) (1.37)
Migi,t (HS) -1.53 3.46 2.45
(11.06) (8.14) (4.83)
Impi,t (LS) × dGi,t 2.15 -0.91 2.76**
(1.44) (1.07) (1.27)
Migi,t (LS) × dGi,t -2.08 0.08 -1.02
(1.50) (1.32) (0.97)
Observations 575 575 575 578 461 470
Pseudo-R2 0.41 0.38 0.52
R2 0.07 0.09 0.04
K-Paap F-stat 14.90 16.01 8.74
Notes: ***, **, and * denote significance at the 1, 5, and 10 percent levels, respectively; clus-
tered standard errors at the country level are reported in parentheses; coefficients presented
in column (1) to (3) have been estimated with PPML using the Stata command ppmlhdfe and
predicted globalization variables from the model estimated in equation (7), while coefficients
in column (4) to (6) have been estimated with 2SLS using the Stata command ivreghdfe.

114

Electronic copy available at: https://ssrn.com/abstract=4273442

Table D-XXXV: Reduced-form IV PPML and 2SLS results – Volume and Mean Margins
Interaction with de-industrialization (dDi,t )

Volume (ΠVi,e,t ) Mean margin (ΠM

i,e,t )
All RW LW All RW LW
(1) (2) (3) (4) (5) (6)
dDi,t 0.93 1.43 -0.47 0.05 -0.02 0.03
(0.58) (0.92) (1.02) (0.05) (0.05) (0.04)
log Imp
d i,t (LS) 0.76 1.57* 0.82
(0.53) (0.84) (0.99)
log Imp
d i,t (HS) -1.29* -2.31** -0.87
(0.70) (0.93) (0.80)
log Mig
d i,t (LS) 0.45 1.89*** -1.59
(0.44) (0.63) (0.98)
log Mig
d i,t (HS) -0.97* -1.92** 0.67
(0.56) (0.82) (1.32)
d i,t (LS) × dDi,t
log Imp 0.28** 0.51*** 0.19
(0.12) (0.19) (0.26)
d i,t (LS) × dDi,t
log Mig -0.06 -0.17 -0.29
(0.11) (0.20) (0.23)
Impi,t (LS) 5.01* 3.71** 1.86
(2.53) (1.71) (1.57)
Impi,t (HS) -0.20 -0.59 0.46
(0.54) (0.36) (0.37)
Migi,t (LS) 0.72 0.12 -0.71
(2.94) (2.94) (1.35)
Migi,t (HS) 1.04 2.23 4.31
(10.81) (8.36) (4.52)
Impi,t (LS) × dDi,t -0.02 0.67 -0.86*
(0.76) (0.48) (0.51)
Migi,t (LS) × dDi,t -0.40 0.94 -0.42
(1.35) (1.62) (0.48)
Observations 575 575 575 578 461 470
Pseudo-R2 0.41 0.38 0.51
R2 0.07 0.09 0.01
K-Paap F-stat 10.72 9.82 6.05
Notes: ***, **, and * denote significance at the 1, 5, and 10 percent levels, respectively;
clustered standard errors at the country level are reported in parentheses; coefficients pre-
sented in column (1) to (3) have been estimated with PPML using the Stata command
ppmlhdfe and predicted globalization variables from the model estimated in equation (7),
while coefficients in column (4) to (6) have been estimated with 2SLS using the Stata
command ivreghdfe.

115

Electronic copy available at: https://ssrn.com/abstract=4273442

Table D-XXXVI: Reduced-form IV PPML and 2SLS results – Volume and Mean Margins
Interaction with internet coverage (dIi,t )

Volume (ΠVi,e,t ) Mean margin (ΠM

i,e,t )
All RW LW All RW LW
(1) (2) (3) (4) (5) (6)
dIi,t 3.85** 5.01 5.27* -0.23 -0.23* -0.16*
(1.94) (3.37) (2.90) (0.17) (0.12) (0.09)
log Imp
d i,t (LS) 1.17** 2.06** 1.13
(0.48) (0.92) (0.76)
log Imp
d i,t (HS) -1.63*** -2.52*** -0.82
(0.58) (0.84) (0.73)
log Mig
d i,t (LS) 0.61 2.17*** -1.87**
(0.41) (0.55) (0.89)
log Mig
d i,t (HS) -1.02* -2.18** 0.74
(0.57) (0.91) (1.18)
d i,t (LS) × dIi,t
log Imp 1.38** 3.25*** 0.81
(0.66) (0.92) (0.95)
log Mig
d i,t (LS)× dIi,t 0.34 -0.60 1.01
(0.50) (0.59) (1.12)
Impi,t (LS) 4.13* 3.70** 0.44
(2.35) (1.78) (1.37)
Impi,t (HS) -0.20 -0.70* 0.45
(0.57) (0.39) (0.34)
Migi,t (LS) -0.30 1.67 -1.79
(3.97) (3.22) (1.58)
Migi,t (HS) 2.43 0.59 5.55
(10.25) (7.86) (3.91)
Impi,t (LS) × dIi,t 1.28 2.55* 1.49
(2.71) (1.39) (1.09)
Migi,t (LS) × dIi,t 2.07 -0.95 1.33
(3.34) (1.68) (1.40)
Observations 575 575 575 578 461 470
Pseudo-R2 0.42 0.40 0.52
R2 0.07 0.10 0.04
K-Paap F-stat 10.11 8.11 8.30
Notes: ***, **, and * denote significance at the 1, 5, and 10 percent levels, respectively; clus-
tered standard errors at the country level are reported in parentheses; coefficients presented in
column (1) to (3) have been estimated with PPML using the Stata command ppmlhdfe and
predicted globalization variables from the model estimated in equation (7), while coefficients
in column (4) to (6) have been estimated with 2SLS using the Stata command ivreghdfe.

116

Electronic copy available at: https://ssrn.com/abstract=4273442

Table D-XXXVII: Reduced-form IV PPML and 2SLS results – Volume and Mean Margins
Interaction with trade diversity (dHHIit ) and genetic distance (dGDit )

Volume (ΠVi,e,t ) Mean margin (ΠM

i,e,t )
All RW LW All RW LW
(1) (2) (3) (4) (5) (6)
dHHIit -1.05 -3.72∗∗∗ -1.66 0.38∗∗∗ 0.26∗∗∗ 0.15∗∗
(1.03) (1.32) (2.46) (0.10) (0.09) (0.06)
dGDit -2.31∗ -1.02 -4.66∗ 0.17 0.04 0.09
(1.26) (2.41) (2.70) (0.11) (0.10) (0.07)
log Imp
d i,t (LS) 0.95∗ 1.86∗∗ 0.96
(0.53) (0.79) (0.85)
log Imp
d i,t (HS) -1.37∗∗ -2.38∗∗∗ -1.20
(0.65) (0.92) (0.89)
log Mig
d i,t (LS) 1.13∗∗ 2.21∗∗∗ -1.02
(0.46) (0.61) (0.98)
log Mig
d i,t (HS) -1.55∗∗ -2.20∗∗ 0.23
(0.70) (0.93) (1.28)
d i,t (LS) × dHHIit
log Imp -0.29 -1.23∗∗∗ -0.22
(0.25) (0.32) (0.56)
d i,t (LS) × dGDit
log Mig -0.63∗∗∗ -0.54 -1.03∗
(0.23) (0.55) (0.57)
Impi,t (LS) 4.86∗∗∗ 4.29∗∗∗ 0.99
(1.53) (1.50) (1.02)
Impi,t (HS) -0.22 -0.59∗ 0.44∗
(0.39) (0.33) (0.23)
Migi,t (LS) 0.54 -0.07 -0.20
(1.91) (1.91) (1.12)
Migi,t (HS) 1.14 4.99 1.69
(7.14) (6.07) (4.16)
Impi,t (LS) × dHHIit -5.20∗∗∗ -3.60∗∗ -2.36∗
(1.68) (1.65) (1.22)
Migi,t (LS) × dGDit -2.85 5.56 -7.95∗∗
(5.21) (4.37) (3.33)
Observations 575 575 575 578 461 470
Pseudo-R2 0.42 0.40 0.52
R2 0.10 0.11 0.04
K-Paap F-stat 53.37 36.02 45.89
Notes: ***, **, and * denote significance at the 1, 5, and 10 percent levels, respectively; clus-
tered standard errors at the country level are reported in parentheses; coefficients presented in
column (1) to (3) have been estimated with PPML using the Stata command ppmlhdfe and
predicted globalization variables from the model estimated in equation (7), while coefficients in
column (4) to (6) have been estimated with 2SLS using the Stata command ivreghdfe.

117

Electronic copy available at: https://ssrn.com/abstract=4273442

D.10 Exploring Diversity Specific results
Figure D-I and Tables D-XXXVIII to D-XL explore the potential interaction effect of diversity in
low-skill imports and immigration on populism. First we compute for low-skill specific inflows f ∈
{M ig, Imp} a Greenberg Index as follows:

I
f
sfc,i,t × (1 − sfc,i,t ) × gc,i × ec,i ,
X
HHIc,t = (8)
i=1

where sfi,t is the low-skill origin specific inflow from country i over the total low-skill inflow to desti-
nation country c at year t. Such index augments the standard Herfindal index by including measures
of time-invariant bilateral genetic distance (gc,i ) and economic distance (ec,i ) to capture relatedness
across origin and destination countries. Bilateral genetic distances are available from Spolaore and
Wacziarg (2009), while economic distances are measured as the difference in GDP per capita between
destination and origin country in the year 2000. Following Alesina et al. (2016) we then compute
two variations of the Greenberg index, which put different weights to groups. A first variation put
higher weight to origin groups that are genetically close but economically distant to the country of
destination (LH). The second variation put higher weight to origin groups that are genetically distant
but economically close to the country of destination (HL). These two extremes are motivated by the
literature that explores the economic effect of migration diversity, and results in the U.S. context show
that the effects are magnified once only one of the two distances has high weight at the time (Docquier
et al., 2020). Since we do not expect that voters and politicians are able to distinguish detailed differ-
ences across origin countries, we regrouped the set of country of origin in the following broad regions,
following the World Bank Classification: Australia and New Zealand, Caribbean, Central America,
Central Asia, Eastern Africa, Eastern Asia, Eastern Europe, Melanesia, Micronesia, Middle Africa,
Northern Africa, Northern America, Northern Europe, Polynesia, South America, South-eastern Asia,
Southern Africa, Southern Asia, Southern Europe, Western Africa, Western Asia and Western Europe.
Finally, to investigate the potential amplifying effect on our low-skill specific variables, we construct
dummies equal to one if the low-skill specific inflows belong to the first decile of the distribution in
terms of Greenberg index and we interact them with our low-skill inflows.
Figure D-I(a) and Table D-XXXVIII shows the results using the simple Greenber Indexes for trade
and migration. The results show that while diversity in imports reduces the positive effect of low-skill
intensive imports on both margins of populism, we find no statistically significant associated to the
interaction with diversity among immigrants. These results suggest, if any, that higher variety in
imports could hamper the trade-specific determinant of the recent rise of populism.
Figures D-I(b) and (c) and Tables D-XXXIX and D-XL reports the results once genetically and
economically distant groups are weighted differently in the construction of the Greenberg Index. The
results suggest an amplifying effect of diversity (both in trade and migration) on low-skill specific
estimates once higher weight is associated to economically distant groups, particularly on the volume
margin. Conversely, the interactions with low-skill intensive imports are negative and statistically

118

Electronic copy available at: https://ssrn.com/abstract=4273442

 Figure D-I: Interactions with amplifiers for volume and mean margins
Reduced-form IV PPML and 2SLS results - Diversity-specific results

10
4
2

0
Volume Margin

Mean Margin
-2 0

-10
-4
-6

-20
Import LS Migration LS Import LS Migration LS

(a) Greenberg Index

10

10
2
3

1
2

0
5
Volume Margin

Volume Margin
Mean Margin

Mean Margin
0
1

-10
-1
0
0

-2

-20
-1

-5

-3

Import LS Migration LS Import LS Migration LS Import LS Migration LS Import LS Migration LS

(b) Greenberg Index (high weight genetically close (c) Greenberg Index (high weight genetically distant
group and economically distant groups) group and economically close groups)

Notes: Black (square), blue (triangle) and red (diamond) objects correspond to overall, right wing and left
wing dimensions, respectively. Dependent variable is the volume margin on the left panels, while is the
mean margin in the right panels. The estimates represent the coefficients of the interaction term between
migration (LS) and imports (LS) with a dummy equal to one (top-decile) as proxy for trade diversity and
migration diversity. 90% confidence intervals are reported.

119

Electronic copy available at: https://ssrn.com/abstract=4273442

significant once higher weight is associated to genetically distant groups. Concerning migration, the
interactions are barely statistically significant. These results, in line with the ones presented in Table D-
XXIX, seems to suggest that more than the ”cultural threat” driven by low-skill flows, the magnifying
role is played by the poor economic conditions of immigrants, which could be perceived as burden on
the welfare state, and imports from poor countries.
Finally, as additional robustness check (available upon request) we include in the above mentioned
specifications dummies that captures inflows characterized by high economic distance and high genetic
distance, respectively. These measures are constructed as weighted average of the bilateral distances,
using the share of origin-specific low-skill flows as weights. The inclusion of these dummies, to better
capture potential unmeasured cultural or economic distance, does not affect our previously presented
results.

120

Electronic copy available at: https://ssrn.com/abstract=4273442

Table D-XXXVIII: Reduced-form IV PPML and 2SLS results – Volume and Mean Margins
Interaction with greenberg trade diversity (dHHIIit ) and greenberg migration diversity index
(dHHIM
it )

Volume (ΠVi,e,t ) Mean margin (ΠM

i,e,t )
All RW LW All RW LW
(1) (2) (3) (4) (5) (6)
dHHIIit -6.30∗∗ -2.64 -19.28∗∗∗ 0.25∗∗ 0.12 0.07
(2.77) (3.90) (4.21) (0.12) (0.10) (0.09)
dHHIM
it -1.53∗ 4.87 -3.49∗∗ 0.11 -0.04 0.17∗∗
(0.91) (3.68) (1.66) (0.11) (0.09) (0.07)
log Imp
d i,t (LS) 0.98∗ 1.34∗ 2.27∗∗
(0.51) (0.77) (1.11)
log Imp
d i,t (HS) -1.00 -1.60∗ -0.43
(0.69) (0.85) (1.14)
log Mig
d i,t (LS) 0.68 2.01∗∗∗ -2.58∗∗
(0.50) (0.58) (1.02)
log Mig
d i,t (HS) -0.92 -2.02∗∗ 2.10∗
(0.61) (0.85) (1.25)
dHHIIit × log Imp
d i,t (LS) -1.89∗∗∗ -1.12 -4.58∗∗∗
(0.70) (0.97) (1.03)
dHHIM
it × log Migi,t (LS)
d -0.36 1.50 -0.86∗
(0.23) (0.99) (0.48)
Impi,t (LS) 5.41∗∗∗ 4.17∗∗∗ 1.49
(1.57) (1.50) (1.02)
Impi,t (HS) -0.13 -0.51 0.48∗∗
(0.39) (0.33) (0.24)
Migi,t (LS) 0.58 0.93 -0.75
(1.91) (1.89) (1.12)
Migi,t (HS) -0.80 1.60 2.99
(7.14) (6.18) (4.24)
dHHIIit × Impi,t (LS) -10.99∗∗∗ -2.86 -7.41∗∗∗
(3.95) (3.27) (2.28)
dHHIM
it × Migi,t (LS) -3.34 2.46 -4.37∗
(4.00) (3.23) (2.39)
Observations 574 574 574 574 457 469
Pseudo-R2 0.43 0.38 0.56
R2 0.07 0.11 0.03
K-Paap F-stat 49.72 37.93 45.66
Notes: ***, **, and * denote significance at the 1, 5, and 10 percent levels, respectively; clustered
standard errors at the country level are reported in parentheses; coefficients presented in column (1)
to (3) have been estimated with PPML using the Stata command ppmlhdfe and predicted globaliza-
tion variables from the model estimated in equation (7), while coefficients in column (4) to (6) have
been estimated with 2SLS using the Stata command ivreghdfe.

121

Electronic copy available at: https://ssrn.com/abstract=4273442

Table D-XXXIX: Reduced-form IV PPML and 2SLS results – Volume and Mean Margins
Interaction with greenberg trade diversity (dHHII,LH
it ) and greenberg migration diversity
index, higher weights genetically close groups and economically distant groups (dHHIM,LH
it )

Volume (ΠVi,e,t ) Mean margin (ΠM

i,e,t )
All RW LW All RW LW
(1) (2) (3) (4) (5) (6)
dHHII,LH
it 5.60∗∗∗ 6.64∗∗ 3.80 -0.10 -0.06 0.06
(2.11) (2.92) (3.94) (0.11) (0.08) (0.10)
dHHIM,LH
it 3.03∗∗∗ 3.40∗∗ 4.29∗∗ -0.01 0.09 -0.01
(0.98) (1.37) (1.75) (0.09) (0.07) (0.06)
log Imp
d i,t (LS) 0.77 1.65∗ 1.05
(0.49) (0.87) (0.89)
log Imp
d i,t (HS) -1.10∗ -1.91∗∗ -0.75
(0.64) (0.92) (0.72)
log Mig
d i,t (LS) 0.53 2.10∗∗∗ -1.81∗
(0.44) (0.56) (0.95)
log Mig
d i,t (HS) -1.18∗∗ -2.36∗∗∗ 0.62
(0.52) (0.72) (1.20)
dHHII,LH
it × log Imp
d i,t (LS) 1.66∗∗∗ 1.96∗∗ 1.08
(0.61) (0.83) (1.04)
dHHIM,LH
it × log Mig
d i,t (LS) 0.60∗∗∗ 0.56∗ 1.01∗∗
(0.22) (0.31) (0.41)
Impi,t (LS) 5.72∗∗∗ 4.30∗∗∗ 1.68∗
(1.51) (1.45) (0.99)
Impi,t (HS) -0.15 -0.55∗ 0.45∗
(0.39) (0.33) (0.24)
Migi,t (LS) 0.96 0.89 -0.62
(1.89) (1.89) (1.12)
Migi,t (HS) -2.67 2.29 2.34
(7.17) (6.15) (4.29)
dHHII,LH
it × Impi,t (LS) 5.88∗∗ 3.50∗ -0.06
(2.38) (1.85) (2.01)
dHHIM,LH
it × Migi,t (LS) 2.54 -2.36∗ 1.39
(1.82) (1.42) (1.19)
Observations 574 574 574 574 457 469
Pseudo-R2 0.42 0.40 0.51
R2 0.09 0.10 0.00
K-Paap F-stat 54.08 40.08 45.95
Notes: ***, **, and * denote significance at the 1, 5, and 10 percent levels, respectively; clustered
standard errors at the country level are reported in parentheses; coefficients presented in column
(1) to (3) have been estimated with PPML using the Stata command ppmlhdfe and predicted
globalization variables from the model estimated in equation (7), while coefficients in column (4)
to (6) have been estimated with 2SLS using the Stata command ivreghdfe.

122

Electronic copy available at: https://ssrn.com/abstract=4273442

 Table D-XL: Reduced-form IV PPML and 2SLS results – Volume and Mean Margins
Interaction with greenberg trade diversity (dHHII,HL
it ) and greenberg migration diversity
index, higher weights genetically distant group and economically close groups (dHHIM,HL
it )

Volume (ΠVi,e,t ) Mean margin (ΠM

i,e,t )
All RW LW All RW LW
(1) (2) (3) (4) (5) (6)
dHHII,HL
it -3.67∗∗∗ -2.79∗ -5.73∗∗∗ 0.78∗∗∗ 0.34∗∗∗ 0.58∗∗∗
(0.71) (1.57) (1.87) (0.14) (0.12) (0.09)
dHHIM,HL
it 1.92 0.52 7.40∗∗ -0.09 -0.08 0.01
(1.97) (3.32) (2.89) (0.11) (0.09) (0.08)
log Imp
d i,t (LS) 1.13∗ 1.91∗∗ 1.66∗
(0.58) (0.88) (0.99)
log Imp
d i,t (HS) -0.68 -1.88∗∗ -0.11
(0.65) (0.81) (1.15)
log Mig
d i,t (LS) 0.62∗ 1.91∗∗∗ -2.43∗∗
(0.37) (0.52) (1.05)
log Mig
d i,t (HS) -0.81∗ -1.89∗∗ 2.11∗
(0.48) (0.85) (1.22)
dHHII,HL
it × log Imp
d i,t (LS) -1.16∗∗∗ -0.91∗∗ -1.86∗∗∗
(0.14) (0.37) (0.41)
dHHIM,HL
it × log Mig
d i,t (LS) 0.41 0.28 1.21∗
(0.43) (0.79) (0.70)
Impi,t (LS) 4.02∗∗∗ 3.75∗∗∗ 0.49
(1.46) (1.45) (0.93)
Impi,t (HS) -0.22 -0.60∗ 0.36
(0.38) (0.32) (0.23)
Migi,t (LS) 1.38 1.20 -0.75
(1.84) (1.86) (1.06)
Migi,t (HS) -1.27 4.56 4.39
(7.10) (6.37) (4.12)
dHHII,HL
it × Impi,t (LS) -15.01∗∗∗ -3.30 -12.24∗∗∗
(4.52) (4.03) (2.77)
dHHIM,HL
it × Migi,t (LS) 0.48 -1.32 -1.57
(3.39) (3.11) (2.17)
Observations 574 574 574 574 457 469
Pseudo-R2 0.45 0.38 0.60
R2 0.14 0.13 0.11
K-Paap F-stat 44.26 24.72 43.56
Notes: ***, **, and * denote significance at the 1, 5, and 10 percent levels, respectively; clustered stan-
dard errors at the country level are reported in parentheses; coefficients presented in column (1) to
(3) have been estimated with PPML using the Stata command ppmlhdfe and predicted globalization
variables from the model estimated in equation (7), while coefficients in column (4) to (6) have been
estimated with 2SLS using the Stata command ivreghdfe.

123

Electronic copy available at: https://ssrn.com/abstract=4273442