Political Homophily in Social Relationships:

Evidence from Online Dating Behavior

Gregory A. Huber, Yale University
Neil Malhotra, Stanford University

Do people form relationships based upon political similarity? Past work has shown that social relationships are more
politically similar than expected by chance, but the reason for this concordance is unclear. Is it because people prefer
politically similar others, or is it attributable to confounding factors such as convergence, social structures, and sorting
on nonpolitical characteristics? Addressing this question is challenging because we typically do not observe partners
prior to relationship formation. Consequently, we leverage the domain of online dating. We first conducted a na-
tionwide experiment in which we randomized political characteristics in dating profiles. Second, we analyzed behav-
ioral data from a national online dating community. We find that people evaluate potential dating partners more
favorably and are more likely to reach out to them when they have similar political characteristics. The magnitude of
the effect is comparable to that of educational homophily and half as large as racial homophily.

S cholars argue that partisan loyalties extend beyond

issue positions and disagreements over policy, bleed-
ing into social interactions. For instance, survey evi-
dence indicates that individuals do not want to be friends
with, or have their children marry, members of the opposing
ners in the domain of online dating. In two different studies,
we observe political preferences and beliefs before people
evaluate and choose from a known set of potential partners.
These data therefore allow us to estimate how shared political
characteristics predict with whom a person would like to form
party (Iyengar et al. 2012). It is unclear, however, whether a relationship, a direct measure of political homophily. Both
these survey responses predict people’s behaviors outside of studies allow us to preclude post-choice convergence or restric-
such a research setting. At the same time, there is long- tions on available partners as explanations for observed polit-
standing evidence that social relationships, including mar- ical similarity. Each study provides different leverage to assess
riages, are more politically homogenous than one would the importance of sorting on other nonpolitical characteristics.
predict by chance (e.g., Alford et al. 2011; Martin et al. 1986). Our first study is an experiment embedded in a general
But such research does not definitively tell us whether this population survey in which we randomly manipulate the
pattern reflects political choice homophily—a preference for political characteristics of online dating profiles presented to
those who are politically similar—or is instead a side effect of participants. We test whether participants’ assessments of
attitude convergence, constrained partner markets, or part- the profiles are predicted by the concordance of their own
ner choice on the basis of other factors that are correlated views with those shown in the profile. We find that par-
with shared political orientations. Disentangling choice homo- ticipants consistently evaluate profiles more positively (e.g.,
phily from these alternative mechanisms is important because had greater interest in dating the individual presented) when
it suggests that ameliorating partisan divisions may be difficult the target’s profile shared their political ideology. Shared
if self-segregation is intentional rather than incidental. levels of political interest also improve evaluations for some
We present novel data that allow us to directly measure outcomes, but the effects are much weaker. Because we in-
revealed preferences for politically similar relationship part- dependently manipulate the political and nonpolitical char-

Gregory A. Huber (gregory.huber@yale.edu) is professor of political science at Yale University, New Haven, CT 06520. Neil Malhotra (neilm@stanford.edu)
is professor of political economy at the Stanford Graduate School of Business, Stanford, CA 94305.
This research was approved by the IRBs at Stanford University (IRB-24611 and IRB-19678) and Yale University (HSC no. 1007007097). Data and supporting
materials necessary to reproduce the numerical results in the article are available in the JOP Dataverse (https://dataverse.harvard.edu/dataverse/jop). An online
appendix with supplementary material is available at http://dx.doi.org/10.1086/687533.

The Journal of Politics, volume 79, number 1. Published online October 13, 2016. http://dx.doi.org/10.1086/687533
q 2016 by the Southern Political Science Association. All rights reserved. 0022-3816/2017/7901-0019$10.00 269
270 / Political Homophily in Social Relationships Gregory A. Huber and Neil Malhotra

acteristics of the profiles, these experimental results isolate system and those who do not. Thus, as with increasing
the causal effects of shared political predispositions, pro- homophily along class and educational divisions, political
viding direct evidence of choice-based homophily. homophily may exacerbate differences in the distribution of
Our second study uses a large, novel data set from a political resources that are associated with the ability to af-
diverse, national online dating community to understand fect public policy. If engaged people are more likely to as-
which factors predict when individuals communicate with sociate with one another, then it could help explain how
other potential dating partners. This behavioral measure of such social inequality, when it affects policy outcomes, can
social discernment is important because it provides evi- sustain political inequality (Verba et al. 1995).
dence not just of stated preference for political similarity
but also evidence that individuals act on those preferences THE DIMENSIONS OF POLITICAL HOMOPHILY
in real social interactions when they are not being moni- How and why does politics affect the social relationships
tored. We examine the effects of three types of political that individuals form? Theoretically, a preference for po-
characteristics—political identity, issue positions, and po- litically similar partners may reflect a general tendency to
litical engagement. After accounting for the range of avail- prefer similarity for any given personal characteristic (Mc-
able online partners on the site, we find that men are more Pherson et al. 2001). Political homophily, by this view, may
likely to message a woman if they share these key political simply be a particular manifestation of a preference for sim-
traits with her, and women are similarly more likely to ilar others. Of course, political attitudes and orientations are
respond to a man’s message if they share these traits with often deeply held, and so a particular preference for politi-
him. For example, online pairings in which men send a cally similar relationship partners may arise because choosing
message and women reply are about 8%–10% more similar a dissimilar partner may invite future relationship conflict
on ideology and partisanship and about 11% more similar (Gerber et al. 2012) or predict differences about other core
in levels of political interest than all potential pairings, ef- values (Graham et al. 2009) that may be implicated in child-
fects that are similar in size to educational homophily and rearing or other salient choices. We therefore distinguish
about half that of racial homophily. We conduct additional theoretically between three different types of political choice
analysis of these data to show that this pattern does not homophily—identity homophily, issue homophily, and en-
appear to arise simply due to sorting on nonpolitical char- gagement homophily—that may generate different patterns
acteristics. of social sorting.
Overall, our work shows that individuals seek politically First, individuals may sort on political identities, such as
similar relationship partners and that this political sorting identification with a political party (e.g., Democrat) or with
takes place even at the earliest stages of relationship for- an ideological disposition (e.g., conservative). These iden-
mation and in an environment in which individuals can tities appear to form early in life and persist throughout
choose from among many different relationship partners time, leading many scholars to argue that they are more
using diverse criteria. We therefore provide behavioral evi- akin to group identities than simple summaries of political
dence that establishes the external validity of survey data opinions (Campbell et al. 1960). Thus, in light of social and
showing a preference for politically similar social partners. group identity theories, we expect political homophily to
Additionally, compared to previous research, we more per- take place along these lines, just as it does for social iden-
suasively show that political homophily is a source of po- tities for which sorting is ubiquitous (e.g., ethnicity).
litical homogeneity in romantic relationships by providing Second, people may sort according to political issue po-
evidence that excludes alternative explanations for this ob- sitions, such as their stances on economic, social, and foreign
served similarity. policy issues. While these attitudes can also be stable over
The political sorting in romantic relationships that we time, they are distinct from party and ideological identifi-
document likely reduces political disagreement within the cation in several ways. For instance, policy issues evolve,
household, which risks creating political enclaves and may moving in and out of the national agenda, which requires
in turn increase polarization and decrease political toler- citizens to develop new beliefs (Carmines and Stimson 1989).
ance (Mutz 2002). We also show that people select rela- Sometimes social groups cleave along policy lines (e.g., pro-
tionship partners on the basis of shared levels of political life vs. pro-choice activists), but unlike party identification,
engagement, rather than solely based on ideological pre- most issues do not define social groups. Additionally, indi-
dispositions, thereby raising the possibility that the country viduals’ policy views often appear malleable and subservient
may become increasingly stratified between those who have to political identities (Levendusky 2009). Thus, despite the
the resources and motivation to engage with the political fact that issues may signal other group and value commit-
 Volume 79 Number 1 January 2017 / 271

ments, particularly on social issues, we expect political ho- The most extensive body of research on the role of politics
mophily for issue positions to be weaker than it is for political in partner choice is the literature on assortative mating. This
identities. work also draws on studies of the intergenerational trans-
Finally, people may sort based on political engagement. mission of political views (e.g., Alford, Funk, and Hibbing
Independent of whether they agree with another person 2005; Jennings, Stoker, and Bowers 2009) and, from social
about politics, they may prefer someone who shares their psychology, work showing that individuals generally rate as
(lack of ) engagement with political debates and issues. Those more desirable those individuals who are similar to them
who are civically engaged may view those who are not as along a range of dimensions (e.g., Byrne 1961; see also
failing the duties of citizenship (Theiss-Morse and Hibbing Berscheid and Reis [1998] for a review on the correlates of
2005), which may be akin to violating a core value. This may perceived attractiveness). Numerous studies show that mar-
in turn cause those who are not engaged with politics to shun ried couples are more alike on many dimensions than one
the engaged so as to avoid their social disapprobation. Such would expect by chance (e.g., Martin et al. 1986). The ques-
a pattern may be exacerbated if those who dislike politics view tion, however, is if this political congruence arises due to
it as conflictual, partisan, corrupt, and uncivil (Hibbing and choice or is instead induced (McPherson and Smith-Lovin
Theiss-Morse 2002). 1987). That is, do people really seek out politically similar
partners, or does this similarity arise for different reasons?
Choice homophily refers to a preference to associate with
PRIOR RESEARCH a similar other along a given dimension. Induced homo-
We study online dating behavior to test these hypotheses phily, by contrast, arises as an indirect consequence of other
about how shared political orientations affect partner pref- factors. Three main causes of induced homophily are
erences. In addition to providing novel behavioral data for (1) convergence, or the tendency of social partners to be-
understanding how people seek relationship partners, online come more similar after meeting, either through persuasion
dating has become an increasingly important means by or sharing the same environment; (2) social structures (e.g.,
which Americans search for romantic partners (displacing residential segregation) constraining the types of people to
traditional institutions such as school and family) and is a whom one is exposed (i.e., might date); and (3) choice ho-
precursor to marriage (Rosenfeld and Thomas 2012). None- mophily along dimensions other than politics such as racial
theless, it is important to note that with online dating data we or religious identity that are correlated with shared political
observe only the initial periods of relationship formation, views.
which are likely akin to the search for nonromantic social The most compelling evidence to date about the role
partners but may not reflect the entire process by which in- of political choice homophily in marriages comes from
dividuals choose marital partners. Additionally, the popula- Alford et al. (2011).1 Drawing on a survey from a sample of
tion of any online dating community is not fully representa- twins and their relatives, Alford et al. show that married
tive of all Americans on the relationship market. As such, partners are similar for a variety of political measures. To
extrapolating our results to other dating sites or populations explore whether convergence explains this pattern, they
requires additional assumptions, which we discuss later. document that this similarity is at most only modestly larger
Our research builds on work that examines how political for couples who have been married for longer periods. They
homophily shapes which social relationships form. The most argue that this demonstrates that if convergence arises it
developed literature is about assortative mating, but we also must occur early in a marriage, although such a pattern
draw from work that examines behavior in online dating could also be due to a cohort effect; that is, if partners in
forums and in speed dating events. Our literature review more recent marriages are more similar at the beginning of
reveals that while prior research establishes clear evidence their relationship than partners who were married in earlier
that married partners are often quite similar politically, we periods (see also Jennings and Stoker [2001], who find in-
know less about why this sorting arises. As we discuss below, creasing partner correlations for partisanship over time).2
prior work offers conflicting evidence about the importance
of political characteristics in partner choice. Nor is it clear 1. A larger set of studies about political homophily in romantic and
which political characteristics are most important in shaping dating relationships is summarized below. We focus on Alford et al. here
because it is closest to our own work and makes the greatest progress in
social sorting. These questions are particularly ripe because
isolating choice homophily.
existing analysis cannot distinguish individuals choosing
2. Alford et al. also draw from a panel data set in Australia where they
political similar partners from other sources of observed observe one member of a future marriage prior to marriage and show that
political similarity. people who become married experience only modestly larger changes in
272 / Political Homophily in Social Relationships Gregory A. Huber and Neil Malhotra

To understand whether this concordance is explained by other factors that might explain partner choice and political
shared backgrounds (searching on other correlated factors views, and (e) includes any experimental manipulation. As
or constrained partner markets), Alford et al. show that the table shows, no prior study of the role of politics in part-
spousal political similarity is also found within salient social ner choice does more than three of these things; our study
groups (defined separately by shared religion, education, does all five. This is important because studies that do not
church attendance, or party voting). This analysis rules out provide a rich set of political measures (a) cannot be used
choice homophily on other dimensions as causing the to assess the importance of different dimensions of politics
similarity in formed relationships only if those partitionings on partner choice. Examining partner preferences only after
fully account for partner choice and all sources of shared relationships have formed (b) means that researchers need
political views. Similarly, it precludes limits on available to make additional assumptions to distinguish convergence
partners only if those subgroups account for all restrictions from homophily. Similarly, it is also difficult to rule out the
on partners that are correlated with politics. effect of constrained partner markets without accounting for
Alford et al. recognize these limitations of their analysis the diversity of those potential partners (c), and distinguish-
and argue that measuring attitudes before individuals in- ing political homophily from sorting on other dimensions is
teract would more definitively rule out the possibility of very difficult without accounting for a broad range of partner
convergence. Thus, they state that “the ideal research design” characteristics that could also explain partner choice (d) or
for studying assortative mating based on attitudes is a lon- an explicit experimental manipulation (e). Thus, at least four
gitudinal sample of individuals “before they met; however, design elements (a, b, c, and either d or e) are crucial for more
this would require the ability to foresee the future” (2011, persuasively identifying political choice homophily as a source
370). As we explain below, this article implements the re- of political similarity in formed relationships, something even
search design that Alford et al. foresee as the next innova- the most promising work on assortative mating in marriage
tion in the analysis of political homophily. Similarly, to un- (e.g., Alford et al. 2011) does not do.
derstand the effect of social structure on observed homophily, In addition to research on assortative mating, there are
one would need to account for the views of the range of also developing and related literatures focusing on speed
potential future partners (e.g., as in Snyder’s [1964] study of dating and online dating. These approaches provide re-
members of sophomore classes from 13 rural Pennsylvania searchers with possible tools for understanding how differ-
high schools, many of whom ended up marrying one another ent factors affect partner choice, including accounting for
later in life, although that study does not contain any measures the range of available partners and measuring (potential)
of political attitudes). Finally, to rule out political homophily partner characteristics prior to partner interactions. In an
as arising due to choosing on other dimensions correlated early study by Carlson (1979), 96 undergraduates each rated
with shared political views would require either experimen- one opposite sex “computer dating profile” composed of
tal manipulation of political characteristics or obtaining a 20 issue items and two measures of political interest. Each
rich set of nonpolitical characteristics to assess whether, af- profile was either entirely matched on political issues items
ter accounting for choice along those dimensions, political and entirely unmatched on nonpolitical issue items or en-
similarity still predicts partner choice. Our study has these tirely unmatched on political issues and entirely matched
features. on nonpolitical issues. Those profiles that were politically
In light of these arguments, table S1 in the appendix aligned were evaluated more positively, but the profiles did
(appendix, tables S1–S8, figs. S1–S4 available online) sum- not include any other content (e.g., demographics) that
marizes prior work examining the role of politics in assor- might affect partner choice. There are two articles that ex-
tative mating (marriage), speed dating, and online dating in amine the effect of shared political orientations on partner
the United States. For each of the 14 previous studies, we preference using small speed dating events.3 Tidwell, East-
tabulate whether it (a) accounts for multiple dimensions of wick, and Finkel (2013) find that shared ideology (their only
political views, (b) measures both partners’ preferences be- measure of political preference) does not significantly in-
fore a relationship is formed, (c) accounts for potential part- crease romantically liking of a speed dating partner, but
ners and their views, (d) includes measures of a wide variety of their analysis is bivariate and does not control for all other

their views than those whose partnership status does not change. Those 3. One concern with using the speed dating environment to under-
data do not include measures of the views of both members of the future stand relationship formation is that people may use different factors to
couple. choose partners in more common dating environments.
 Volume 79 Number 1 January 2017 / 273

partner characteristics. Similarly, Luo and Zhang (2009) amine communication conditional on browsing (viewing)
find that shared ideology does not explain partner choice in an online profile (i.e., they examine communication after ac-
a speed dating forum, but their analysis is also bivariate in counting for initial searching/screening). They find that both
nature. Unfortunately, given the prior implementation of men and women who are liberal (conservative) are less likely
dating and speed dating experiments (the three studies listed to contact someone who is conservative (liberal). While their
B1–B3 in table S1), we know relatively little about how a study has three of the five design elements included in our
broad range of political factors affect partner choice and study (see table S1, row C3), they have a geographically
whether it is politics per se, or correlated factors, that affect limited sample and cannot account for the role of politics in
evaluations of partners. initial partner search. Additionally, they do not consider a
Turning to analysis using online dating behavior (the broader range of measures of political characteristics, include
three studies listed C1–C3 in table S1), two articles by Klof- only a limited set of covariates that are likely correlated with
stad et al. (2012, 2013) examine how people present them- political views (e.g., the many reasons people are seeking
selves politically in online dating forums.4 They do not ex- dating partners or more detailed measures of religious iden-
amine actual partner communication or other measures of tity), and do not have access to the rich data on personal views
dating behavior but instead focus on a sample of public that might be correlated with political characteristics that we
profiles collected from a national dating site. These profiles present below.
include a single measure of political preference (ideology).
Klofstad et al. (2012) find that people claim to be moderate at
OVERVIEW OF STUDIES
a rate that exceeds that found in other surveys, perhaps be-
We conducted two studies: (1) an experiment in which we
cause expressing a moderate preference may maximize one’s
randomly manipulated the political characteristics of online
appeal to a broad range of partners. Those who do express a
dating profiles presented to participants and (2) an analysis
political preference are more civically engaged, but they
of communication behavior in a large, national online
conclude that “individuals do not appear to initially select
dating community. The two studies have nonoverlapping
potential dates along political lines” (100). They also find that
strengths and weaknesses for identifying choice-based po-
few other characteristics are correlated with expressing a
litical homophily. The first study maximizes internal va-
preference for one ideological extreme over the other. Build-
lidity via controlled manipulation of the information shown
ing on these findings, Klofstad et al. (2013) conclude that
to participants. Thus, political characteristics are uncorre-
there is little evidence for political homophily in choosing
lated with other profile features, making it easier to generate
relationship partners and that political similarity in
unbiased estimates of the effect of political factors on part-
marriages likely indicates either that dating is different than
ner preference. However, these subjects are making choices
searching for a spouse or that political concordance in
in a survey environment, so one might be concerned that the
marriages arises due to sorting on dimensions correlated
same behavior might not manifest outside of the experi-
with political views.
mental context, which is a problem of external validity. The
In terms of observed partner communication behavior
second study maximizes external validity because it exam-
in online dating forums, while there are a number of arti-
ines people’s actual choices made when they are not being
cles focusing on topics such as preferred partner race (e.g.,
monitored in a research setting and also allows us to explore
Robnett and Feliciano 2011), we are aware of only a single
more of the potential dimensions of political homophily.
article that assesses the role of shared political orientations.
However, because we do not fully control the choice envi-
Hitcsch, Hortacsu, and Ariely (2010) examine the behavior
ronment, we need to make additional assumptions to in-
of 6,485 participants in an online dating forum who lived in
terpret our effects as causal, which is a problem of internal
Boston or San Diego. In a multivariate model that includes
validity. Because both studies yield similar results, we reduce
a single measure of political preferences (ideology) they ex-
concerns that these results are a methodological artifact of
either approach.

4. A related but distinct literature considers abstract partner preferences

outside of the online dating setting (but not actual dating or partner prefer- STUDY 1: A NATIONAL EXPERIMENT
ence behavior). See, e.g., Doosje, Rojahn, and Fischer (1999). Kofoed (2008) Procedures
conducts a convenience survey of 168 undergraduates and finds that 55% of
We recruited approximately 1,000 subjects aged 18–35 and
participants had dated someone who did not share their political beliefs and
that 18% of respondents indicated they would not date someone with dif- interested in dating members of the opposite sex from a
ferent political beliefs. nationwide sample provided by Survey Sampling Interna-
274 / Political Homophily in Social Relationships Gregory A. Huber and Neil Malhotra

tional (SSI) to participate in a study about online dating.5 attractiveness, and (6) their interest in being friends with
Subjects were compensated by SSI for their participation in the person.7 These questions tap multiple dimensions of
the experiment, which did not involve deception and took evaluations, from simple measures of physical attractiveness
place over the Internet. The study took about 15 minutes to interest in dating and beliefs about long-term compati-
to complete. After we obtained informed consent, subjects bility and shared values. Full question wordings and re-
took a short survey about their personal characteristics and sponse options are presented in the appendix.
prior online dating behavior. (The complete text of the
survey is presented in the appendix, and tabulations of Method of analysis
subject demographics appear in table S3.) Next, they were Our primary theoretical interest is in the effect of the ma-
asked to evaluate 10 randomly constructed opposite-sex dat- nipulation of the political description in the profile on eval-
ing profiles. We describe how those profiles were con- uations of the person portrayed in the profile. Politics was
structed and the evaluations gathered in greater detail be- randomly assigned with equal probability to: none (i.e., the
low. Finally, subjects were asked a series of questions about field for political affiliation was left blank), “not interested in
the factors they considered important in selecting dating politics,” “conservative,” “moderate,” and “liberal.” This manip-
partners and why they thought the study was being con- ulation allows us to assess political sorting along two di-
ducted.6 Our final analysis data set consists of 979 individ- mensions. First, we can ascertain whether or not partic-
uals who evaluated 10 profiles each (a total of 9,790 obser- ipants evaluate more favorably those individuals who share
vations) and excludes respondents who did not provide their political ideology than those who do not. Second, we
their year of birth or failed to answer a question about their can assess whether or not participants react more favor-
political orientations. ably to individuals who have similar levels of political
The profiles were presented in a standard template sim- interest.
ilar to those found on many popular online dating sites. We Because this is an experiment and both politics and other
used elements of the most popular online dating websites in features of the profiles were randomly assigned, our analysis
our design, enhancing the realism of the experimental stim- of these data is straightforward. In particular, for each out-
uli. A sample profile appears in appendix figure S1, and the come measure Y we estimate participant i’s evaluation of
complete list of information included in the profile is pre- profile j using this equation:
sented in table S2, where we also show which characteris-
tics were randomly manipulated. These independent random Y ij p ai 1 b1
Match ideologyij
manipulations included a person’s picture, textual descrip- 1 b2
Match not interested in politicsij
tion, user ID, age, height, religion, educational attainment, 1 b3
No politics in prof ilej
1 b4
Not interested in politics in prof ilej
and political orientation. We can therefore assess whether ð1Þ
1 b5
Liberal in prof ilej
shared politics matters in explaining partner preference
1 b6
Conservative in prof ilej
when other nonpolitical dimensions are available on which 1 Z
Other characteristics
to choose. The photographs, user IDs, and profile text were 1 g
Picturej 1 h
Prof ile text j 1 εij :
harvested from public dating profiles. Textual descriptions
were edited for length and to remove inconsistencies with Of primary interest is the effect of shared political ide-
the other manipulations. ology and shared levels of political interest. “Match Ideol-
Participants were shown a profile and asked to evaluate it ogy” is coded “1” if the participant’s self-reported ideology
by answering six questions in closed response format about (liberal, conservative, or moderate) matches the ideology
(1) their interest in contacting the person, (2) their interest shown in the profile, and “0” otherwise. “Match not in-
in responding to a message from the person, (3) whether terested in politics” is coded “1” if the respondent reports
or not they could get along with the person in the long term, “hardly at all” or “only now and then” for how often he or
(4) their assessment of the person’s values, (5) the person’s she follows politics and the profile states “Not interested in
politics.” Our theoretical prediction is that b1 and b2 will be
greater than 0.
5. This experiment is a replication of a smaller study conducted in a
university laboratory using a student population. Results in that context
were similar.
6. Twelve subjects use the word “politics” or its variants in describing 7. When a respondent did not provide a response for a given item, the
why we conducted the study. Excluding these respondents does not alter missing value was replaced with that person’s average response for that
on our results (see the appendix for details). item across all of the profiles they were shown.
 Volume 79 Number 1 January 2017 / 275

In addition to these political variables, we also account come in study 2. We find that shared ideology increases as-
for the nonpolitical characteristics included in the profiles. sessments of interest in initiating contact by about .014 units
We include indicators for each individual photograph and (pp.05, two-tailed tests throughout) on a 5-point scale that
the textual profile descriptions, as well as for each of the age ranges from 0 to 1 and a shared lack of interest in politics
and height options. For age we include a measure of ab- improves evaluations by .019 units (pp.13). Within respon-
solute difference in age between the respondent and the dent, by comparison, the average standard deviation of inter-
profile. For education and religion, where we do not have est in dating is about .23 units (see the bottom row of table 1),
clear ex ante predictions about evaluations of different char- so substantively these manipulations affect average evalu-
acteristics, we include each possible pairing of respondent and ations by about 6% and 8% of the within-person standard
profile characteristics (e.g., for education one variable indi- deviation, respectively. In short, shared ideology affects how
cates that the respondent has a graduate degree and the person much participants stated they would like to contact the people
in the profile has a high school degree, while for religion one in the profiles we presented to them, and the effect of shared
variable represents that the respondent is Jewish and the per- interest is marginally insignificant.
son in the profile is Catholic). We estimate equation (1) using Results in column (2), where the outcome measure is
OLS regression with fixed effects for the individual respondent stated willingness to respond to a message, are similarly
(ai) and cluster standard errors at the respondent level to ac- strong for shared ideology but are smaller and far from
count for correlated responses by each respondent. We pre- statistically significant for shared lack of interest in politics.
sent abridged results (estimated using Stata 12.1 for Windows) In columns (3) and (4) we present two measures of deeper
in table 1, only displaying the coefficients of substantive in- evaluations of compatibility: interest in a long-term rela-
terest. The full regression models can be found in table S4. All tionship and assessments of shared values, respectively. For
dependent variables are scaled to range from 0 to 1, with 1 the both outcomes, the effect of shared ideology is larger than for
most positive outcome and 0 the most negative one. the column (1) contact outcome. Shared ideology increases
interest in long-term dating by 11% (p ! .01) of the within-
Results respondent standard deviation of that measure and 13% (p !
In column (1), we estimate the effect of the experimental .01) for the assessment of shared values. The effect of share
manipulations on interest in initiating contact, a key out- lacked of interest is close to zero for long-term dating and

Table 1. Effect of Shared Political Orientations on Profile Evaluations, Study 1

Interest in Would You Interest in Do They Share Would You Like

Dating Respond to Person? LT Dating Your Values? Attractiveness to Be Friends?
(0–1) (0–1) (0–1) (0–1) (0–1) (0–1)
(1) (2) (3) (4) (5) (6)

Ideology matches .014 .014 .025 .024 .003 .029

[.007]* [.007]* [.007]** [.007]** [.007] [.007]**
Match not interested in
politics (explicit not
interested in profile) .019 .008 2.009 .015 .015 .009
[.012] [.013] [.012] [.012] [.012] [.013]
Observations 9790 9790 9790 9790 9790 9790
No. of respondents
(fixed effects) 979 979 979 979 979 979
R-squared .141 .126 .100 .081 .195 .099
Avg. SD of DV within
respondent .229 .228 .222 .192 .231 .222

Note. OLS coefficients with standard errors clustered at respondent level in brackets. See table S4 for complete model results.
* Significant at 10%.
** Significant at 5%.
276 / Political Homophily in Social Relationships Gregory A. Huber and Neil Malhotra

modestly positive but insignificant for shared values (7.8% of After reporting this public information, the user then
the within-person standard deviation, p p .23). has the option of answering other questions about his or
We asked two additional questions about participants’ her personality, interests, and opinions about relationships.
evaluations of the people shown in the profiles. The first was These items are referred to as “match questions,” distinct
about whether or not the person was attractive (column 5). from the profile questions that are part of setting up an
Neither measure of shared orientation has statistically sig- account. These questions, which include both items sub-
nificant effects on that outcome. The second general ques- mitted by users and the operators of the site, are eclectic.
tion was about whether the profile displayed someone the par- Users also report how important each question they answer
ticipant would like to be friends with (column 6), a measure of is to them in partner choice. Answers to these questions are
nonromantic social sorting. Shared ideology increases desire not made public unless the user specifically requests that
for friendship by 13% relative to the average standard devia- this occur, which we are told is rare. As we describe in
tion by respondent (p ! .01), but shared lack of interest has greater detail below, our new political questions were in-
small and insignificant effects. This pattern is most similar to cluded in this list of nonprofile questions (a complete list of
assessments of shared values. profile and available match questions appears in the ap-
We explore the robustness of this finding in additional pendix).
analyses reported in the appendix (see table S5). These ro- Once a user has completed a profile, he or she can search
bustness checks include (1) employing listwise deletion for for others based on the profile characteristics mentioned
missing values; (2) excluding moderate profiles, which may above. Users can also filter based on whether their potential
be perceived as more attractive than nonaligned profiles; match has a photo as well as recent account activity. The
(3) comparing the effect of shared ideology between men site also presents suggested matches to respondents based
and women; (4) using a more flexible coding of ideological on what they state they are looking for in their profile.
matching; and (5) considering whether the effect of shared Finally, the site creates a “match score” based on the match
ideology varies by political interest. These robustness checks questions answered by the user and all potential partners,
do not provide any evidence that lead us to call into question taking into account the user’s stated importance of each of the
our original reported findings. Overall, the results of this questions and the other users’ responses (including questions
experiment provide strong causal evidence that shared ide- for which the participant chooses not to make their answer
ology affects social evaluations in the online dating context. public).8 Unlike fixed algorithms for suggesting potential
Shared interest may have an effect for some outcome mea- partners used by most other online dating sites, the site’s al-
sures, but the effects are less definitive. People appear to gorithm allows users to assign greater weight to those factors
prefer ideologically similar others when evaluating potential they believe are important in partner choice, including poli-
dating partners. tics. For this reason, any role of shared politics in shaping the
match score is a function of a user’s own explicit preferences
for shared political views. In other words, the fact that users
STUDY 2: AN OBSERVATIONAL STUDY may use political characteristics to shape their partner search
OF BEHAVIORAL DATA is the choice homophily we are attempting to identify.
The online dating site Our analysis here focuses on the acts of sending a message
The data for study 2 come from a major, national online and responding to it, which is a means by which individuals
dating site, which we refer to as “the site” (due to our agree- can converse with one another and, if desired, arrange in-
ment with the site, we cannot reveal its name). Users interact teractions outside of the online dating environment (e.g.,
with the site using an Internet web browser and can also private email communications or dates). A user can send a
choose to be notified about activities on the site by email. message to someone selected from a list of potential partner
Details of how the site operated at the time of the study appear profile(s), and lists are generated using the different ways
in the appendix and are summarized here. When a user first that users can search for potential dating partners outlined
joins the site, he or she constructs a public profile, which can above. As such, initial messaging reflects both sorting and an
include a photograph and a free-form textual description. affirmative decision to contact some subset of those partners
As part of their profile, users can list a series of attributes identified through the initial search process.
that they are looking for in their potential dating partners:
sexual orientation, relationship status, age, location, and what
the other user is looking for in a relationship (e.g., casual 8. The site would not provide us with these weights or the proprietary
dating). algorithm for calculating the match score.
 Volume 79 Number 1 January 2017 / 277

Sample and data set overview unique men and 119,569 women in which women respond to
We obtained a list of all users who were registered on the 24.09% of the messages sent to them.
site between October 1, 2010, and December 15, 2010. We Finally, our third outcome of focus is the joint occur-
began by removing from this list those users whose site rence of a man messaging a woman and a woman replying.
membership was revoked for inappropriate behavior, who Insofar as bilateral communication is required to develop
withdrew from the site before the end of the sample period, a social relationship, this analysis allows us to assess the
or who resided outside the 50 US States or the District of cumulative effect of a man’s initial messaging and a wom-
Columbia. Additionally, for reasons of analytic tractability, an’s replying behavior on the formation of potentially en-
we restricted our analysis to heterosexual men and women during relationships. This analysis uses the same data set
age 18 to 64 who reported being single and seeking opposite as in the analysis of men’s initial messaging behavior. The
sex partners.9 Finally, to eliminate inactive users and low- third measure differs from the second in that our analysis is
quality profiles, we removed all men who were never re- not restricted to cases where men initiate communication,
plied to and all women who never replied to a message. but it still accounts for the fact that bilateral communica-
These restrictions yield 142,964 men and 119,754 women. tion is required to develop a further relationship. In 0.136%
The final data set for our analysis is a set of dyads, in of all dyads both parties send messages.
which for a given geographic area, we form all possible pair-
wise comparisons between each man and each woman.10 Geo- Independent variables
graphic areas are identified by two-digit zip codes (e.g., 60, The site, in cooperation with us, fielded seven new political
which includes all zip codes from 60000 to 60999), which questions. We describe our cooperation with the site in
is the lowest level of geographic reporting provided to us greater detail in the appendix. These seven questions tap
by the site and which is chosen so that users could readily the three overarching types of political characteristics in-
meet in person if they wanted to. This yields a data set of troduced above: (1) political identity, (2) issue positions,
367,047,169 dyads, which we use for our analysis. and (3) political engagement. The question wordings are
presented in the appendix. In addition to partisanship and
Outcome measurement liberal-conservative ideology (two traditionally studied iden-
We examine three outcomes in our analysis. The first out- tities in research on political behavior) we also asked about
come is a man sending a message to a woman in a shared users’ media choices, which might indicate political orienta-
geographic area. While both men and women send mes- tion in a less-explicit manner. We also asked users about their
sages, men initiate 88% of online conversations, and so we positions on specific issues, one economic and one social/
begin by exploring men’s initial efforts to reach out to women. cultural. Finally, we asked two questions designed to tap users’
In the complete data set, men send messages to 0.565% of political engagement—how important politics is to them
their potential dyadic partners. Selected demographics for personally and the level of participation they expect of other
this sample appear in appendix table S5. citizens.
The second outcome we analyze is each woman’s mes- Our goal is to analyze how well the responses to these
saging behavior in response to a man having sent her a political questions predict sorting on the site. As with any
message. Here we begin by taking all cases where a man sent other match question, users had the option to decline an-
a woman a message in the same geographic area and retain swering each of these new questions. We sought to use
cases where there is variation in the woman’s responding standard question wording from existing surveys (e.g., the
behavior and whether each man’s messages yield responses. American National Election Study and the General Social
This yields a data set of 2,073,902 dyads composed of 142,622 Survey) where possible, but negotiations with the site and
technical concerns affected the final questions that were
fielded.
In addition to these seven new questions, we also used
9. We remove the few users younger than 18 to comply with human
subjects requirements. The number of users over 64 is very small. We
site participants’ responses to a range of other questions
exclude those already in a relationship because some of these users may included in the nonanonymous profile to account for user
not be seeking relationship partners. Our focus on heterosexual partner preferences and characteristics. The public profile variables
search is driven by the relatively small number of users seeking same-sex include age, height, education, race, what one is looking for
partners in certain geographic areas.
10. For example, if there were two men, X1 and X2, and two women,
in a dating partner, tobacco/alcohol/drug use, whether one
Y1 and Y2, in a geographic area, we would create four dyads: X1Y1, X1Y2, has/wants kids, length of short textual description, and re-
X2Y1, and X2Y2. ligion. As with study 1, we adopt a flexible modeling spec-
278 / Political Homophily in Social Relationships Gregory A. Huber and Neil Malhotra

ification by creating indicator variables for each pairwise reply, a proportional increase of about 4%. In other words,
combination of responses to each item included in the model women are reinforcing men’s initial sorting based on height .11
(e.g., for religion, one indicator variable is “man Christian, Finally, the rightmost set of numbers is the same data set as
woman Catholic”). Note that many of these public questions used for the analysis of men sending messages, but here the
may be correlated with politics (e.g., religion). outcome is joint messaging behavior. We find that the cu-
mulative rate of shared height is even larger: relative to all
Results dyads shared height increases by 2.3 points (9.8%) in cases
We examine the level of political and nonpolitical sorting in where both parties send a message.
online dating markets. To do so, we compare dyads in which As this table makes clear, matching on height is not an
parties communicate to the entire set of dyads in each geo- anomalous case. Focusing on the remaining nonpolitical
graphic area (i.e., compared to random selection among items, age differences (i.e., users not in the same 5-year bin)
geographically close online partners, the available set of are reduced in joint messaging cases by over a half, shared
online partners). We focus on three outcomes: men’s initial race increases by 16.6%, shared religion by 50.0%, and
messaging behavior, women’s replying behavior (condi- shared education by 10.6%.
tional on having been sent a message), and the joint prob- By comparison, the political items vary considerably in
ability a man sends a woman a message and she replies. We their discernment. At the low end, shared beliefs about the
do so separately for five nonpolitical characteristics on which role of religion in shaping policy (0.6% proportional de-
individuals may seek to match (height, age, race, religion, crease for both parties messaging, which is the only case in
and education) and the three categories of political questions which dyads where messaging takes place are less similar
mentioned above. The nonpolitical characteristics provide a than those in which it does not), the duty to vote (1.3 points,
baseline with which to calibrate the effects of political pre- 2.1%), and media preferences (1.6 points, 3.8%) are modestly
dispositions. associated with joint messaging. By contrast, common ide-
In table 2, we compare similarity in the population of ology (3.9 points, 7.7%), partisanship (4.0 points, 9.5%),
eligible dyads to those in which our outcome of interest levels of political interest (3.0 points, 10.7%), and ideas about
occurs. So, for example, the first row of table 2 begins by how to balance the budget (5.6 points, 10.8%) are substan-
displaying the proportion of all dyads in which men and tially more frequent in dyads where both parties commu-
women are in the same height quintile. By this calculation, nicate. The small effects of social issues in determining
23.4% of all dyads match on this measure of height. This sorting may be due to the fact that our survey question did
baseline matching rate is attributable to the underlying not deal with specific “hot button” issues such as same-sex
distribution of the characteristic (in this case height) in the marriage. These larger effects are on par with the effects of
geographically segmented partner market. It is how much shared educational attainment and about half the size of the
homophily we would observe if individuals simply ran- effect of shared race and concord with our theoretical
domly chose to contact someone in their geographic area. expectations that political identities and engagement should
Thus, because we compare observed similarity to what be especially strong predictors of political homophily. We
we would expect given restricted partner markets, we can note that the effect of shared interest is larger and more
rule out the possibility that homophily arises simply due to robust here than in study 1.
a similar set of potential partners. Among those cases in Several patterns in these results are important to high-
which a man sends a message, however, height is matched light. First, of the 21 differences in proportions for the po-
24.7% of the time, a proportional increase in height ho- litical items shown in table 2, 19 are positive and statistically
mogeneity of about 1.3 percentage points or 5.5% [(24.72 significant, indicating increased homogeneity relative to
23.4)/23.4]. (Not surprisingly, given the large number of all dyads. Second, as with the nonpolitical items, women ap-
observations in our data set, most differences between dyads pear to reinforce the initial increase in political homogeneity
that message and those that do not, for each messaging
outcome, are statistically significant at p ! .001. For this 11. Whether or not one expects women to reinforce men’s initial
reason, we instead highlight cases where results are not sta- sorting depends on one’s hypotheses about men’s strategic behavior. For
tistically significant at conventional levels.) The second group these reasons, it is useful to think of three potential patterns in this analysis:
of numbers is for the universe of dyads included in the (1) women reinforcing the initial sorting created by men; (2) women un-
doing men’s sorting; or (3) null effects in which women are not contributing
woman’s replying analysis (i.e., cases where a man first mes- to sorting above and beyond what men have already done, either because
saged a woman). Here, 24.7% of all dyads match on height, a men anticipate women’s preferences or because women do not consider
number that increases to 25.7% among cases where women these factors at all.
Table 2. Observed Homogeneity for Selected Characteristics (Study 2) for All Potential Dyads and Those in Which Communication Occurs

Men’s Sending Behavior, at Least One Message Women’s Replying Behavior, at Least One Message Joint Communication Behavior, at Least One Message Each
Proportion Matching Proportion Matching Proportion Matching

If Man Among All

Sends Percentage p-Value of Proportional Dyads Where If Percentage p-Value of Proportional If Man Sends Percentage p-Value of Proportional
Among First Point Difference Increase in Man Sent First Woman Point Difference Increase in Among Message and Point Difference Increase in
Characteristic All Dyads Message Difference in Means Match Rate Message Replies Difference in Means Match Rate All Dyads Woman Replies Difference in Means Match Rate

Match height
quintile 23.4% 24.7% 1.3% !.001 5.5% 24.7% 25.7% 1.0% !.001 4.0% 23.4% 25.7% 2.3% !.001 9.8%
Match age (5 year
bins, beginning
with 18–22) 20.6% 38.4% 17.9% !.001 87.0% 38.4% 41.5% 3.1% !.001 8.0% 20.6% 41.5% 20.9% !.001 101.9%
Match race 54.7% 59.5% 4.7% !.001 8.7% 59.5% 63.8% 4.3% !.001 7.3% 54.7% 63.8% 9.1% !.001 16.6%
Match religion 21.4% 29.6% 8.2% !.001 38.6% 29.6% 32.0% 2.4% !.001 8.2% 21.4% 32.0% 10.7% !.001 50.0%
Match education 43.2% 46.6% 3.4% !.001 7.9% 46.6% 47.7% 1.2% !.001 2.5% 43.2% 47.7% 4.6% !.001 10.6%
Match ideology 50.6% 52.6% 1.9% !.001 3.8% 52.6% 54.5% 2.0% !.001 3.7% 50.6% 54.5% 3.9% !.001 7.7%
Match partisanship 42.1% 44.1% 2.1% !.001 4.9% 44.1% 46.1% 2.0% !.001 4.5% 42.1% 46.1% 4.0% !.001 9.5%
Match media
preferences 42.4% 43.4% 1.0% !.001 2.5% 43.4% 44.0% .6% .055 1.4% 42.4% 44.0% 1.6% !.001 3.8%
Match role
of church 89.7% 88.4% 21.3% !.001 21.4% 88.4% 89.2% .8% !.001 .9% 89.7% 89.2% 2.5% !.001 2.6%
Match how balance
budget 51.5% 55.5% 3.9% !.001 7.7% 55.5% 57.1% 1.6% !.001 2.9% 51.5% 57.1% 5.6% !.001 10.8%
Match political
interest 28.2% 30.1% 1.9% !.001 6.9% 30.1% 31.2% 1.1% !.001 3.6% 28.2% 31.2% 3.0% !.001 10.7%
Match duty to vote 60.3% 61.4% 1.0% !.001 1.7% 61.4% 61.6% .2% .088 .4% 60.3% 61.6% 1.3% !.001 2.1%
Dyads 367,047,169 2,073,902 367,047,169
Men 142,964 142,622 142,964
Women 119,754 119,569 119,754
Outcome rate .565% 24.091% .136%

Note. Cell entries in each block (e.g., Men’s Sending Behavior) are first the proportion of dyads that match for a given row variable (e.g., both heights in the same quintile; see appendix for complete coding), second
the proportion matching for the subset of dyads in which the outcome of interest occurs (e.g., the man sends the woman a message), third the percentage point increase in matching from the first to the second
(col. 22col. 1), fourth the p-value of the hypothesis test that the two proportions are equal, and fifth the proportional increase in matching from the first to the second [(col. 22col. 1)/col. 1]. All statistics exclude
individuals who did not provide a response (there are no missing cases for age or height).
280 / Political Homophily in Social Relationships Gregory A. Huber and Neil Malhotra

generated by men’s messaging behavior (women also appear testing whether shared political orientation predicts com-
to select on shared views about the role of religion in shaping munication behavior even after accounting for other factors
policy, whereas men do not). This provides direct evidence that users might use to select partners. We note that the set of
that women’s choices are important in shaping the observed conditioning variables we had access to was extensive com-
levels of homogeneity. (Additionally, it shows that political pared to prior research.
sorting likely does not arise solely due to the site’s match- Due to space constraints, we present the full details of
ing algorithm because women are deciding which messages model specifications and complete regression results in the
to reply to only among the set of men who have contacted appendix and summarize our findings for the occurrence of
them.) Third, there are large baseline differences in match joint messaging (the clearest evidence of bilateral social in-
rates across items, which reflect the range of available online terest) here. The OLS model specification is similar to what
partners. At one extreme, 89.7% of geographically constructed we use for our analysis of study 1. For computational reasons
dyads match on beliefs about the role of religion in setting the data are a subset of the observations used in the table 2
policy, but only 28.2% match on levels of political interest. analysis. Briefly, to account for baseline differences across
One potential concern with this analysis is that com- individuals in their desirability as partners, choosiness in
munication behavior in the online dating environment may communication, and care in crafting messages, we measure
not reflect more serious decisions about forming enduring the rates at which individuals send and receive messages and
social relationships. We take two different approaches to reply to or have their messages replied to. To account for
address this concern (complete results appear in appendix nonpolitical sorting we measure concordance for each dyad
table S7). First, we replicate our results among users more on the all of the traits from the publicly available profile
likely to be seeking serious relationships: those seeking long- information discussed earlier (e.g., age, race, education, re-
term dating relationships and also wanting to have children. ligion).
Among these more “serious” subsamples, results are similar to Figure 1 summarizes results of our analyses for models
those reported above. Second, we replicate our earlier analysis with the political items entered as blocks (e.g., all the indi-
but require more extensive communication to take place cators for potential partner combinations of ideology, com-
(five messages being sent) before coding an outcome as plete regression results are in appendix table S8). We present
having occurred. Despite the fact that these data are sparse differences in predicted probabilities of messaging for cases
(in only 0.026% of cases do both parties send at least five in which individuals match on each trait versus when they
messages), for the seven political items, we find 16 pro- are out of alignment (e.g., man conservative, woman con-
portional increases in match rates (14 statistically signifi- servative vs. man conservative, woman liberal). The figure
cant at p ! .05). reveals that even after accounting for the range of available
Overall, these data provide clear evidence that when po- partners, average user behavior, and sorting on the nonpo-
litical views are measured before relationships are formed and litical profile characteristics, dyads in which both parties
after accounting for baseline sorting expected given available communicate are more similar. For example, the baseline
online partners, there is increased political homogeneity in the rate of communication in this data set is about 0.4%. When
subset of online dyads in which men message, women re- both the man and woman were conservative—rather than
spond to an initial message, and both parties send a message. the woman being liberal—this rate is predicted to increase
Confirming our earlier experimental finding, these data show by 0.22 points, an increase of more than 50% relative to
that after accounting for available online partners and pref- the baseline. This effect is similar in size to the rate at which
erences as measured prior to relationship formation, indi- white men communicate with white rather than black women
viduals outside of the laboratory setting choose relationship (.18 points), and about 4/5 the size of the rate at which
partners in a way that increases political homogeneity. Christian men communicate with Christian rather than athe-
ist/agnostic women (.26 points) (see fig. S2). (As a reminder, all
Multivariate analysis of these estimates are from models that account for the avail-
Although these observational data do not allow us the same able online partners in a geographic area.)
control as the experimental setting, we can nonetheless un- Sorting is similar but less precisely estimated for shared
dertake additional analysis to assess whether the observed party orientation and is not statistically significant for shared
homogeneity in online dating communication reveals po- media preferences. For the political issues, shared views about
litical homophily—a choice to select similar others—or is church state separation predict messaging, but the effects of
instead caused by sorting on other characteristics (one form shared spending preferences are not consistently statistically
of induced homophily as described above). We do so by significant. Finally, shared views about the importance of
 Volume 79 Number 1 January 2017 / 281

Figure 1. Effect of political characteristics on joint messaging behavior estimates from models with individual sets of political items. Mean of DV in this sample
is .0039. See appendix table S8 for model specification and full regression results.

politics (.26 points for agreement that politics is very impor- with measures of concordance on 40 additional match ques-
tant and .17 points for shared lack of interest) have effects tions (see the appendix for a complete list). These 40 ques-
about as large as of shared ideology and are statistically sig- tions were provided to us by the site because they were
nificant (the effects of shared views about the duty to vote are answered by many users and were predictive of messaging.
smaller, with an average estimate of about .06 points, but are Some of these additional match questions include salient
still statistically significant). Overall, the general pattern that political referents (e.g., views about abortion), so finding
emerges is that sorting is strongest based on ideology, parti- evidence of political homophily after accounting for those
sanship, and political interest; it is weaker based on media views is a difficult test. Nonetheless, as appendix figure S3
choice, social and fiscal policy positions, and attitudes about shows, we continue to find that ideological similarity and
whether voting is a civic duty. This is in accord with our shared views about the importance of politics explain which
theoretical expectation that political identities will be the relationships form, reducing the possibility that our earlier
most powerful predictors of political homophily, while sorting findings are due to sorting on other characteristics. Overall,
along policy positions will be less pronounced. In the appendix the most striking finding from study 2 is that these behav-
we use these models to assess the relative importance of homo- ioral data yield patterns very similar to those from study 1
phily in explaining observed sorting in social relationships. about the role of shared ideology; individuals seek politically
Figure S4 shows that models accounting for political homo- similar dating partners. Unlike study 1, we also find evidence
phily predict increased political sorting relative both (1) to that shared views about the importance of politics more con-
random communication in this constrained partner market sistently explain partner communication.
and (2) a baseline nonpolitical sorting model (i.e., a model
that accounts only for the control variables in the regression). DISCUSSION AND CONCLUSION
We also consider the robustness of the figure 1 results to The analysis presented here provides strong evidence of
including all of the political items simultaneously along purposive political sorting in the formation of social rela-
282 / Political Homophily in Social Relationships Gregory A. Huber and Neil Malhotra

tionships. We find that individuals seek out relationship Do political predispositions influence how people select
partners who share their political identities and degree of their social relationships? With two novel data sets we have
engagement with politics, and that this preference for po- addressed this important question. People do seem to con-
litical homophily can be distinguished from three forms struct their social lives around politics, and such sorting
of induced homophily: post-choice convergence, restricted appears substantively consequential in explaining which rela-
partner markets, and sorting on nonpolitical factors. No tionships form. We also find that political homophily is more
prior work can simultaneously rule out these three alterna- than the result of restricted partner markets or selecting on
tive explanations for the observed correlation of political other demographic and social characteristics, explanations
orientations among already-formed social relationships. Our previous work cannot easily and definitively reject. Of con-
experimental study revealed that people react more posi- tinuing importance, this provides direct evidence of social
tively to ideologically congruent profiles and perhaps also sorting along political lines and may also drive future polar-
to those that exhibit similar levels of political interest. The ization through the increased homogenization of political be-
observational study of behavioral data demonstrated that liefs within households and social networks.
the dyads in which men message women, women respond,
and both men and women communicate, are all more po-
litically similar than would be expected by chance. Political ACKNOWLEDGMENTS
choice homophily is substantively large, rivaling the sizes of We thank Sigal Alon, Kevin Arceneaux, Dan Biggers, John-
racial and educational homophily. Paul Ferguson, James Fowler, Sanford Gordon, Sharique Hasan,
These findings have important implications for major Peter Hatemi, John Hibbing, Seth Hill, Shigeo Hirano, Shanto
research areas in political science. In particular, it appears Iyengar, Keith Krehbiel, Casey Klofstad, Annette Lareau, Rose
as if in the contemporary period political orientations di- McDermott, Nolan McCarty, Lincoln Quillian, Sarah Soule,
rectly affect the social relationships people seek to form, and Laura Stoker for their assistance. We also are in debt
which results in increased political homogeneity in formed to the founder and employees of the online dating company
relationships. This has the potential to amplify polarization who generously provided the data. There were several other
through the creation of homogenous social networks and people, including seminar and conference participants, who
households. Further, political homophily is not simply con- provided additional feedback whom we do not have space to
fined to political identities and issue positions but also extends thank here. Finally, we thank anonymous reviewers and the
to engagement with politics itself. editors for their valuable feedback.
Future research should examine how these findings gen-
eralize to social interactions outside of the online dating con-
text. On the one hand, because more information about REFERENCES
potential social partners is available during in-person in- Alford, John R., Carolyn L. Funk, and John R. Hibbing. 2005. “Are Po-
teractions, our findings may represent lower bounds on polit- litical Orientations Genetically Transmitted?” American Political Sci-
ence Review 99:153–67.
ical homophily in other contexts. On the other hand, people
Alford, John R., Peter K. Hatemi, John R. Hibbing, Nicholas G. Martin,
may eschew politics in other forums, making it difficult to sort and Lindon J. Eaves. 2011. “The Politics of Mate Choice.” Journal of
on that basis. Our work, however, identifies the key problems Politics 73:362–79.
of research design that would be necessary to isolate political Berscheid, Ellen, and Harry T. Reis. 1998. “Attraction and Close Rela-
choice homophily in other contexts. Further, for those who tionships.” In D. T. Gilbert, S. T. Fiske, and G. Lindzey, eds., The Hand-
book of Social Psychology, 4th ed., vol. 2. Boston: McGraw-Hill, 193–281.
wish to understand the effect of information transmission in Byrne, Donn. 1961. “Interpersonal Attraction and Attitude Similarity.”
social networks, we provide evidence of purposive selection Journal of Abnormal and Social Psychology 62:713–15.
in the formation of these social relationships. This means Campbell, Angus, Phillip E. Converse, Warren E. Miller, and Donald E.
subsequent analysis of information effects must account for Stokes. 1960. The American Voter. New York: Wiley.
Carlson, James M. 1979. “Politics and Interpersonal Attraction.” American
this systematic variation in which types of political views dif-
Politics Quarterly 7:120–26.
ferent individuals are exposed to. Additionally, it suggests an Carmines, Edward G., and James A. Stimson. 1989. Issue Evolution.
important question for ongoing work is about what factors Princeton, NJ: Princeton University Press.
cause individuals to look beyond political differences when Doosje, Bertjan, Krystyna Rojahn, and Agneta Fischer. 1999. “Partner
Preferences as a Function of Gender, Age, Political Orientation and
forming relationships. Finally, this research project yielded
Level of Education.” Sex Roles 40:45–60.
valuable lessons for scholars who seek to work with firms that Gerber, Alan S., Gregory A. Huber, David Doherty, and Conor M. Dowling.
generate valuable online data (for a discussion, see appendix 2012. “Disagreement and the Avoidance of Political Discussion.”
section 5). American Journal of Political Science 56:849–74.
 Volume 79 Number 1 January 2017 / 283

Graham, Jesse, Jonathan Haidt, and Brian A. Nosek. 2009. “Liberals and Luo, Shanhong, and Guangjian Zhang. 2009. “What Leads to Romantic
Conservatives Use Different Sets of Moral Foundations.” Journal of Attraction?” Journal of Personality 77:933–63.
Personality and Social Psychology 96:1029–46. Martin, N. G., L. J. Eaves, A. C. Heath, R. Jardine, L. M. Feingold, and H. J.
Hibbing, John R., and Elisabeth Theiss-Morse. 2002. Stealth Democracy. Eysenck. 1986. “Transmission of Social Attitudes.” Proceedings of the
Cambridge: Cambridge University Press. National Academy of Sciences 83:4364–68.
Hitsch, Gunter J., Ali Hortacsu, and Dan Ariely. 2010. “What Makes You McPherson, Miller, and Lynn Smith-Lovin. 1987. “Homophily in Vol-
Click?” Quantitative Marketing and Economics 8:393–427. untary Organizations.” American Sociological Review 52:370–79.
Iyengar, Shanto, Gaurav Sood, and Yphtach Lelkes. 2012. “Affect, Not McPherson, Miller, Lynn Smith-Lovin, and James M. Cook. 2001. “Birds
Ideology.” Public Opinion Quarterly 76:405–31. of a Feather.” Annual Review of Sociology 27:415–44.
Jennings, M. Kent, and Laura Stoker. 2001. “Political Similarity and In- Mutz, Diana C. 2002. “Cross-Cutting Social Networks.” American Political
fluence between Husbands and Wives.” Institute for Governmental Science Review 96:111–26.
Studies Paper 2001-14, University of California, Berkeley. Robnett, Belinda, and Cynthia Feliciano. 2011. “Patterns of Racial-Ethnic
Jennings, M. Kent, Laura Stoker, and Jake Bowers. 2009. “Politics across Exclusion by Internet Daters.” Social Forces 89:807–28.
Generations.” Journal of Politics 71:782–99. Rosenfeld, Michael J., and Reuben J. Thomas. 2012. “Searching for a
Klofstad, Casey A., Rose McDermott, and Peter K. Hatemi. 2012. “Do Mate.” American Sociological Review 77:523–47.
Bedroom Eyes Wear Political Glasses?” Evolution and Human Be- Snyder, Eloise C. 1964. “A Study of Homogamy and Marital Selectivity.”
havior 33:100–108. Journal of Marriage and Family 26:332–36.
Klofstad, Casey A., Rose McDermott, and Peter K. Hatemi. 2013. “The Dating Theiss-Morse, Elizabeth, and John R. Hibbing. 2005. “Citizenship and
Preferences of Liberals and Conservatives.” Political Behavior 35:519–38. Civic Engagement.” Annual Review of Political Science 8:227–49.
Kofoed, E. 2008. “The Role of Political Affiliations and Attraction in Tidwell, Natasha D., Paul W. Eastwick, and Eli J. Finkel. 2013. “Perceived,
Romantic Relationships.” Advances in Communication Theory and Re- Not Actual, Similarity Predicts Initial Attraction in a Live Romantic
search 2:1–23. Context.” Personal Relationships 20:199–215.
Levendusky, Matthew. 2009. The Partisan Sort. Chicago: University of Verba, Sidney, Kay Lehman Schlozman, and Henry Brady. 1995. Voice
Chicago Press. and Equality. Cambridge, MA: Harvard University Press.