Supplementary Material

For: “AI in essay-based assessment: Student adoption, usage, and performance”

Table S1
Descriptive statistics for survey participants versus class peers

Variable Survey No Survey Difference p-value

Age 21.68 21.71 -0.03 .931
Female 0.46 0.38 0.08 .272
English 0.51 0.50 0.01 .924
International 0.48 0.42 0.06 .405
Economics 0.42 0.38 0.04 .602
GPA 5.28 4.81 0.47 .001***
External 0.20 0.32 -0.12 .080*
Full-time 0.93 0.81 0.11 .032**
Number of students 69 118

Note. The table displays the mean of each row characteristic (column 1) for the survey respondents (column
2) compared to their class peers (column 3). ‘English’: English is the main language at home.
‘International’: International student. ‘Economics’: Economics degree. ‘GPA’: Degree GPA (7-point scale)
up to and including the previous semester, i.e., before the release of ChatGPT. ‘External’: Study mode was
fully online. The p-values are calculated from two-sample t-tests with equal variance.
*
p<.10; **p<.05; ***p<.01.

1
Table S2
Effect of AI on Student Performance: Number of AI components

Marks (out of 100)

Variable (1) (2) (3) (4) (5)
Number of AI
1.207 -0.233 0.663 -0.140 -0.236
components
(1.310) (1.168) (1.258) (1.062) (1.036)
GPA 8.594*** 3.695 4.437**
(1.932) (2.402) (1.934)
Economics 5.349 8.747** 0.349 0.078
(3.769) (3.989) (3.663) (3.598)
Test 1 score 0.065 -0.022
(0.123) (0.106)
**
Test 2 score 0.260 0.075
(0.112) (0.107)
Age 0.661 0.636
(0.566) (0.553)
Female 2.044 1.899
(3.453) (3.368)
International -6.802 -6.737
(6.164) (6.081)
English 11.807* 11.807*
(6.348) (6.267)
*** * ***
Constant 63.844 18.515 37.066 24.285 25.246
(2.926) (9.948) (9.838) (17.233) (16.868)
Observations 69 69 69 69 69
2
R 0.013 0.296 0.182 0.492 0.488
Adjusted R2 0.002 0.264 0.131 0.415 0.429
Note. This analysis replicates Table 2 in the main results after substituting the binary indicator of any AI use
with a more granular measure of the number of research essay components in which a student reported using
AI assistance. The dependent variable is the student’s mark on the assessment, graded out of 100. Standard
errors in parentheses. *p<.10; **p<.05; ***p<.01. Student ability is controlled for via their pre-semester GPA
and their scores on Test 1 and Test 2, the course’s earlier assessment items. Because the Tests were
conducted after the release of ChatGPT (even though AI use was not permitted), Model 5 omits the Test
scores.

2
Table S3
Split-sample regressions of the effect of AI use on performance

Marks (out of 100)

Male Female Domestic International Non-English English Economics Non-Economics Low GPA High GPA
Variable (1) (2) (3) (4) (5) (6) (7) (8) (9) (10)
Used AI 0.608 1.647 3.497 -5.112 -5.107 3.819 -1.219 2.818 -4.834 2.101
(5.328) (4.326) (5.035) (5.120) (5.473) (4.522) (4.559) (6.742) (5.224) (6.094)
GPA 1.978 5.585** 0.049 7.192*** 6.468** 2.207 6.333** 3.061
(3.416) (2.142) (3.449) (2.285) (2.457) (3.038) (2.448) (3.701)
Economics -2.766 -4.136 3.355 -2.726 -0.384 -0.395 3.223 -3.870
(5.134) (5.270) (4.476) (5.889) (6.113) (4.246) (6.279) (5.114)
Age 0.439 0.323 0.449 2.377 3.168* 0.186 1.114* -0.423 0.557 0.513
(0.642) (1.285) (0.585) (1.570) (1.558) (0.534) (0.650) (1.061) (1.052) (0.675)
International -19.565* -0.179 -14.383* -2.935 -0.599 -21.646* -22.244** 6.574
(10.549) (7.094) (7.947) (8.605) (7.755) (11.080) (10.387) (8.187)
Female 6.014 -1.182 -3.797 10.686** -1.014 8.908 -0.164 4.609
(4.763) (4.549) (4.840) (4.433) (4.645) (5.733) (4.583) (5.379)
English -4.313 28.053*** 10.262 19.596** 11.979 6.037 4.024 24.078**
(9.934) (8.578) (8.393) (9.299) (8.294) (11.327) (9.477) (9.129)
Constant 58.601** 18.707 49.686** -27.042 -24.806 54.162** 3.019 58.078* 62.396** 40.839**
(26.428) (28.583) (23.518) (36.276) (37.757) (20.700) (21.837) (28.816) (26.196) (16.727)
Observations 37 32 36 33 34 35 40 29 35 34
R2 0.329 0.711 0.130 0.427 0.380 0.205 0.511 0.492 0.564 0.336
Adjusted R2 0.195 0.642 -0.050 0.295 0.243 0.035 0.422 0.354 0.471 0.189

Notes The dependent variable is the student’s mark on the assessment, graded out of 100. The model title indicates the sample
restrictions, with each pair of models using a common demographic variable to split the sample. E.g. (1) uses the sample of males,
(2) uses the sample of females. ‘Non-English’: English is not the spoken language at home. ‘Low GPA’: Student’s GPA is below
the sample median. Standard errors in parentheses. *p<.10; **p<.05; ***p<.01.

3
Table S4
Effect of AI use on performance: Intention-to-treat analysis

Dependent variable:
Research Proposal Test 1 Score Test 2 Score
Variable (1) (2) (3) (4) (5) (6) (7) (8) (9)
Year = 2023 -0.468 1.812 2.637* -1.212 0.897 1.635 -3.687* -0.773 1.971
(1.702) (1.471) (1.461) (1.794) (1.614) (1.847) (2.177) (1.880) (2.167)
GPA 8.227*** 6.118*** 7.639*** 7.080*** 11.081*** 11.229***
(0.808) (0.800) (0.887) (1.012) (1.033) (1.187)
Economics 5.982*** 0.513 5.435*** 4.289** 5.562*** 7.582***
(1.509) (1.567) (1.656) (1.981) (1.930) (2.324)
Age 0.218 0.171 0.301
(0.310) (0.393) (0.460)
Female 0.301 0.376 -1.350
(1.479) (1.871) (2.194)
International -13.287*** -3.287 -1.614
(2.398) (3.033) (3.558)
English 3.270 -2.242 -5.568
(2.379) (3.009) (3.530)
Constant 66.286*** 20.621*** 31.801*** 70.242*** 27.891*** 29.097*** 74.519*** 14.153*** 7.414
(1.179) (4.248) (8.385) (1.242) (4.661) (10.605) (1.507) (5.431) (12.440)
Observations 390 390 323 390 390 323 390 390 323
R2 0.0002 0.270 0.440 0.001 0.209 0.189 0.007 0.275 0.271
Adjusted R2 -0.002 0.265 0.428 -0.001 0.203 0.171 0.005 0.270 0.255

Note: The table displays the cohort effect on student assessment scores. ChatGPT was released in
November 2022, and so the cohort year is used as an instrument for access to AI tools. The coefficient on
Year = 2023 can be interpreted as the “intention-to-treat” effect – the overall effect of allowing the use of
AI, whether adopted by the student or not – on student performance. There is no strong evidence that access
to ChatGPT increased performance on the Research Proposal assessment (columns 1-3; marks out of 100 for
each assessment). For comparison, there is no effect on the earlier assessment items, which were exams with
a mixture of multiple-choice and calculation questions. These tests were checked by the lecturer to ensure
they were “ChatGPT-proof” (that is: ChatGPT suggested the incorrect answer more often than not).
Standard errors in parentheses. *p<.10; **p<.05; ***p<.01.

4
Figure S1.
AI Detection and Actual Use.

Note. The x-axis displays Turnitin’s AI writing detection score, which estimates the overall percentage of
the document that AI tools like ChatGPT may have generated. ‘Major use’: Student reported using AI in at
least one of the four major written sections of the assessment (Introduction, Literature, Method, Conclusion).
‘Minor use’ is defined as using AI for generating the topic and/or writing the Title, Abstract or References.

5
Thematic Analysis: Coding process description

The coding process involved a systematic approach to analysing the 24 open-ended responses from students.
First, all responses were entered into the software NVivo (version 14), and initial codes such as ‘Research
question formulation’ and ‘Data analysis assistance’ were assigned to the key points of each segment of text.
Next, these codes were grouped into broader themes that categorized how students used AI tools in their
research proposals and their reflective experiences (Table S5).

Table S5
Frequency of themes

Theme Number of Responses

Idea generation and brainstorming 14
Writing and structuring assistance 12
Literature search and reference assistance 11
Grammar and error checking 4
Statistical and data analysis assistance 8
Overall usefulness and learning impact 9

Methodological Details
Additional methodological details supporting the thematic analysis include:
 Inter-rater reliability: The open-ended responses were independently evaluated by three economics
lecturers. Each response was marked for thematic alignment, and the binary assessments were
averaged across the lecturers to categorise the responses. This ensured consistency and reliability in
the coding process.
 Criteria for assigning codes: Codes were assigned based on the relevance and significance of
segments of text in relation to the themes. For example, segments describing the use of ChatGPT for
generating research ideas were coded under “Idea generation and brainstorming,” while segments
highlighting grammar corrections were coded under “Grammar and error checking.”

Illustrative Quotes
Below are illustrative quotes for each theme, along with an identifier for the respective student.

1. Idea generation and brainstorming

 “I found ChatGPT useful to provide a starting point for this assessment. I used it for coming
up with basic ideas, so I was then able to expand upon these.” (S3)
 “I used ChatGPT mainly for brainstorming and also workshopping my research idea.” (S11)
2. Writing and structuring assistance
 “I used ChatGPT to help understand how I should structure my method section. I also used it
when going through journal articles as a time saver by asking it to simplify and summarise
what the article was testing.” (S10)
 “ChatGPT helped me structure my research proposal and ensure coherence in my writing.”
(S9)
3. Literature search and reference assistance
 “I used ChatGPT only for idea generation and directing me to obscure references when I was
having trouble finding appropriate articles.” (S12)
 “ChatGPT assisted me in finding relevant literature, although some suggestions were not
reliable.” (S2)
4. Grammar and error checking
6
  “ChatGPT was useful in identifying grammatical mistakes, but it was very bad/useless in
generating 'good' ideas as they were all very basic.” (S1)
 “I used ChatGPT mostly as a Grammarly type tool to restructure my sentences and make
them more concise.” (S9)
5. Statistical and data analysis assistance
 “ChatGPT can effectively help me propose corresponding detections based on the
intelligence and data I have collected, such as t-testing, binomial distribution testing, and chi
square testing.” (S6)
 “I used ChatGPT to recommend a method of statistical analysis for my project, as it seemed a
necessary inclusion but something I had not been taught during the course.” (S15)
6. Overall usefulness and learning impact
 “I believe it had no effect on my learning. ChatGPT could potentially help generate ideas, and
even write the whole proposal, but it will be extremely poor, basic and lacking everything we
learnt in class.” (S22)
 “ChatGPT greatly reduced the time I spent in the initial stages of planning and researching.”
(S23)
 “ChatGPT was very helpful for brainstorming and structuring my proposal, but I still had to
do most of the work myself. Overall, I would have learned less without it.” (S24)

7
 Appendix A

Research Proposal Instructions for Students

The following pages contain the research proposal instructions provided to students for their final
assessment. Students were additionally supplied with examples of past submissions and their marks for a
range of performance levels.

Summary of task
The goal of this assessment is to write an original and interesting proposal for a research project related to
behavioural economics. You will come up with your own research question and describe a practical and
feasible method for how you would go about answering it. The word limit is 1,500 words, which excludes
the title page, abstract, tables and figures, references, and any appendixes. This is a hard word limit, and
you will be penalised for exceeding it. Submission will be through Turnitin.

Getting started
Your first step is to decide your topic, and specifically, your research question. For example, it could be a
test of a mainstream economics concept, an extension of a behavioural economics concept, or the
application of a behavioural economics theory to a novel population or setting (such as a behavioural
‘nudge’). Typically (but not always), a good question can be rewritten in the form “Does X cause Y?”,
where X and Y are some characteristics or outcomes of interest.

There is quite some flexibility with the format of your proposal, and you should not feel constrained by the
guidelines below if you think you can do it better with your particular topic. These guidelines are there to
help improve the quality of your proposal, but if you can write a high-quality paper in another way, you will
not be penalised for unorthodoxy. Likewise, feel free to include anything in an appendix that is not an
appropriate fit in the main document but that you do refer to in the text, such as specific experiment
instructions, a variable dictionary for an existing dataset, or other background information or material.
Most proposals will not need an appendix, but, seeing as it does not count for your word limit, you may wish
to exploit it for material that is relevant but not critical to your research proposal.

Regardless of your preferred format, however, your structure must include the following sections:

1. Title page
2. Abstract
3. Introduction
4. Related literature (with mandatory recent journal article reference)
5. Method (with mandatory inclusion of a table or figure)
6. Conclusion
7. References

You may add other sections or subheadings for readability if you wish, but be careful of your word limit.

Details of each section

1. Title page

8
Your title page should contain your proposal’s Title, your name and student number, date of submission,
your abstract, and an accurate word count (which excludes the title page, abstract, tables and figures,
references, and any appendixes).

2. Abstract

The abstract (placed on your title page) should be a four-sentence summary of your entire research
proposal, so make sure it includes what you believe are the absolute key ingredients of your paper. The
structure and content are flexible, but a typical format would be:
 First sentence: jump right into the topic or research question. (In some cases, you might prefer a
first sentence of motivation instead.) It would typically start with something like “This research
proposal…” followed by what it is you are proposing to investigate/answer.
 Second sentence: State what the contribution is of your research proposal. What has the previous
literature proposed or found, and what will your proposal contribute to fill the gaps in our
knowledge? E.g. “While previous literature has found that/assumes that/predicts that […], this
proposal contributes to our knowledge of this issue by testing whether […]”
 Third sentence: This should be about how you plan to answer your question. State your method (or
‘empirical’ approach). E.g.: “I propose…” followed by the method, e.g. “I propose to run a lab
experiment in which I will test how changing X affects people’s behaviour towards Y.” If there is a
treatment/explanatory variable and an outcome variable (and there should be!), or details of a
specific context or sample that is being tested, they should be clearly specified.
 Fourth sentence: Add any remaining critical details of the method, and/or include the bare
minimum of your analysis plan: what do you plan to test, how do you plan to test it (e.g. a t-test
between control and treatment groups, or a linear regression on your data), and what conclusions
will you draw depending on these results? E.g. “I will run a t-test to determine whether there are
differences in [Y] between the treatment and control groups, and if the treatment group’s [Y] is
significantly larger, this would support the hypothesis that […]”.

3. Introduction

This section should be started on a new page to your title page/abstract, and should be short, typically 1-3
paragraphs. Provide a motivation for the broad topic and why you think the reader should be interested in
it, and introduce any relevant economic theories (mainstream or behavioural). You may wish to highlight
where there is currently a gap in our knowledge (which your research proposal will aim to fill). But don’t
dwell for too long on setting up the context; make sure your research question clearly appears by the end of
the first paragraph. If different school of academic thought predict different answers to your research
question, you may then want to spend 1-2 sentences outlining these predictions, and/or to briefly introduce
which method you are proposing to use to test these predictions.

In terms of your writing, try to avoid flowery language, embellishments, or ambiguities. Write clearly and
matter-of-factly. Academic papers are often considered ‘dry’ in style, which can be true; the purpose of an
academic paper is not to entertain with the writing, but to convey the material as clearly as possible; how
‘interesting’ an academic paper is will be typically judged on the worthiness of the topic and the quality of
the research.

Here are some examples of introductions of behavioural economics papers that broadly follow the structure
that is expected of you.

Most children think of their potential future occupations in terms of what they will be (firemen,
doctors, etc.), not merely what they will do for a living. Many adults also think of their job as
an integral part of their identity. At least in the United States, “What do you do?” has become
as common a component of an introduction as the anachronistic “How do you do?” once was,

9
 yet identity, pride, and meaning are all left out from standard models of labour supply. This
omission is understandable: identity, pride, and meaning are difficult to quantify and are thus
hard to incorporate into the empirically driven field of labour economics.

In this article, we focus on minimal perceived meaning by the labour producing force and
investigate how it influences labour supply in controlled laboratory experiments. Our intention
is to compare situations with no meaning (or as low a level of meaning as we can create) with
situations having some small additional meaning. Thus, our investigation will focus not on
occupations highly endowed with meaning, like medicine or teaching, but on the least-common
denominator of meaningfulness that is shared by virtually all compensated activities.
– Ariely, Kamenica and Prelec (2008)

Neoclassical models include several fundamental assumptions. While most of the main tenets
appear to be reasonably met, the basic independence assumption, which is used in most
theoretical and applied economic models to assess the operation of markets, has been directly
refuted in several experimental settings
(Knetsch 1989; Kahneman, Knetsch, and Thaler 1990; Bateman et al. 1997). These
experimental findings have been robust across unfamiliar goods, such as irradiated
sandwiches, and common goods, such as chocolate bars, with most authors noting behaviour
consistent with an endowment effect. Such findings have induced even the most ardent
supporters of neoclassical theory to doubt the validity of certain neoclassical modelling
assumptions. Given the notable significance of the anomaly, it is important to understand
whether the value disparity represents a stable preference structure or if consumers’ behaviour
approaches neoclassical predictions as market experience intensifies.

In this study, I gather primary field data from two distinct markets to test whether individual
behaviour converges to the neoclassical prediction as market experience intensifies.
– List (2003)

Charitable contributions in the United States were estimated to exceed $300 billion annually in
2007, 2008, and 2009. This is roughly $1000 for each person in the US, a not insignificant
amount. Given the reliance of charitable organizations on these contributions, it is quite
important to try to identify and implement effective methods for enhancing the revenue
received. There has been some recent work on suggested donations to public radio, and some
study of the notion of paying-what-you-want as a pricing device. We extend both of these
notions to fund-raising in a restaurant venue, exploring whether the suggested amount (if any)
mattered with respect to the contributions raised.

Businesses like grocery stores and restaurants often ask customers (typically through having a
donation jar at the check-out register) to donate money to a certain charity organization. One
often sees a suggested certain donation level. But there has been little by way of systematic and
controlled study regarding how the suggested donation level affects behaviour in this
environment. Our research question is to attempt to determine the optimal amount to suggest,
or whether it is better to make no suggestion.
– Charness and Cheung (2013)

Improving energy efficiency reduces costs for firms and mitigates CO 2 emissions. This is
particularly important in the transportation sector, which is responsible for approximately
25%–28% of greenhouse gas emissions in Western industrialized countries (cf. EEA, 2018,
EPA, 2018). Fuel accounts for around 40% of variable costs for transportation companies. We
conducted an analysis to determine if loss aversion helps motivate drivers to drive in a fuel-
efficient manner. If successful, this could reduce fuel consumption by about 22%.
– Hoffman and Thommes (2020)

10
 Nudging has been found to affect human behavior across a wide range of domains. In
particular, it has been used to improve the payment morale of citizens when they owe money to
public institutions. While the traditional view (Allingham and Sandmo, 1972) considered
citizens’ tax compliance as a matter of audits and harsh fines, it is by now well understood that
tax morale is also a very important factor for compliance (Kirchler, 2007). In fact, nudging has
been frequently applied to improve tax morale, even though with mixed results. In the realm of
taxation, taxpayers are very likely to anticipate, however, that the government will ultimately
enforce correct tax payments, which is why nudges might have a good chance to work. In other
situations, however, public institutions may not want to enforce the collection of citizens’
payments for social or ethical reasons. Whether or not nudging also works in such a setting
and whether it can have persistent effects even after abolishing the nudge again are the key
questions of this paper.
– Sutter, Rosenberger and Sutter (2020)

4. Related literature

Please make sure that you clearly relate your research question to the existing academic literature. What
are the possible answers to your research question that have been discussed in past studies? What papers
answer a similar question to yours, and what do they find? (For example, if you are researching “Do tennis
players exhibit loss aversion?”, you would want to cite studies that investigate whether other sporting
players exhibit loss aversion). It is possible that closely related studies come from fields other than
economics, such as psychology or even more specialised fields (for example, in the previous example,
papers from sports journals might be relevant). But when in doubt, prioritise economics papers.

What you must definitely avoid is proposing a study that has already been carried out. So, make sure that
your literature search is thorough. Google Scholar is the best place to start, and once you find a close
paper, use the “Cited by” feature to filter by recent, related papers.

A common question is “How many papers should I cite?” This is hard to answer other than the general
comment “The most important ones, but no more”. While it is important not to omit any critical paper, it is
equally important not to spread yourself too thin such that you cite many papers but with insufficient detail
for the relevance to be clear to the reader. Here are some types of examples.
 If your proposal is an extension of one specific paper, then you may justify citing only this paper, so
that you can go into deep detail about this paper and what your extension contributes to it. For
example, you may be adding an original extension to the design of Niederle and Vesterlund’s (2007)
competitiveness experiment.
 If your extension has a very similar design to one study but applies it in a different domain – for
example, you apply the Apesteguia and Palacios-Heurta (2010) paper about soccer penalty kicks to
rugby union – you would want to (at least) cite both this paper and the most relevant paper about
psychological pressure in your new domain (rugby union).
 If your research proposal tries to reconcile two or more papers that reached contradictory
conclusions, then you would want to describe these papers in detail (and you may not need to cite
more). For example, Albrecht and Smerdon’s (2022) design references three contradictory theories
in its review, and cites the main papers for each theory.
 If your research proposal covers several topics – for example, you are comparing whether
confirmation bias or the sunk cost fallacy can best explain why people don’t sell their crypto
investments – you may need to cite more papers (in this case, ones on both biases in general, and
also on broad psychological biases in the crypto market). Many research topics fall into this
category.

11
At the end of this section, you should state in one sentence what the specific contribution of your research
would be to this literature. What is the gap that you are filling in the academic landscape?

5. Method

This should be the longest section of your proposal, roughly half of your allocated word count. Typically,
the method for your research proposal will be either an experiment (lab or field) or an empirical study of
existing data. Your proposed method must be: (a) able to answer your research question, (b) practical, and
(c) ethical.

If you choose an experiment, your proposal must include the following details:
 The experimental design, including the type and number of subjects, the groups, and how you will
administer the treatment(s)
 The experimental procedure. This can be in broad terms, but all critical information must be
included such that another researcher who reads your proposal would be able to implement the
experiment you describe
o You may wish to check the experimental papers assigned as readings in this course for
examples of how to describe the experimental design and procedures.

If you choose an empirical study of existing data, your proposal must include the following details:
 The source of the data (e.g. “OECD PISA data wave 2015”), or how you would plan to collect it
(e.g. “Scrape all Champion’s League football games from 2021-22 from the UEFA website”)
 The key explanatory variable(s) (this would be the ‘X’ variable) and outcome variables(s) (the ‘Y’
variable) from the data
 An explanation of how you plan to address any potential statistical biases such as selection bias in
your analysis. This may involve describing additional control variables that you propose to include
in your analysis of the data. If your design will make use of a natural experiment, clearly detail the
source of the randomisation and why it means that your proposed method will accurately answer
your research question.
o For example, in Apesteguia and Palacios-Huerta (2010), the method of choosing which
football side gets the first penalty kick is random, which prevents selection bias. In Gong
(2015), the author made use of an existing program (the VCT) that randomly assigned HIV
testing.

No matter which method you use, you should clearly describe your treatment variable (or variables; the
‘X’) and outcome variable (or variables; the ‘Y’).

Next, you should state how you plan to analyse the (experimental or natural) data, including any statistical
tests that you propose to use (such as a t-test). If there are other variables that are important in your dataset
(either an existing data set or one you will collect from your experiment), describe them and how you will
use them.

Finally, you should clearly state your hypotheses as they relate to the variables and tests. This will include
any sub-sample effects (also known as “heterogeneous effects”), e.g., does your effect differ for males and
females? (and how would you test this)?

6. Conclusion

Your conclusion should be short (1 paragraph). You may wish to describe what you will conclude about
your hypotheses or the motivating theories depending on which way your results turn out, as well as any
limitations of your research or risks for its implementation (and how these might be mitigated). You should

12
also describe the implications that you think your results might have, for either existing economic theories
or for policy-makers / industry / other relevant groups.

7. References

Your research proposal should be fully (and correctly) referenced, both within the text and by including a
full bibliography. You are free to use any of the standard referencing styles so long as you are consistent.
To save time and guarantee accuracy, especially if you use a lot of references, you may wish to use a
referencing software like Zotero or Endnote. For instance, Zotero (free!) can be installed as a web browser
extension, which is very handy because once you find a paper online, you can import it into your Zotero
library with one click. It also has a Word extension, meaning that you can import your library references
into your Word document and also add an automated bibliography of references that updates by itself. (If
you don’t use many references, it’s just as easy or easier to do things manually.)

13
Criteria & Marking:

Criteria Maximum possible marks

Presentation 20
 Clarity and quality of written expression (15)
 Referencing (5)
Research Topic 10
 Originality (5)
 Clarity and justification of research question (5)
Literature and economic concepts 20
 Identification and critical use of literature (10)
 Understanding and explanation of relevant economic
concepts (10)
Research Methods 40
 Appropriateness and feasibility of proposed method
(5)
 Suitability to test a causal claim while avoiding
selection and other biases (5)
 Description of method, including setting, sample,
variables, design, procedures (20)
 Hypotheses and analysis plan (10)
Implications and limitations 10
 Implications of research (5)
 Limitation/risks of research proposal (5)
Total 100

Submission:

Submit via the Turnitin link on Blackboard by the specified deadline. Requests for the granting of extensions
must be made online with supporting documentation before the submission due date/time. If an extension
is approved, the new agreed date for submission will be noted on the application and the student notified
through their student email. Extensions cannot exceed the number of days you suffered from a medical
condition, as stated on the medical certificate.

If your proposal is submitted late without an approved extension, a penalty of 10% of the maximum possible
mark will be deducted per day for up to 7 calendar days, at which point any submission will not receive any
marks unless an extension has been approved. Each 24-hour block is recorded from the time the submission
is due.

14
 Appendix B

Survey

The following pages contain the Qualtrics survey sent to students after their assessment submissions had
been marked.

Start of Block: section_use

studentID Student ID: (e.g. 12345678)

________________________________________________________________

ChatGPT_use Did you use ChatGPT or any similar AI chatbot (e.g. Bing Chat, Google Bard) for help with
your research proposal?

o Yes (1)
o No (2)
End of Block: section_use

Start of Block: section_noChatGPT

Display This Question:
If Did you use ChatGPT or any similar AI chatbot (e.g. Bing Chat, Google Bard) for help with your re... = No

noChatGPT_why Could you briefly list your reason(s) why you didn't use ChatGPT (or similar) on your
assignment?

o It didn't feel ethical (1)

o I didn't know it was allowed (2)
o Too lazy (3)
o I didn't know how (4)
o I didn't trust its output (5)
o I wanted it to be all my own work (6)
o Other (please explain): (7) __________________________________________________
15
End of Block: section_noChatGPT

Start of Block: section_ChatGPT

Display This Question:
If Did you use ChatGPT or any similar AI chatbot (e.g. Bing Chat, Google Bard) for help with your re... = Yes

ChatGPT_idea Did you use ChatGPT (or similar) for help with coming up with your research proposal idea?

o Yes (1)
o No (2)

Display This Question:

If Did you use ChatGPT or any similar AI chatbot (e.g. Bing Chat, Google Bard) for help with your re... = Yes

ChatGPT_sections Finally, please select all the sections of your research proposal in which you used
ChatGPT (or similar).

▢ Title (1)
▢ Abstract (2)
▢ Introduction (3)
▢ Related Literature (4)
▢ Method (5)
▢ Conclusion (6)
▢ References (7)
End of Block: section_ChatGPT

Start of Block: section_feedback

ChatGPT_feedback If you want to leave any comments, feel free to do so in the text box below.

________________________________________________________________

________________________________________________________________

________________________________________________________________

________________________________________________________________

________________________________________________________________

16
End of Block: section_feedback

Start of Block: Consent

Consent [Redacted for Anonymity]

End of Block: Consent

Start of Block: section_use

studentID Student ID: (e.g. 12345678)

________________________________________________________________

ChatGPT_use Did you use ChatGPT or any similar AI chatbot (e.g. Bing Chat, Google Bard) for help with
your research proposal?

o Yes (1)
o No (2)
End of Block: section_use

Start of Block: section_noChatGPT

Display This Question:
If Did you use ChatGPT or any similar AI chatbot (e.g. Bing Chat, Google Bard) for help with your re... = No

17
noChatGPT_why Could you briefly list your reason(s) why you didn't use ChatGPT (or similar) on your
assignment?

o It didn't feel ethical (1)

o I didn't know it was allowed (2)
o Too lazy (3)
o I didn't know how (4)
o I didn't trust its output (5)
o I wanted it to be all my own work (6)
o Other (please explain): (7) __________________________________________________
End of Block: section_noChatGPT

Start of Block: section_ChatGPT

Display This Question:
If Did you use ChatGPT or any similar AI chatbot (e.g. Bing Chat, Google Bard) for help with your re... = Yes

ChatGPT_idea Did you use ChatGPT (or similar) for help with coming up with your research proposal idea?

o Yes (1)
o No (2)

Display This Question:

If Did you use ChatGPT or any similar AI chatbot (e.g. Bing Chat, Google Bard) for help with your re... = Yes

18
ChatGPT_sections Finally, please select all the sections of your research proposal in which you used
ChatGPT (or similar).

▢ Title (1)
▢ Abstract (2)
▢ Introduction (3)
▢ Related Literature (4)
▢ Method (5)
▢ Conclusion (6)
▢ References (7)
End of Block: section_ChatGPT

Start of Block: section_feedback

ChatGPT_feedback If you want to leave any comments, feel free to do so in the text box below.

________________________________________________________________

________________________________________________________________

________________________________________________________________

________________________________________________________________

________________________________________________________________

End of Block: section_feedback

19