Hello, dear friend, you can consult us at any time if you have any questions, add WeChat: daixieit

MASTERCLASS ASSIGNMENT

Food Advertising, Emotions, and Food Choice

An Analysis of Experimental Data

In this assignment, you will analyse experimental data collected during the in-class session to study how food advertising and emotions affect unhealthy food choices. You will combine economic reasoning, existing literature, and statistical analysis to test three hypotheses about the causal effects of advertising and emotional states on behaviour. The final output is a short empirical report presenting your analysis, results, and policy implications.

Submission format	Written report (Word or PDF), max 4 pages excluding tables and references
Software	Stata or R (both accepted; include your code in your submission)
Group size	Individual
Data file	data_masterclass_simulation.dta (Stata) / data_masterclass_simulation.csv (R)

Background

You have been given simulated data from a between-subjects experiment investigating how exposure to unhealthy food advertisements and experimentally induced emotions jointly affect food choices. Understanding these mechanisms has direct relevance for public health policy, advertising regulation, and behavioural interventions.

Participants (N = 159) were randomly assigned to one of six conditions in a 2 × 3 design:

	Positive emotion	Neutral emotion	Negative emotion
Food advertisements (F)	+F	=F	−F
Non-food advertisements (NF)	+NF	=NF	−NF

After watching the advertisements and emotion-eliciting film clips, participants chose 5 food items from a list of 20 (10 healthy, 10 unhealthy). Incentive compatibility was ensured: 1 in 7 participants received their chosen items by mail.

Research Hypotheses

Your analysis should aim to test the following three pre-registered hypotheses:

H1 — Main effect of food advertising Exposure to online unhealthy food advertising leads to more unhealthy food choices compared to exposure to non-food advertising.

H2 — Main effect of emotions Experimentally induced emotions influence food choices: participants in a negative affective state will make more unhealthy food choices than those in a positive or neutral state.

H3 — Interaction (moderation) Emotions moderate the impact of food advertising on food choice: the effect of food advertising on unhealthy choices will differ depending on the induced affective state.

Part 1 — Literature Review (max 1 page)

Before presenting the data, you need to situate your analysis within the existing literature. You are writing as a health economist or policy analyst: justify why the question matters, and what the evidence says about whether and how advertising and emotions affect food choices. Write a short literature review (maximum 1 page, approximately 400–500 words). Below is a suggestion on how you could structure it.

1. What is the public health and economic case for caring about food choices? Think about the economics of obesity: external costs, internalities, and long-run health trajectories. Advertising restrictions are an increasingly common policy lever, what is the economic rationale for intervening in this market at all?

2. What does the causal evidence say about the effect of food advertising on food choices?

3. How do emotions enter the picture, and why should a policy analyst care? Emotions are not just a psychological curiosity. If they systematically shift food preferences, they become a confound that any advertising study must account for, and potentially a mechanism that advertisers deliberately exploit. Summarise what is known about how induced affective states shift food choices.

4. What is the policy implication if H3 (the interaction) is supported? Explains the rationale for expecting an interaction between advertising and emotion. If food advertising effects are amplified or dampened depending on emotional state, what does this mean for the design of advertising regulations? Does this create targeting concerns? Might advertisers place food ads strategically around content designed to evoke specific emotions? Frame this as a policy question.

Suggestion of journals to consult AER, Journal of Health Economics, Review of Economic Studies, European Economic Review, Journal of Behavioral and Experimental Economics, Agricultural Economics, American Journal of Agricultural Economics, Journal of Marketing Research, Health Economics, American Journal of Health Economics, Journal of Public Economics, Journal of Political Economy, Annual Review of Public Health. Be selective: 8 to 12 references total is plenty. Cite in APA 7th format. References do NOT count toward the 1-page limit.

Part 2 — Descriptive Statistics

2a. Sample Characteristics

Produce a summary statistics table describing your sample. The table should cover:

• Demographics: gender, area of residence (urban/suburban/rural)

• Internet use: frequency of going online, daily hours, years of internet experience, frequency of each online activity

• Self-reported attention to advertisements and food advertisements

• Hunger level at the time of the food choice task

• Baseline emotions: mean positive affect and negative affect before the manipulation

For each variable report: N, mean (or %), standard deviation (for continuous variables), and min/max. Format this as a publication-quality table — not raw software output.

2b. Manipulation Check — Did the Emotion Induction Work?

Before testing the hypotheses, you must verify that the experimental manipulation successfully induced the intended emotions. This is called a manipulation check.

The PANAS was administered before (p_a0, n_a0) and after (p_a1, n_a1) the film clips and advertisements. The difference scores (p_a_diff, n_a_diff) capture the change in affect.

5. Produce a table showing the mean positive affect scores and negative affect scores before, after, and their difference by emotion condition (positive / neutral / negative).

6. Test whether these differ significantly across the three emotion conditions using F-test.

7. Interpret your results in 3–4 sentences: was the manipulation successful?

Part 3 — Descriptive Statistics of Outcome Variables by Condition

Now turn to the main outcome variables. Your primary outcome is unhealthy_p (the proportion of unhealthy choices out of 5). Secondary outcomes include unhealthy_savory_p, unhealthy_sweet_p, and the parallel healthy counts.

3a. Outcome Variables by Experimental Condition

Produce a table showing means and standard deviations for the following outcome variables, broken down by the two marginal treatments separately: (i) food ad vs. non-food ad (foodad), and (ii) emotion condition (emotion). This will give you a first visual impression of the main effects.

• unhealthy_p — proportion of unhealthy choices (primary outcome)

• unhealthy_savory_p — proportion of unhealthy savory choices

• unhealthy_sweet_p — proportion of unhealthy sweet choices

• healthy_choices — count of healthy items chosen

• unhealthy_choices — count of unhealthy items chosen

 

An addition would be to add the t-test comparing pairs of treatments (see examples below)

 

	(1)	(2)	(1) vs (2)
	Non-food ads	Food ads	P-value
Proportion unhealthy
	(SD)	(  )
Proportion unhealthy savory
	(  )	(  )
Proportion unhealthy sweet
	(  )	(  )
N			159

 

	(1)	(2)	(3)	(1) vs (2)	(1) vs (3)	(2) vs (3)
	Positive	Neutral	Negative	P-value	P-value	P-value
Proportion unhealthy
	(SD)	(  )	(  )
Proportion unhealthy savory
	(  )	(  )	(  )
Proportion unhealthy sweet
	(  )	(  )	(  )
N

Part 4 — Regression Analysis

Parts 2 and 3 described the data. Now you will use regression to formally test the three hypotheses.

4a. Setting Up the Regression

Before running any models, you need to make a few preparatory decisions. 

8. Choose your reference category. For foodad, the natural reference is non-food ads (NF). For the emotion dummies, neutral is the natural reference.

9. Consider whether to control for covariates. Gender, hunger, and baseline affect (p_a0, n_a0) may predict food choices independently of the treatment. Including them as controls may improve precision.

4b. Model Sequence

Estimate one model per hypothesis. Present them side-by-side in a single regression table.

Model 1 (Bivariate)	Model 2 (Main effects)	Model 3 (Interaction)
Test H1 in isolation: does food ad exposure raise unhealthy choices?	Test H1 and H2 simultaneously, controlling for baseline differences between participants.	Test H3: does the effect of food ads differ by emotion condition?

4c. Interpreting the Regression Output

For each model, briefly interpret the results, mentioning whether the effect is significant and its meaning (e.g., 'exposure to X is associated with a Y percentage point increase in the proportion of unhealthy choices').

4e. Robustness Checks (optional but encouraged)

Stronger analyses test whether results are sensitive to modelling choices. Consider at least one of the following:

• Re-run Model 3 using the count variable unhealthy_choices (0–10) as the outcome, or separately for unhealthy_savory_p and unhealthy_sweet_p. Do the conclusions change?

• Because unhealthy_p is a proportion bounded between 0 and 1, OLS may produce predicted values outside [0,1]. As an extension, estimate a fractional logit (Stata: fracreg logit; R: glm with family=quasibinomial) and compare the results.

Part 5 — Discussion and Conclusion

In approximately 200 words, write a discussion section that summarises the main findings with respect to each of the three hypotheses (H1, H2, H3). State clearly whether each hypothesis is supported, partially supported, or not supported by the data. Interprets your results in the context of the literature reviewed in Part 1.

Grading Rubric

Section	Marks	Key criteria
Part 1 — Literature review	20%	Relevance of sources; clarity of theoretical argument; clear link to hypotheses
Part 2 — Descriptive stats	20%	Completeness of tables; quality of manipulation check; correct interpretation
Part 3 — Outcomes by group + F-tests	20%	Correct computation of group means; correct F-tests; clear figure; sensible interpretation
Part 4 — Regression	30%	Correct model specification; correct interpretation of coefficients and interactions; joint F-test; marginal effects; at least one robustness check
Part 5 — Discussion	10%	Accurate summary of findings; honest treatment of limitations; quality of writing
Code appendix	Bonus (+5%)	Clean, commented, reproducible code in Stata or R

Quick Reference — Key Variables

Variable	Type	Description
foodad	Binary (0/1)	1 = exposed to unhealthy food ads; 0 = non-food ads
emotion	Categorical (3 levels)	Positive / Neutral / Negative — assigned emotion condition
condition	Categorical (6 levels)	+F, =F, −F, +NF, =NF, −NF — full treatment cell
p_a0 / n_a0	Numeric (5–25)	Baseline positive / negative affect (PANAS sum, pre-treatment)
p_a1 / n_a1	Numeric (5–25)	Post-treatment positive / negative affect (PANAS sum)
p_a_diff / n_a_diff	Numeric	Change in affect (post − pre). Use for manipulation check.
fc_1 – fc_20	Binary (0/1)	Individual food choice indicators (1 = item selected)
unhealthy_p	Proportion (0–1)	PRIMARY OUTCOME: share of 5 choices that are unhealthy
unhealthy_savory_p	Proportion (0–1)	Share of choices that are unhealthy AND savory
unhealthy_sweet_p	Proportion (0–1)	Share of choices that are unhealthy AND sweet
healthy_choices	Count (0–10)	Number of healthy items chosen
unhealthy_choices	Count (0–10)	Number of unhealthy items chosen
hunger_1	Scale (1–10)	Self-reported hunger at time of food choice task
gender	Categorical	Male / Female / Non-binary / Prefer not to say
age	Numeric	Age

General Tips

• Always report your results in substantive terms, not just 'p < 0.05'. A statistically significant result with a tiny effect size may not be practically meaningful.

• Every table and figure needs a title, labelled axes/columns, and a note explaining abbreviations. Do not paste raw software output into your report.

• Use heteroskedasticity-robust standard errors throughout.