Experimental Design Details
We will run the following regressions, including dummies for each treatment assignment. The outcome variables for each regression are:
Door open (0,1)
Purchase F&N (0,1)
Purchase Sweets (0,1)
Purchase cond. on door open (0,1)
Purchase F&N cond. on purchasing (0,1)
Purchase Sweets cond. on purchasing (0,1)
We will alternatively report the results using these additional measures:
Added sugar purchased
Saturated fat purchased
All regressions will include RA, time-of-day and city (the experiment spans across different suburbs) fixed effects.
We will separately estimate these with additional demographic data that we collect, including:
1) RAs will record estimated race, gender, age bracket (young, adult, older adult) and whether individual or household made the decision, which we include as covariates
2) Census-tract level income data, and the prices of each food at 2 nearby stores (averaged by store/week).
Our post-estimation tests of interest are:
1) Comparing flyer vs. no flyer door opening rates. Lower door opening rates may be an indication of lower welfare, and higher door opening rates may be an indication of higher welfare. We will compare flyer, flyer with information about health benefits, and no flyer.
2) Comparing food choice in flyer vs. no flyer conditions. Different purchase rates and food choices may be an indication of sorting. For example, we may expect healthier food purchases in Flyer-Information relative to No Flyer-Information if the flyer induces higher health-conscious people to sort in.
3) Comparing food choice in immediate versus delayed. Different purchase rates and food choices may indicate present bias. For example, we expect more purchases of the Sweets box in immediate versus delayed.
4) Since the design is 2x2 (Immediate/Delayed and Information/No Information) crossed with Flyer/No Flyer, we also wish to look at different 2-way comparisons to understand the impact of the Information nudge under each treatment and with/without pre-notification as we expect these to vary.
5) Comparing $1 vs $2 treatment we expect a simple demand effect: more purchases at $1 than at $2. But this is not one of our key questions for the reduced form analysis.