Resolving Pasquale's Comments for Manuscript

[EmilHafeez]

• [x] 2.

There is only one red number he wants us to review, that I can see, in Table 1. We can use App Table 1 data to get it. "*Approximately 9.5% of restaurants were excluded from the analytic sample because of a lack of data available in the relevant baseline period (i.e., 3 to 8 months prior to menu labeling implementation)."

I vote we just use App Tab 1, take the proportion of: restaurants with data available in baseline period / restaurants with data available in baseline period, matched, open ever

which is (461+782)/(470+804)=0.9756672; we lose 2.4%.

The problem with this is that we do not have a variable like post6 that describes only the number of restaurants with 6 data rows between -3 and -8; we could both make this easily like we did post6 and produce this, though it's annoying for such a needless footnote

Alternatively, if it's percent of all restaurants that were ever available, our number looks pretty bad; the number of controls is inflated by duplicated restaurants across entry times, necessary for the unmatched analysis and App Table 1. Even if we ignore these duplicates and take the unique restaurants ignoring entry (so all unique control restaurants only), the number isn't great, something like 1200 out of 4500 retained

-- resolved, 470/473

• [x] 3. The test for heteroskedasticity rejects the null, suggesting it's present; we'll need to mount an argument/response.

[EmilHafeez]

• [x] Also, perhaps discuss ATT defense if it's relevant to ML meeting
• [x] Sig preds of calorie implemented before dropping bev calories, not after.

Here is vector of CA sig calorie predictors with beverage in

A tibble: 18 × 1

term

1 (Intercept) 2 concept 3 ownership 4 drive 5 meal 6 calorie3 7 slope_calorie 8 count3 9 dollar3 10 slope_calorie_all 11 count_all3 12 dollar_all3 13 total 14 white 15 hisp 16 capital_income 17 hsbelow 18 open18 Here is vector of CA sig calorie predictors with beverage excluded # A tibble: 19 × 1 term 1 (Intercept) 2 concept 3 ownership 4 drive 5 meal 6 calorie3 7 slope_calorie 8 count3 9 dollar3 10 slope_dollar 11 calorie_all3 12 slope_calorie_all 13 count_all3 14 slope_dollar_all 15 white 16 hisp 17 capital_income 18 hsbelow 19 open18

[EmilHafeez]

Also,

• [x] Appendix Table 3 print covariate balance table by location into cells rather than image; LLOYD
• [x] Emil and Lloyd to add footnotes to flag which variables are used in which matching approach; page 4 of 10; EMIL to comment, can we direct a reader to Appendix Table 3? Lengthy
• [x] Transactions during lunch (34.6%) and dinner (27.8%) represented the majority of sales by time of day (Appendix Figure 3); and the average calories per transaction was highest during dinner [1,143 (SD=142)] and evening [1,088 (SD=186)].
• • • THE SAME
• [x] Menu labeling impact on calories purchased during late night [β=-54.3 (95% CI: -69.5, -39.2)] and evening [β=-53 (95% CI: -55.9, -50.2)] were significantly larger than calories purchased in transactions made during other times of day in the 3-12 month period; whereas, the impact was larger during breakfast [β=-72.2 (95% CI: -73.5, -70.9)] relative to other times in the 13-24 month period.
• • • slightly different, appexndix table 6
• [x] Taken together, our analysis of XX million Taco Bell transactions demonstrates that menu labeling legislation may lead to modest and sustained population-level improvements in health,31 especially in combination with labeling in other settings and other types of healthy eating policies. EMIL

[EmilHafeez]

For menu labeling meeting:

• Heteroscedastic; CIs distant from 0
• Tod has not emailed re: ATT vs ATE; unclear where he's coming from, would need explanation. IPW specifies ATT.

[EmilHafeez]

Weighted plm into clubSandwich's CR setting; output 1.

Repeat the process and use the plm's equivalent in lm, and then pass that into clubSandwich CR, same settings; output 2

These results are different. SE's are (much) smaller in the LM though SE estimates that we start from are identical in each iteration.

[EmilHafeez]

Closing until reviewer comments on paper 1; if robust SEs required, we can switch to felm() package