35.4 Policy Applications of Discrete Choice Models
35.4.1 (Khan, Misra, and Singh 2015)
Variation: prices vary wiht fat content level (
Price is determined at a regional level, and independent of local demand conditions (i.e., exogenous shocks)
Examine price sensitivity and substitution patterns (heterogeneous for different socioeconomic groups).
Higher price leads to more likely consumption of lower calorie milk.
- Especially for low-income households.
Recommendation: tax scheme based on relative prices of healthier options.
Interesting choice of presenting data in the introduction section
Data: IRI
35.4.2 (A. Rao and Wang 2017)
Demand reduced after the termination of the claims, (12 - 67 % monthly loss in revenue)
- The decline effects come mainly from newcomers.
35.4.3 (Tuchman 2019)
Descriptive evidence for e-cig ads reducing traditional cig (i.e., e-cig is a sub of traditional cig)
From structural models, propose counterfactual evidence for banning e-cig ad (but might increase traditional cig demand again)
35.4.4 (Seiler, Tuchman, and Yao 2020)
Examine the impact of sugar-sweetened beverages (SBB) tax on Philadelphia, where they found that cross-shopping to stores outside the area accounted for half the reduction in sales and decreases the net reduction in sales 22%
Key findings:
Tax pass through at an average rate of 97% (i.e., 34% price increase)
Price increase reduce quantity purchased by 46% (but half went to other stores outside of the city). Hence, the net sales of SSB decreased by 22%
Bottled water is not a substitute for SSB, but natural juices might.
Low income neighborhood just decreased demand (no increase in cross-shopping) due to limitation of transportation.
Counterfactuals:
15 cents per ounce is close to the revenue-maximizing tax rate (but 2 cents higher could be optimal because it lowers sales while costs marginally to tax revenue).
Initial plan of 3 cents per ounce could be detrimental (tax revenue decreases by 75%)
The authors have to argue for the paper’s contribution above the one studied in Berkeley and other places (i.e., representative demographics, and results)
Tax on distributors and only on artificial sweetener because of financial purposes
Data: IRI retail point-of-sale data 2015-2018, tax date = Jan 2017
Product aggregate at the brand/diet status/pack size level. (i.e., total units sold and quantity-weighted prices at the product/store/week level.
861 products (489 taxed, 372 untaxed)
data cover 28% of sales of taxed beverages
Demo data from Census Bureau and obesity rates from the CDC
Dif-n-dif research design:
Treatment; tax area
Control: 3-digit surrounding zipcode - 6-mile away (non-taxed)
Parallel trend pretax data.
\[ y_{st} = \alpha(\text{Philly}_s \times \text{AfterTax}_t) + \gamma_s + \delta_t + \epsilon_{st} \]
where
\(y_{st}\) = quantity sold and price
\(\gamma_s\) = store fixed effect
\(\delta_t\) = week fixed effect
\(\epsilon_{st}\) = error
\(\alpha\) = dif-in-dif coefficient
To assess heterogeneity
\[ y_{st} = \tilde{\alpha}_0 (\text{Philly}_s \times \text{AfterTax}_t) + (\text{Philly}_s \times \text{AfterTax}_t \times \mathbf{X}_s)' \tilde{\alpha}_1 + (\text{afterTax}_t \times \mathbf{X}_s)' \tilde{\mathbf{\beta}} + \tilde{\epsilon}_{st} \]
where
\(\tilde{\gamma}_s\) = store fixed effects
\(\tilde{\delta}_t\) = week fixed effects
\(\mathbf{X}_s\) = a set of store characteristics
\(\tilde{\mathbf{\beta}}\) = vector of coefficients capturing the change in the outcome in stores outside of Philly after the tax took effect as a function of \(\mathbf{X}_s\)
\(\mathbf{\tilde{\alpha}}_1\) = the differential change in the outcome in Philly stores relative to control group as a function of \(\mathbf{X}_s\)
\(\tilde{\mathbf{\alpha}}_0\) = baseline (i.e., uninteracted dif-in-dif estimate)
two-way clustered SE at the store and the week level
No single-term \(\mathbf{X}_s\) because fixed store effects already absorb all store characteristics.
Quantity:
The reason why drugstores and convenience stores experience modest to no decrease in quantity sold is because
- They already have higher pretax price level
- Consumers who buy at those places are less price sensitive
“Quantity decreases more in high-income areas” (contrary to intuition, high-income should respond less to changes in price, may be because of lower transportation costs).
“Obesity rates do not predict a differential quantity response.”
Provided evidence for revenue maximum relating quantity sold and price elasticity.