31.2 Estimating QTT with CiC
CiC relies on four distributions from a 2 × 2 Difference-in-Differences (DiD) setup:
- \(F_{Y(0),00}\): CDF of \(Y(0)\) for control units in period 0.
- \(F_{Y(0),10}\): CDF of \(Y(0)\) for treatment units in period 0.
- \(F_{Y(0),01}\): CDF of \(Y(0)\) for control units in period 1.
- \(F_{Y(1),11}\): CDF of \(Y(1)\) for treatment units in period 1.
The Quantile Treatment Effect on the Treated (QTT) at quantile \(\theta\) is:
\[ \Delta_\theta^{QTT} = F_{Y(1), 11}^{-1} (\theta) - F_{Y (0), 11}^{-1} (\theta) \]
To estimate the counterfactual CDF:
\[ \hat{F}_{Y(0),11}(y) = F_{y,01}\left(F^{-1}_{y,00}\left(F_{y,10}(y)\right)\right) \]
This leads to the estimation of the inverse counterfactual CDF:
\[ \hat{F}^{-1}_{Y(0),11}(\theta) = F^{-1}_{y,01}\left(F_{y,00}\left(F^{-1}_{y,10}(\theta)\right)\right) \]
Finally, the treatment effect estimate is:
\[ \hat{\Delta}^{CIC}_{\theta} = F^{-1}_{Y(1),11}(\theta) - \hat{F}^{-1}_{Y(0),11}(\theta) \]
Alternatively, CiC can be expressed as the difference between two QTE estimates:
\[ \Delta^{CIC}_{\theta} = \Delta^{QTE}_{\theta,1} - \Delta^{QTE}_{\theta',0} \]
where:
- \(\Delta^{QTT}_{\theta,1}\) represents the change over time at quantile \(\theta\) for the treated group (\(D=1\)).
- \(\Delta^{QTU}_{\theta',0}\) represents the change over time at quantile \(\theta'\) for the control group (\(D=0\)).
- The quantile \(\theta'\) is selected to match the value of \(y\) at quantile \(\theta\) in the treated group’s period 0 distribution.
Marketing Example
Suppose a company introduces a new online marketing strategy aimed at improving customer retention rates. The goal is to analyze how this strategy affects retention at different quantiles of the customer base.
- QTT Interpretation:
- Instead of looking at the average effect of the marketing strategy, CiC allows the company to examine how retention rates change across different quantiles (e.g., low vs. high-retention customers).
- Rank Preservation Assumption:
- This approach assumes that customers’ rank in the retention distribution remains unchanged, regardless of whether they received the new strategy.
- Counterfactual Distribution:
- CiC helps estimate how retention rates would have evolved without the new strategy, by comparing trends in the control group.