PH784: Problem Set #6

Author

Koichiro Shiba

Published

November 11, 2025

Read the following article and answer the questions.

Notice that sections in this paper do not follow the conventional order (Introduction, Methods, Results, and Discussion). You will see Methods at the end of the paper, following the guidelines from the Science portfolio journals.

Briefly, this study estimated the heterogeneous effects of disaster-related traumatic experiences (home loss and loss of loved ones) on cognitive disability among older adult survivors of the 2011 Japanese earthquake and tsunami. We will focus on findings for the “home loss” exposure in this problem set. The home loss exposure was binary (0 = no loss or partial damage, 1 = complete home loss), and the continuous levels of cognitive disability were assessed in 2013 (2.5 years after the disaster onset) and 2016 (5.5 years after). The authors used the causal forest algorithm to estimate CATEs of trauma exposures conditional on various pre-disaster characteristics.


Q1

Table 1 shows the distribution of pre-disaster characteristics by exposure status. Explain how the exposed and unexposed individuals differ in terms of their socioeconomic status.

Your answer


Q2

Fig. 1 shows the estimated ATEs (\(E[Y^{a=1}-Y^{a=0}]\)) of the home loss exposure (\(A\)) on cognitive disability levels (\(Y\)). Provide an interpretation of the estimate \(0.14\) using counterfactual language.

Your answer


Q3

Fig. 2 shows the distribution of estimated CATEs (\(E[Y^{a=1}-Y^{a=0}|L]\)) conditional on 51 pre-disaster characteristics (\(L\)). Provide an interpretation of this estimand.

Your answer


Q4

Focus on the red curves in Fig. 2 (i.e., CATE distributions for the home loss exposure). Which tail of the distribution shows “resilient” individuals for whom home loss appeared less harmful—right tail or left tail? Justify your answer.

Your answer


Q5

One of the 51 pre-disaster characteristics is education. One could fit the following parametric regression model to examine effect heterogeneity by education, adjusting for other covariates \(L_{-edu}\).

\[E[Y|A, Education, L_{-edu}] = \beta_0 + \beta_1 A + \beta_2 Education + \beta_3 A \times Education + \gamma L_{-edu}\]

What are the underlying assumptions about effect heterogeneity this model is making?

Your answer


Q6

Table 3 compares pre-disaster characteristics of the “resilient” vs “vulnerable” groups (i.e., extreme tails of the CATE distribution as discussed in Q4 above). The results showed that the vulnerable group tended to be those with lower educational attainment compared to the resilient group.

Based on this finding and your answer to Q1, explain whether you think disaster-induced home loss widened or narrowed education-based disparities in cognitive disability levels.

Your answer

Q7

Discuss what might explain the observed effect heterogeneity by education. The first author of this paper apparently did a poor job providing alternative explanations comprehensively, so you need to think beyond what’s written in the paper.

Your answer