PH784: Problem Set #6
Read the following article and answer the questions.
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.
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.
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.
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.
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?
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.
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.