Empirical Exercise 12

In this exercise, we use data form the paper Enhancing Young Children’s Language Acquisition Through Parent-Child Book Sharing: a Randomized Trial in Rural Kenya by Lia C.H. Fernald, Heather A. Knauer, Pamela Jakiela, and Owen Ozier. The study examines the short-term impact of dialogic reading training combined with mother tongue storybooks. The authors use data from this short-term study to calculate the statistical power of a subsequent cluster-randomized trial, as described in Evaluating the Effect of an Early Literacy Intervention. Because child outcomes are highly correlated over time, selecting the right covariates can increase statistical power substantially.

Getting Started

Begin by creating a do file that reads in the data set E12-storybooks-data.dta1. Familiarize yourself with the dataset and the range of baseline variables included in it. The key outcome variable is e_zstoryexp, a measure of the expressive vocabulary children might have picked up from the storybooks. The data set also contains a large set of baseline covariates. Extend your do file as you answer the following questions.

Empirical Exercise

Question 1

Regress the outcome e_zstoryexp on treatment. What is the p-value associated with the test of the hypothesis that treatment has no impact on children’s vocabulary?

Question 2

Predict the residuals from the regression above. What is the variance of the residuals?

Question 3

Use lasso to identify the baseline covariates that predict treatment. Use cross-validation to choose the tuning parameter, lambda. Which variables does lasso select as predictors of treatment? Store the selected variables in a local named Tvars.

Question 4

Use lasso to identify the baseline covariates that predict e_zstoryexp. Use cross-validation to choose the tuning parameter, lambda. Make sure you tell lasso to always include the treatment variable. Which variables does lasso select as predictors of the outcome? Store the selected variables in a local named Yvars.

Question 5

Regress the outcome e_zstoryexp on treatment including all the controls selected by lasso. What is the p-value associated with the test of the hypothesis that treatment has no impact on children’s vocabulary?

Question 6

Predict the residuals from the regression that includes the controls selected by lasso. What is the variance of these residuals?

---

This exercise is part of the module Choosing Covariates.