Reproduction Attempt for "Twenty Year Economic Impacts of Deworming"

Overview

The is a reproduction attempt for "Twenty Year Economic Impacts of Deworming" by Hamory et al. (2021). The effort followed the Guide for Accelerating Computational Reproducibility (ACRE) from the Berkeley Initiative for Transparency in the Social Sciences (BITSS). The reproduction attempt is logged here on the Social Science Reproduction Platform (SSRP).

The attempt follows the provided reproduction package, which includes an appendix.

Description

Hamory et al. expand the deworming literature by assessing impacts 20 years in the future. Data from the Primary School Deworming Project (PSDP), a randomized intervention conducted across control and treatment schools, is cross-referenced with the Kenya Life Panel Survey (KLPS) to assess a variety of living standards. Regression is conducted on two sets of KLPS data with the following specification:

where Y is an outcome for an individual i in school j in survey round t, T indicates treatment, C and P are cost-sharing and saturation indicators, and X denotes a panel of school and individual covariates.

Kenyan school children who received deworming interventions.

The treatment group experienced:

1. 305 USD PPP more consumption.

2. 80 USD PPP greater individual earnings.

3. 239 USD PPP greater household earnings.

4. 0.14 USD PPP greater hourly earnings.

5. 0.04% greater likelihood to live in urban areas

6. 1.91 more nonagricultural weekly work hours.

The social IRR is 37% and the 25 year, 10% discount rate NPV is USD PPP 230.71.

Package

Installations of R and Stata IC are required to run the analysis and cleaning scripts.

The following is a diagram of all inputs and outputs, where "manual" indicates a manual transcription. A full diagram is available on the ACRE platform.

  Table 1 - KLPS4_E+_earnings_consumption_main_byage.tex
  └── Earnings_consumption_hhearn_main_byage.do
      ├── Worms20_master_main.do
      │   └── Worms20_MasterDo.do
      │       └── Worms20_Globals.do
      └── Worms20_Analysis.dta
  Table 2 - KLPS4_E+_earnings_labor_occchoice_main_byage.tex
  └── Earnings_wealth_labor_occchoice_main_byage.do
      ├── Worms20_master_main.do
      │   └── Worms20_MasterDo.do
      │       └── Worms20_Globals.do
      └── Worms20_Analysis.dta
  In-Text Results
  └── Worms20_intext_calc.do
      ├── Worms20_Attrition.dta
      └── Worms20_Heterogeneity.dta
  Table S12
  └── Cost_benefit_table_pooled.Rmd
      └── Parameters

Improvements

Edits to the original file include:

• Master R script that removes the need for manually setting the directory, removes repetitive library calls, and compiles Cost_benefit_pooled.R, Consumption_graph.R, Earnings_graph.R, and Cost_benefit_table_pooled.Rmd.

• Swap "Figure 1" and "Figure 2" in Worms20_master_main.do.

• Added missing package texdoc on line 33 of Worms20_MasterDo.do.

• Removed

  set mem ___ set matsize ___ set maxvar ___

  from each file for compatibility with Stata IC.

• Added LaTeX template with \bettershortstack and \sym macros.

Robustness

A robustness check is conducted on results 4 through 6 (and, indirectly, result 2). A regression is conducted on only male and older subjects to isolate the confluence of the two variables. We adopt the following regression specification

where Y includes an individual's gender, age, and an interaction variable between gender and age.

The results, with the additional specification in column 4, are below:

Considering only older males increases the significance of every effect over male or older subjects alone. We observe particularly large increases in effect size for wage earnings (301 vs. 162), non-zero earnings (0.05 vs 0.02), total hours worked (3.94 vs 2.2), and non-agriculture hours worked (4.44 vs. 2.77). These results suggest that the overall increase may be driven in large part by older males specifically.