Abstract

Aiken et al. (2015) and Davey et al (2015) draw the conclusion that the evidence for a relationship between deworming and school attendance is “weak” based on two fundamental errors in their data analysis. First, the authors redefine treatment to include pre-treatment control periods. Second, while the original research design was based on a stepped-wedge analysis that was adequately powered, the re-analysis authors undertake a clearly under-powered alternative analysis which ignores the time series element of the data, and then splits the cross-sectional analysis into two separate components, each of which has inadequate power. Examining the fully powered analysis, they report that in a fully-adjusted logistic regression model making maximum use of the data available, there is strong evidence of an improvement in school attendance.  If either error is corrected, deworming significantly increases school attendance under the full range of statistical analyses considered by Davey et al. Their analysis also underestimates the impact of deworming on school attendance by neglecting violations of the SUTVA assumption generated by transmission of worm infection to nearby schools (as in Miguel and Kremer 2004).

Supplementary Materials and Data