Replication package (Stata) for "How Many Observations Are Behind Each Cell? Weighting and Estimator Choice, Not Aggregation, When Georeferenced Surveys Meet Conflict Data"

Résumé fourni par la source

Complete replication package for the paper. Eighteen seeded Stata do-files and a shared Mata library rebuild both estimation panels from the raw Demographic and Health Surveys and ACLED sources and reproduce every table, figure and simulation in the article and in its online appendix, driven by a single master file. The package covers the cell-year panel used to implement the standard grid-cell design, the woman-month hazard panel used at the individual level, the constant-estimator decomposition, a Monte Carlo over 96 configurations of the cell-size distribution, randomization inference on 5,000 within-cell permutations, a 300-draw simulation of the DHS displacement algorithm, a paired bootstrap of the difference between the aggregated and individual-level estimates, two placebo designs and an event study. Two results are exact rather than statistical and serve as reproduction checks: the aggregated and individual-level estimates coincide at −0.0096427 to the seventh decimal place, and at a zero distance cutoff the Conley estimator returns the cluster-robust standard error of 0.000212. A full pass takes about two hours and twenty-four minutes once the panels exist; a two-second smoke test validates the Mata library on its own. All code is released without restriction under the MIT licence. The underlying microdata are not redistributed: the DHS and ACLED terms of use forbid it. The README explains how to obtain them free of charge, and ACLED_extract_specification.md gives the exact export parameters so that the identical conflict extract can be regenerated.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.