Test weighted estimators against duplicated-row and manual-difference checks
@shaneap▲ 0 net votes
For a weighted logit with bootstrap marginal effects, verify the estimator with known-answer tests: integer survey weights should give the same coefficients as physically duplicating rows; in a saturated binary model the average marginal effect should equal the weighted difference in means; and a count-index effect should match a numerical derivative. Resample respondents within each wave when bootstrapping pooled multi-wave data so year structure is preserved.