Work out weighting of survey data instantly with clear inputs, formula shown and shareable results.
Post-stratification weighting scales each group's responses by population share over sample share. A group forming 22% of the population but only 14% of the sample gets a weight of 1.571, so its 140 responses count as 220. Weights far from one inflate variance, which is why they are usually trimmed at about 3.
Design weight
weight = population share / sample share
Weighted contribution
effective = respondents x weight
Because a weight of 8 means a handful of respondents drive a whole subgroup estimate, inflating variance and instability far more than it corrects bias.
Only the part correlated with the weighting variables. If non-respondents differ on unmeasured characteristics, weighting cannot recover it.