Survey Weighting: Post-Stratification, Raking (IPF) & Design Effects

What you'll build

What you will be able to do

Each lab ends with a concrete result that you build yourself and the checker confirms, cell by cell.

• Compare the 1,200 interviews of a transit survey with the census margins for age, gender and region, measure every gap and find the group that needs the biggest weight.
• Build expansion weights and relative weights for nine age × region cells, check that the weighted sample still adds up to 1,200 interviews and find the cells with risky large weights.
• Rake the 3 × 3 table of interviews to the age margin and the region margin in turn, watch the row sums converge after two rounds and read off a raked weight.
• Cap raked weights at 0.5 and 2.5, rescale them so they add up to 1,200 interviews again and count how many interviews the trimming touched.
• Calculate weighted means and shares from respondent records with SUMPRODUCT, including a weighted share for one region, and compare them with the unweighted numbers.
• Compute Kish's effective sample size and design effect for the raked and the trimmed weights and see how many interviews trimming wins back.
• Calculate the textbook margin of error, widen it with the design effect of the weights, do the same for the small rural subgroup and report the upper end of the interval.
• Turn the weighted interviews into weighted shares, measure the absolute gap to the population before and after weighting and chart population, sample and weighted shares side by side.