Weighting
Weighting is a statistical adjustment that gives some survey respondents more influence and others less, so the sample's makeup matches the population it represents on characteristics such as age, gender, region or education. It corrects imbalances without changing anyone's answers.
How it works
Samples drift from their populations because of how they were drawn, who could be reached and who chose to answer. Weighting gives each respondent a number. Members of under-represented groups count for more than 1, over-represented ones for less. Post-stratification sets each group's weight directly from population and sample shares, as in the formula. Raking handles several variables at once, adjusting the weights iteratively until the sample lines up with the population on each of them.
Weighting corrects the variables it uses and helps other estimates only as far as those variables relate to them. In Pew Research Center's test of weighting on online opt-in samples, choosing the right adjustment variables mattered more than choosing the method, and even the best procedure removed only about 30 percent of the bias.
Weight for a group = the group's share of the population / the group's share of the sample
A worked example
A survey of 1,000 customers has 400 under 35, although under-35s are 30 percent of the customer base. Each young respondent gets a weight of 0.30 / 0.40 = 0.75, and each older respondent 0.70 / 0.60 = 1.17. Interest in a mobile-only plan is 74 percent among younger respondents and 24 percent among older ones, so it falls from 44 to 39 percent after weighting.
Related
Automate this in Pulse
The screener answers each Pulse participant gives, such as age band and region, are the variables a weighting scheme needs to compare the sample with the population.