What is Raking?
Raking (also called rim weighting or iterative proportional fitting) is a survey-weighting method that adjusts sample responses so their marginal distributions match known population totals (for example by age, gender, region). It’s used to reduce bias when the people who answered a survey don’t perfectly reflect the population you care about.
Raking is an iterative process that reweights survey respondents so the sample’s margins for selected variables (such as age groups, sex, geography or language) align with external population benchmarks (for example census or administrative data). You choose a set of control margins, give each respondent an initial weight, and then adjust weights repeatedly across the margins until the sample proportions match the target proportions within a tolerance. The method is flexible because it works with many variables without requiring every combination of categories to appear in the sample.
Usage example
A city council runs an online survey about local services but finds respondents are disproportionately older and live in a few neighbourhoods. Using raking, the analyst adjusts respondent weights so the survey matches the city’s known age, gender and district distributions. After raking, the survey estimates better reflect the whole community rather than just the people who were most likely to respond.
Practical application
Raking matters because it helps organisations make survey results more representative without having to collect more data. For Hearo customers, raking can be used alongside multilingual outreach to correct for remaining imbalances (for example if some language communities responded at lower rates). It improves the accuracy of aggregated estimates used for decisions and reporting, but it’s not a cure-all: raking depends on good population benchmarks, can inflate variance when weights vary a lot, and may require trimming or combining categories to avoid extreme weights. Always report that weighting was applied and consider design-effect or effective sample size when interpreting results.
FAQ
What information do I need to perform raking?
You need reliable external benchmarks (marginal distributions) for each variable you’ll rake to — for example census totals for age bands, sex, region or language. You also need the same variables collected in your survey so respondents can be classified into the matching categories.
How many variables can I include when raking?
Raking can handle several variables because it adjusts margins iteratively rather than requiring every cross-classified cell to exist. However, adding many variables increases the chance of small or empty cells, which can produce extreme weights. If that happens, combine categories, use fewer margins, or apply weight trimming.
Does raking change individual responses?
No — raking does not alter what respondents answered. It assigns weights to responses so aggregated totals and proportions reflect the target population. Individual records remain the same; only their contribution to estimates changes.
Can raking fix a biased sample entirely?
Raking helps correct bias for the variables you include and to the extent your population benchmarks are accurate. It can’t correct for unmeasured sources of bias (for example systematic differences in opinions within categories you didn’t measure) and can increase estimate variance if weights are very uneven. Raking is a useful tool, but good survey design and broad outreach remain important.