How to weight data in survey sampling

Market Research

Survey sample groups should represent the whole population as accurately as possible. Sometimes, you may end up with an imbalanced sample group, such as one including too many men or not enough young people. When this happens, you can weight the data to ensure you have a representative sample.

Methods of weighting data

Depending on your needs, there are several ways to weight data:


With raking, a researcher adjusts the weights for a set of variables with a known population distribution until the sample is balanced. The researcher will start by adjusting one variable and then adjust subsequent variables until they are correctly proportioned. If an adjustment for one variable affects another too much, more adjustments are performed until a balance is achieved.


Matching involves taking a sample of cases like survey interviews that represent the population and using it as a template for what a randomly selected survey sample would look like. Each case in this target sample is then paired with the most similar case from your sample. After the closest matches have been found for all the target sample cases, any unmatched cases from your survey are discarded.

Ideally, you’ll end up with a set that closely matches the target population. However, there is the chance that some cases in the target sample won’t match your survey data. Keep in mind that the larger your starting sample, the more potential matches you’ll have with the target sample.

Propensity weighting

With propensity weighting, survey respondents are weighted by the inverse of their probability of being randomly selected. This method is designed to ensure that the target population is represented in accurate proportions. As such, propensity weighting allows the researcher to use all collected responses. However, the large number of respondents can lead to highly variable weights and larger margins of error.

Matching and propensity weighting

In some cases, weighting survey data may require two weighting methods. With matching followed by propensity weighting, matched cases are combined with the cases in the target sample. All of these cases are then fit to the propensity weighting method to create weights for the matched cases.

Matching and raking

When you use matching and raking, only the matched cases are raked, allowing for a more precise sampling weight calculation.

Propensity weighting and raking

If you start with propensity weighting, those weights can then be trimmed and fed into your raking procedure.

Matching, propensity weighting and ranking

When matching and propensity weighting is followed by raking, the propensity weights of the matched cases are used as a starting point for the raking process.

Weighted T-Test

If you want to compare two groups of continuous data, a weighted T-test is an ideal choice. This statistical calculation adjusts means and standard deviations to generate weighted values based on the proper representation.

Your source for survey data solutions

Contact us for more information about how our solutions can help you weight survey questions and reveal actionable insights.