Survey weighting, defined

Summary:

Survey weighting is a statistical adjustment applied to collected response data so the sample looks more like the population it's meant to represent. Instead of treating every response equally, weighting assigns each respondent a numeric value based on how over- or under-represented their demographic group is compared to the target population.

Weighting happens after data collection, which is why it's sometimes called post-stratification. It's a correction, not a substitute for good sampling. A well-designed sample still needs weighting sometimes, because response rates rarely match a study's original targets exactly.

Survey weighting matters because an unweighted sample can quietly mislead a team into confident, wrong conclusions. If younger respondents answer surveys less often than older ones, an unweighted report will overstate whatever older respondents believe, even if the underlying targeting was sound.

This shows up constantly in national and brand-tracking research, where a sample that skews toward one region, income bracket, or education level can shift topline numbers by several points. Left uncorrected, that skew doesn't just blur a single result. It can create a false uptick or dip in a tracking study, making a brand's numbers look like they moved when the sample composition simply shifted between waves.

Two methods account for most real-world weighting:

Cell-based weighting applies when a researcher knows exactly how many respondents each demographic combination should represent. If a study needed 80 men aged 18 to 24 but collected 100, each of those 100 responses gets a weight of 0.8 to bring that cell back in line.

Raking, also called iterative proportional fitting, handles cases where a study needs to balance several variables (age, gender, region) without knowing exactly how those variables interlock in the population.

Raking adjusts the sample on one variable at a time, cycling through each variable repeatedly until every one is reasonably balanced. It's the standard method used by Pew Research Center and most public opinion pollsters.

Less common approaches include propensity weighting, which corrects for the likelihood that a certain type of person chose to respond at all, and matching, which pulls a representative comparison set from an external dataset.

A general rule of thumb: weight on as few variables as possible, and avoid letting any single respondent's weight fall below 0.5 or climb above 2.0. Weights outside that range tend to distort results more than they correct them.

A consumer brand runs a national tracking survey and finds their sample skews toward college-educated respondents in the Northeast. Before reporting results, the team applies raking weights on region and education using the latest Census figures, bringing the sample back in line with the true US population before comparing this wave to the last.

For a full worked example, including how to choose weighting variables, set population targets, and trim extreme weights step by step, see SurveyMonkey's complete guide to weighting surveys for market research.

Weighting corrects for imbalance after the fact, but reaching a well-targeted audience from the start reduces how much correction you'll need. SurveyMonkey Audience gives researchers a global panel with built-in demographic targeting, so studies start closer to representative. Explore market research templates built to pair with weighted analysis once your results are in.