Survey weighting adjusts your dataset so the demographic distribution matches the population you're trying to represent. If your sample has 60% women but your target population is 51% women, weighting downweights the women's responses (or upweights men's) so the reported percentages reflect the population.

Done correctly, weighting recovers honest population estimates from imperfect samples. Done poorly, it amplifies noise from small subsamples and produces brittle conclusions.

When weighting is necessary

Online panels rarely match population demographics. Younger, more urban, more tech-comfortable.
B2B studies of "all enterprise customers" when sample skews toward power users.
Tracker studies where wave-to-wave demographic mix shifts.
Stratified sampling that intentionally over-sampled small segments.

When weighting is unnecessary

Convenience samples where there's no claim of population representation in the first place.
Within-respondent comparisons. If the same respondent rates 5 concepts, ranking those concepts doesn't need weighting.
Census-style enumerations where you've actually contacted everyone.

Three flavors of weighting

Cell weighting

The most precise. Define cells by combinations of demographics (e.g., "Female × 35–54 × West"), and weight each cell to its known population proportion.

Pros: exact match to population on the demographics you specify. Cons: cells get small. With 4 ages × 2 genders × 4 regions = 32 cells, a 1,000-respondent sample averages 31 per cell. A few cells will have 5 respondents and unstable weights.

Marginal weighting

Weights each demographic separately. Female = 51%, 35–54 age = 28%, West = 23%. Each respondent's weight is the product of their marginal weights, normalized.

Pros: stable even with small samples. Cons: doesn't guarantee accuracy on combinations (you'll match overall % female and overall % young, but not necessarily % young women).

Rim weighting (also called raking)

The standard middle path. Iteratively adjusts weights until each marginal target is hit, accepting whatever combination cells fall out. Most commercial cross-tab tools default to rim weighting because it's the best trade-off between precision and stability.

Pros: matches all marginals exactly, stable with small samples. Cons: combinations may still be slightly off.

Practical guidance

Cap weights at ~5×. A respondent weighted 8× because they're an under-represented demographic is now contributing as much as 8 typical respondents — which is a lot of leverage on a single answer. Capping weights and re-raking is standard.
Effective sample size drops with weighting. A 1,000-respondent study with heavy weighting may have an effective sample size of 700. Confidence intervals widen.
Document the weights. Stakeholders should be able to see the weighting scheme. Tabular Pro stores it in project metadata.
Don't weight on outcome variables. Only weight on demographics or known stable characteristics.

How Tabular Pro implements weighting

Cell, marginal, and rim weighting all native
Apply, audit, and switch weighting schemes without re-exporting
Weighted frequencies, weighted cross-tabs, weighted significance testing
Effective sample size displayed on every weighted output
Weight caps configurable per study
Full audit trail — every weighting decision logged

Survey Weighting — Cell, Marginal, and Rim