Weighting is like a chiropractor for your sample. A small adjustment makes everything line up better. Try to fix a broken leg with it and you'll make things worse.

Key takeaways

  • Weighting corrects small, known imbalances against reliable population targets.
  • Big weights on a few respondents inflate variance and can swing results.
  • Track weighting efficiency. If it's low, fix the sample in field.
  • Always report effective sample size next to weighted results.

What weighting does

You compare your sample to known population totals (census, reliable reference studies) on a few variables, usually age, gender, region and sometimes education. Then you give each respondent a weight so the weighted sample matches. Cell weighting works for a couple of variables; raking (RIM weighting) handles several at once.

Good reasons to weight

Your quotas were close but not exact. Mobile dropout skewed a little young. Region fell a few points short in one wave. These are small, predictable gaps, and weighting handles them cleanly.

Warning signs

  • Weighting efficiency below roughly 70% means you're losing a lot of statistical power.
  • Maximum weights above 4 or 5 mean a handful of people are speaking for many.
  • Results that move a lot between weighted and unweighted runs mean the sample itself is off.

Picture 12 respondents over 65 each weighted to 6. If two of them are low-quality, your "senior" results are now noticeably wrong, and nobody will know from the chart.

Fix it in field instead

Good quotas beat heroic weighting. Set quotas on the variables you'll weight on, monitor fill daily, and blend sources if one is skewing the sample. This is where a provider with broad global sample coverage helps: when one source runs dry, there's another verified one to pull from rather than stretching weights.

Report it honestly

Say what you weighted to, what the targets were, and the effective sample size (n divided by the design effect). Readers can then judge how much confidence to put in a subgroup. This matters most on multi-market work; see sample size for multi-country surveys.

FAQ

What is a good weighting efficiency?

Above 80% is comfortable. Between 60% and 80% is workable with caution. Below 60% usually means the sample needs fixing, not just weighting.

Should every survey be weighted?

No. If the sample matches targets well, weighting adds little. It's most useful for nationally representative reads and trackers where small skews add up over time.