Survey Weighting, Design Effect and Effective Sample Size Calculator
Calculate survey weights, Kish design effect, weighting efficiency and effective sample size from weights, cell targets or rim weighting (raking) on two variables.
inquiry@globalvoxpopuli.com · globalvoxpopuli.com
How to use the weighting calculator
- Cell targets: list your weighting cells (e.g. age × gender), the number of respondents in each, and each cell's share of the population. The calculator gives the weight for each cell.
- Rim weighting: paste a cross-tab of two variables (e.g. region by age) and the population targets for each. The calculator rakes the weights until both sets of targets are met.
- Existing weights: paste the weight column from your data file to check its efficiency.
- Optionally set trim limits to see how capping extreme weights improves precision.
Weighting efficiency and design effect explained
Weighting corrects a sample that does not match the population, but it has a price: respondents with large weights count for more, so random noise in their answers counts for more too. Kish's design effect measures that cost.
- Design effect (deff): how much weighting inflates the variance. 1.0 means no cost.
- Weighting efficiency: 1 ÷ deff. 80% means the weighted sample is worth 80% of its size.
- Effective sample size: n ÷ deff. Use it to calculate margins of error and significance tests on weighted data.
As a rough guide, efficiency above 80% is good, 60–80% is acceptable, and below 60% suggests the sample needs better quotas or the weighting scheme needs simplifying.
Cell weighting vs rim weighting
Cell weighting needs the population share of every combination, such as men aged 18–34 in the North. Those interlocking targets are often unavailable, and small cells create extreme weights. Rim weighting (raking) only needs each variable's own distribution. It adjusts weights in turn until the sample matches every margin. Most tracking studies use rim weighting on age, gender, region and one or two other variables.
Worked example
A 1,000-interview survey is weighted on region (3 groups, targets 40/35/25%) and age (3 groups, targets 25/35/40%). Rim weighting gives a design effect of about 1.32, so the 1,000 interviews are worth about 760. The few respondents who are both in the first region and the oldest age group carry weights of about 3. Press Example to load it and see how trimming at 2.5× and 0.3× the mean changes the efficiency.
Frequently asked questions
What is a good weighting efficiency?
Above 80% is good. Between 60% and 80% is common for online samples weighted on several variables. Below 60%, look at the weights driving the loss.
Should I trim weights?
Trimming extreme weights, often at 3 to 5 times the mean, usually improves precision with little bias. Report that weights were trimmed and check that key results don't move much.
How many variables can I rim-weight on?
As many as the sample supports, but each one lowers efficiency. Three to five variables with collapsed categories is typical.
Do I use the effective sample size for significance testing?
Yes. Divide each base by the design effect before using the significance calculator or margin of error calculator.
Need the respondents, not just the numbers?
Global Vox Populi runs quantitative and qualitative fieldwork in 170+ countries through its own proprietary panels of consumers, B2B and IT decision-makers, physicians, nurses, patients, caregivers and payers. ISO 9001, ISO 20252 and ISO/IEC 27001 certified and HIPAA compliant, following ESOMAR and Insights Association guidelines.
Check feasibilityLink to this tool
Free to use in proposals, courses and articles. If it helped, a link back is appreciated:
<a href="https://globalvoxpopuli.com/tools/weighting-efficiency-calculator/">Weighting Efficiency Calculator</a> by Global Vox Populi