Statistics Glossary → Sampling weight (design weight)
Sampling weight (design weight)concept
In complex samples, a weight showing how many population units each observation represents, usually defined as the inverse of its probability of selection.
If a unit's probability of selection is π_i, its design weight is w_i = 1 / π_i: a person selected with probability 1/500 represents 500 people in the population. Many surveys deliberately oversample some groups (e.g. small regions); without weights these groups would contribute too much to the results. The final weight is usually the design weight multiplied by adjustments for nonresponse and calibration (matching known population totals).
Weighted estimates reduce bias, but the more unequal the weights, the larger the variance; by Kish's approximation, the design effect due to unequal weights alone is about 1 + CV² (CV: the coefficient of variation of the weights). A sampling weight should not be confused with a frequency weight, which records how many times the same value occurs and leads to different standard errors. For correct standard errors, the stratum and primary sampling unit information must be used together with the weight.
An everyday example
A national survey selects 1,000 people from a sparsely populated region and 1,000 from a densely populated one. In the small region each respondent represents 200 people, in the large region 5,000. If the weights are ignored when computing the national average, the small region's views count 25 times more than they should.
In MerQur
The 'Weight' field in Weighted Mean, Weighted Frequency, Weighted Total, Weighted Linear and Logistic Regression and Survey-PHREG (the inverse of the selection probability; where available, the final weight adjusted for nonresponse or calibration).
Analyses where this concept appears
Weighted Mean (Survey Means)Weighted Frequency (Survey Freq)Weighted Total (Survey Total)Weighted Linear Regression (Survey Regression)Weighted Logistic Regression (Survey Logistic Regression)Survey-PHREG (Weighted Cox)
Related concepts
Stratified sampling (design stratum)Primary sampling unit (PSU) and cluster samplingDesign effect (DEFF)