Statistics Glossary → Frequency weight (summary-table data)
Frequency weight (summary-table data)concept
A weight stating how many identical observations a data row stands for; used when data are entered as a count table rather than as individual records.
Cross-tabulations and count data are often stored in summary form, with each row a cell: for example 'Female, Yes, 34'. A frequency weight makes that row count as 34 separate observations, so the analysis gives exactly the same result as if the 34 rows had been entered one by one. The sample size is the sum of the weights.
A frequency weight is different from a sampling weight (how many people in the population an observation represents) and from an analytic weight (weighting by measurement precision); the same numbers entered as one of these other kinds give very different standard errors and p values. Frequency weights are by nature non-negative whole numbers; fractional or negative values usually signal a data-preparation error.
An everyday example
Instead of sending 1000 individual answers, a polling company shares a four-row table: female-yes 320, female-no 180, male-yes 260, male-no 240. Once the count column is declared a frequency weight, the chi-square test gives the same result as it would with the 1000-row raw data.
In MerQur
The 'Weight (frequency) column' in Chi-Square Independence and the Crosstab, the 'Weight / frequency column' in Chi-Square Goodness of Fit, the 'Frequency (count) column' in the Log-Linear Model, and the 'Frequency weight' option in Poisson / Negative Binomial, Probit, Logistic, Multinomial Logistic and Ordinal Regression.
Analyses where this concept appears
Chi-Square IndependenceChi-Square Goodness of FitLog-Linear ModelCrosstabLogistic RegressionPoisson / Negative Binomial Regression (Count Regression)Multinomial Logistic RegressionOrdinal Logistic RegressionProbit Regression