Weights

Raking (also called rim weighting) creates a new dataset version. SPSS Weight Cases mutates the open file; this does not.

Edit on GitHub

Data → Data preparation → Rake Weights. Submit is Create raked version. Raking is also called rim weighting.

Raking creates a new dataset version. It does not overwrite the open file (SPSS Weight Cases mutates the open file; this does not).

Targets

Fill margins in the dialog, or attach a CSV. The file needs columns for variable, category, and proportion or count. Accepted headers:

  • Variable: variable, var, column, margin
  • Category: category, level, label
  • Value: proportion, percent, percentage, count, population, target, n, freq, frequency

Excel targets must be saved as CSV, or attached in chat.

Missing or untargeted categories

Rows whose weighting category is blank or outside the targets are not raked silently. There are three choices:

ChoiceWhat happens
ExcludeThose rows stay in the file with weight 0, and are counted in the weighting summary.
ImputeThose rows are filled to a named category that is already in the targets. In chat: impute gender to Female.
Missing targetThose rows are coded as missing and that level is added to the targets. In chat: ask for a missing target.

The Rake Weights dialog always excludes. Its warning is: rows with a missing or untargeted category are excluded and given weight 0; they stay in the dataset and are counted in the weighting summary; to impute them or add a “missing” target, say so in chat.

If chat does not choose, Tensr asks before weighting: exclude them (weight 0), impute them to a category, or add a “missing” target.

On a banner

The weight picker on Custom Tables:

  • Unweighted (original file)
  • This file's weight when the open version is a raked version
  • Raked weight (id…) for other raked siblings

Set weight chooses an active weight column, or No active weight.

Crosstabs, means, t-tests, chi-square, and regression use the active weight. Other analyses refuse while a weight is active.

Cells show n (unweighted), wn (weighted), and ESS (Kish effective sample size). Nested banners use that same weight — the nest itself is on Banner table.