Case Semantics & Transforms

Two related ideas: case semantics change which cases an analysis sees and how much each counts, and transforms create or reshape variables. Both live under the Data and Transform menus, and both are respected by every analysis automatically.

These three settings travel with every analysis, chart and transform you run. The status bar shows when any of them is active.

Select Cases

Data → Select Cases… (Ctrl+Alt+S). Restrict the analysis to a subset of cases, for example, only participants where Age >= 60, or a random sample. Unselected cases are filtered out of every subsequent statistic until you clear the selection. You can choose to filter (hide) or to mark unselected cases. The comparison value is checked against the variable's type: a numeric variable needs a number, and a date variable needs a calendar date in YYYY-MM-DD form (compact YYYYMMDD is accepted and canonicalized), compared chronologically, so visit_date >= 2024-01-15 selects that day and everything after it.

Weight Cases

Data → Weight Cases… (Ctrl+Alt+W). Give each case a frequency weight from a variable, so a case with weight 3 counts as three observations. Used for data stored as a frequency table. Counts, means and tests all honour the weight. For survey (sampling) weights use Survey Design instead. A frequency weight reproduces cases, which is the wrong model for a sampling weight and understates survey standard errors; the two settings are mutually exclusive.

Survey Design

Data → Survey Design… Declare the dataset's complex-survey structure: strata, clusters (PSUs) and sampling weights. The covered procedures then report design-based estimates with correct standard errors, and every other analysis says plainly that it ignored the design. The full story is on its own page: Complex Samples (Survey Designs). Under a declared design, Select Cases and Split File become domain estimation automatically (the design-correct way to analyze subgroups).

Split File

Data → Split File… (Ctrl+Alt+T). Repeat every analysis separately for each group of a categorical variable (e.g. run all your descriptives once per StudyZone). Output is produced per split group until you turn the split off.

Order of operations. Select narrows the cases, Weight scales their counts, and Split partitions the result into groups, and every analysis you run sees all three at once.

Transforms (Transform & Data menus)

Compute Variable

Transform → Compute Variable… (Ctrl+Shift+C). Create a new variable (or overwrite one) from an expression. The expression language is missing-aware and typed (number, string or date):

Arithmetic: + - * /, parentheses, MOD.
Numeric functions: SQRT, LN, LOG10, EXP, ABS, MIN, MAX, the trig family, ROUND(x[, digits]), and the SD / VARIANCE aggregates.
Logic: comparisons (< <= > >= = ~=), AND OR NOT, and IF(cond, then, else).
String: CONCAT, LENGTH, UPCASE, LOWCASE, LTRIM, RTRIM, SUBSTR, INDEX, REPLACE, STRING, NUMBER, and quoted literals.
Date (ISO YYYY-MM-DD): DATE.DMY, DATE.MDY, YEAR, MONTH, DAY, DATEDIFF.

The new variable takes the type of its expression: a string expression makes a string variable, a date expression a date variable.

Example. BMI_category = IF(BMI >= 30, "Obese", IF(BMI >= 25, "Overweight", "Normal")).

Recode

Transform → Recode… (Ctrl+Shift+R). Map a variable's values into new ones, into the same variable or a new one. Use it to collapse categories (e.g. five education levels into three) or to band a scale variable.

Grouping a scale variable vs. histogram bins

Two things look alike but do different jobs:

The histogram's bins slider groups a scale variable in the picture only (display, no data change). Use it to explore shape.
A Recode into Different Variables (or Bin Variable, below, for the common numeric-cutpoints case) makes a real categorical variable (BMI → Underweight/Normal/Overweight/Obese) that crosstabs, boxplots and ANOVA can use, and that is saved with the document. Use it when the groups are part of the analysis.

The tutorial workbook's transforms module walks this exact recode end to end.

Rank Cases

Transform → Rank Cases… (Ctrl+Shift+K). Create a new variable holding each case's rank on a chosen variable, with the usual tie-handling options.

Bin Variable

Transform → Bin Variable… (Ctrl+Shift+N). The purpose-built front end for grouping a scale variable into a new ordinal one. Recode works for this too, but means typing every cutpoint and label by hand. Pick a variable and a method:

Equal width: N groups of equal size along the value range (e.g. four 10-point-wide BMI bands).
Equal count (quantiles): N groups with (about) the same number of cases in each, e.g. splitting into quartiles.
Custom: your own ascending cutpoints (e.g. 18.5, 25, 30 for the standard BMI categories).

A live preview shows the resolved boundaries as cut lines over a histogram of the variable, and the generated group labels (< 18.5, 18.5 – 25, …), before you commit. The new variable is numeric with Ordinal measure and value labels already set: nothing left to finish in Variable View. Cutpoints are resolved against the currently active Select Cases/Weight Cases, like every analysis.

Sort Cases (Ctrl+Shift+D): reorder rows by one or more variables.
Transpose (Ctrl+Alt+P): swap rows and columns.
Merge Files (Ctrl+Alt+M): add cases or add variables from another dataset, matched by a key (or by row position). A key that is numeric in both datasets matches by value, so 1 in one file meets 1.0 in the other; any other key matches its exact text. Each case takes the first row with its key. The status line says how many cases matched no row and how many of the file's rows matched no case (an unknown key, a blank key, a repeated key, or rows past the end of a positional join), so a join that misses is never silent.
Define Variable Properties (Ctrl+Alt+D), Bulk Rename and Variable Sets: covered in Working with Data.

Reproducibility note

Every transform on this page is journalled: it appears in the Syntax console as a command, so a data-preparation pipeline can be saved and replayed on fresh data exactly. That includes the case-semantics settings (Select Cases, Weight Cases, Split File, as select_cases, weight_cases, split_file), Recode (its ordered rules ride the recode command as one rule string each, in order), Transpose, and both Merge Files modes (merge_add_cases, merge_add_variables; the journaled command records the file's path and the SHA-256 of its bytes, so a replay proves it merged the same file). Bin Variable's journaled command always records the resolved cutpoints, even for the Equal width/Equal count methods, so replaying it later reproduces the same groups rather than silently re-deriving different ones from changed data. A failed or cancelled dialog writes no journal line.

Three operations remain outside the journal: Bulk Rename, Import Data Dictionary and Define Variable Properties. If a session used one, the saved syntax and any exported reproducibility bundle say so explicitly (a # not replayable: note), rather than implying the journal reconstructs the whole session.