Transforms

Entry: project workspace → Transforms, or the Data menu item "Transform…". Transforms act directly on the current dataset.

Commands

Command	Purpose
COMPUTE	new variable from an expression (+ - * /, abs, sqrt, ln, exp, log10, sin, cos, tan, round, trunc)
RECODE	map old values to new ones
AUTORECODE	encode values (numeric or string) into 1..k codes with value labels
COUNT	count matching cells per case
RANK	rank-derived variables: rank, percentile, normal score, n-tiles
SAMPLE	draw cases randomly (by fraction, or exactly "n FROM N")
SORT CASES	sort by one or more keys
SELECT IF	filter cases (> < == != && `
AGGREGATE	grouped means
FLIP	transpose
COMBINE FILES	append cases (ADD FILES) or match by key (MATCH FILES) from another dataset

RANK: rank-derived variables

RANK VARIABLES=score (A) /RANK /TIES=MEAN.
RANK VARIABLES=score (D) /NTILES(4) /INTO=quartile.

• (A) / (D) after a variable selects ascending / descending order, ascending by default;

• output functions may be combined:

  Option	Meaning
  /RANK	rank
  /N	number of cases in the tie group
  /RFRACTION	rank / number of valid cases
  /PROPORTION	Blom proportion (r − 3/8) / (w + 1/4)
  /PERCENT	percentile r · 100 / w
  /NORMAL	normal score Φ⁻¹ of the proportion
  /NTILES(k)	n-tile group 1..k

• /TIES=MEAN|LOW|HIGH|CONDENSE decides the rank of tied values (average / lowest / highest / condensed); the default MEAN is the familiar average rank;

• /INTO=name names the target variable; the default derives it from the function (R_ / N_ / RF_ / PR_ / P_ / Z_ / NT_ prefix);

• missing cases (including user-missing values) do not participate and get an empty result.

AUTORECODE: one code for strings and numbers alike

AUTORECODE VARIABLES=city /INTO=city_n.
AUTORECODE VARIABLES=a b c /INTO=x y z /GROUP /DESCENDING.

• each source variable has its distinct non-blank values numbered 1..k in ascending order (/DESCENDING reverses it), with value labels filled in automatically (an existing value label wins over the displayed text);
• user-missing values are still encoded, but always come last (they take the highest codes) and are marked missing on the target variable; with more than three such codes the target switches to a missing range, as PSPP does;
• /GROUP shares one coding table across all source variables (they must have the same type);
• target variables are integers on a nominal scale and inherit the variable label.

SAMPLE: drawing cases

SAMPLE 0.5.            * keep each case with probability 50% (count varies)
SAMPLE 5 FROM 10.      * keep exactly 5 of the first 10 cases
SAMPLE 0.5 /SEED=42.   * fix the seed so a run is reproducible

MRSETS: multiple response sets

MRSETS /MDGROUP NAME=$brands LABEL='Brands bought' VARIABLES=q1a q1b q1c VALUE=1.
MRSETS /MCGROUP NAME=$colors VARIABLES=c1 c2 c3.
MRSETS /DISPLAY NAME=ALL.
MRSETS /FREQUENCIES NAME=$brands.
MRSETS /CROSSTABS NAME=$brands BY=gender.

• MDGROUP (multiple dichotomy): VALUE= decides whether a variable counts as "selected";
• MCGROUP (multiple category): every value of the listed variables is itself a response;
• once defined, /DISPLAY lists the sets, /FREQUENCIES reports response counts and /CROSSTABS ... BY=group crosstabulates responses by a group (PSPP builds these tables with CTABLES; the two extensions here are app-specific);
• response tables report both denominators, responses and cases with a response; counts are unweighted (see missing values for the weighting rules);
• /DELETE NAME=$m|ALL removes a set; redefining a name overwrites it (as PSPP dict_add_mrset does).

COMBINE FILES: merging datasets

ADD FILES /FILE=* /FILE=other /IN=source.
MATCH FILES /FILE=* /TABLE=lookup /BY id.

• ADD FILES appends the cases of the second dataset (variables are aligned by name, missing ones are filled with blanks);
• MATCH FILES matches the second dataset in by the /BY key with a hash join (input need not be sorted);
• /IN=var adds a 0/1 source indicator; /KEEP, /DROP and /RENAME=(old=new) adjust the variables;
• dictionaries are merged with PSPP's merge_dictionaries rules: the first file that defines a variable wins, later files only fill gaps, string width takes the maximum and a type conflict is an error; with MATCH FILES a duplicate key keeps the first case;
• "Attach dataset" in the syntax editor adds another .sav or project file, and * stands for the current dataset; the Data menu also offers a dialog-driven "Merge datasets…".

Syntax editor

A syntax subset runner is available from the Data menu:

DESCRIPTIVES pre post.
FREQUENCIES gender method.
T-TEST /TESTVAL=60 /VARIABLES=post.
CORRELATIONS /VARIABLES=pre post.
COMPUTE gain = post - pre.
RECODE gender (1=0) (2=1) INTO sex.
AUTORECODE VARIABLES=city /INTO=city_n.
RANK VARIABLES=post (A) /RANK /NTILES(4).
SAMPLE 0.5 /SEED=20261016.
MRSETS /MDGROUP NAME=$m VARIABLES=a b VALUE=1.
ADD FILES /FILE=* /FILE=other.
MATCH FILES /FILE=* /TABLE=lookup /BY id.
SORT CASES BY post DESCENDING.
SELECT IF post >= 70.
LIST.
HELP.

The template bar above the editor inserts a command; the output pane reports each command. A whole line starting with * is a comment.

Comment semantics, macros, INCLUDE, error locations and the .sps round-trip are covered in Syntax language.

Other data operations

The Data menu also offers:

• Weight cases (WEIGHT CASES)
• Split file (SPLIT FILE)
• Find cases
• CSV import / export