SPSS command: SPSS SELECTPRED
Predictor screening and ranking (SELECTPRED) in your browser
Predictor selection narrows a long list of candidate variables before modelling.
Free · No account · Files stay on your device · Updated 18 October 2026
What it tells you
Crosstabs runs SPSS SELECTPRED on your device: it first screens out variables that are mostly missing, nearly constant or have too many categories, then ranks every remaining predictor by how well it alone predicts a categorical or scale target, and marks the top of the list as selected.
- Which variables are unusable and why: all missing, constant, too many missing values, one category holding too many cases, too little variation or too many categories.
- How strongly each remaining predictor is related to the target on its own, with the statistic, df, significance and N used.
- A ranked shortlist of predictors to carry into a model such as logistic regression, a tree or a neural network.
When to use it
Use it when
- Hundreds of survey or CRM variables and you need a defensible shortlist before modelling.
- A quick data-quality pass that flags near-empty and near-constant variables.
- Ranking many categorical predictors of a categorical target by Cramér's V or lambda.
Use something else when
- You need each predictor's effect adjusted for the others: ranking is one variable at a time; use a regression or automatic linear modeling.
- Predictors that matter only in combination (interactions) can rank low here; check them in a multivariable model.
- You want a final model: the list is a screening step, not a model.
- Case weights must be applied: the SELECTPRED algorithm defines none, so results are unweighted.
How to run it in Crosstabs
- 01
Open your data
Open the file in the workspace and switch to Statistics mode.
- 02
Choose Predictor selection
In the Machine learning group, pick Predictor selection.
- 03
Set target and candidates
Choose the Target measurement (Categorical (classification) or Scale (prediction)) and the Target, then tick Categorical predictors and Continuous predictors.
- 04
Set ranking and screening
Under Ranking and screening choose the Criterion for categorical predictors (Pearson chi-square p-value, Likelihood-ratio chi-square p-value, Cramér's V or Lambda), the Number of predictors to select (blank = automatic), and the thresholds: Maximum percent missing (70), Maximum percent in a single category (95), Maximum categories as percent of cases (90) and Minimum coefficient of variation (0.001).
- 05
Run
Click Run. Cases with a missing target, or with every predictor missing, are removed first.
Worked example
Computed by the Crosstabs engine from the synthetic data below when this page was built — not typed in.
Data
240 synthetic web leads with a Yes/No conversion target. Lead source, pages viewed and minutes on site drive conversion; device and age are noise; country is “home” for about 97% of leads; income was collected for only about a quarter of leads. Default screening thresholds, Pearson chi-square criterion, Number of predictors to select set to 3. Fixed seed.
| Rank | Predictor | Statistic | Value | df | Sig. | N | Selected |
|---|---|---|---|---|---|---|---|
| 1 | pages_viewed | F | 87.8429 | 1, 238 | <.001 | 240 | Yes |
| 2 | source | Pearson chi-square | 29.2585 | 3 | <.001 | 240 | Yes |
| 3 | minutes_on_site | F | 15.9210 | 1, 238 | <.001 | 240 | Yes |
| 4 | device | Pearson chi-square | 0.2390 | 1 | .625 | 240 | No |
| 5 | age | F | 0.1121 | 1, 238 | .738 | 240 | No |
Each predictor is judged alone on cases where it and the target are both present. Number of predictors to select: 3.
| Predictor | Reason |
|---|---|
| country | One category holds too many cases |
| income | Too many missing values |
How to read it
Screening removed country (one category holds too many cases) and income (too many missing values) before any test was run. Because the remaining predictors mix categorical and continuous types, they are ranked by p-value: categorical ones by the Pearson chi-square, continuous ones by a one-way ANOVA F across the target classes. The top 3 are pages_viewed, source, minutes_on_site; these are exactly the three predictors that drive conversion in this file, while device and age rank below them. Each test looks at one predictor alone, so the list is a screening step for a model, not a model itself.
Output tables explained
- Predictor Ranking
- Rank, predictor, statistic (Pearson or likelihood-ratio chi-square, Cramér's V, lambda, F or r), its value, df, significance, the N used and whether it is selected, followed by a −log10(Sig.) chart.
- Screened Predictors
- Each variable removed before ranking with the reason, such as too many missing values or one category holding too many cases.
Assumptions and notes
- Each predictor is judged alone on the cases where it and the target are both present (pairwise); ties follow SPSS's secondary keys, then file order.
- Categorical target: categorical predictors use the chosen criterion and continuous ones a one-way ANOVA F. Scale target: categorical predictors use F and continuous ones the Pearson correlation.
- When categorical and continuous predictors are mixed, only p-values are comparable, so Cramér's V and lambda fall back to the Pearson chi-square p-value.
- With the number left blank, the list length follows SPSS's rule round(min(max(30, 2√L0), L0)) for L0 ranked predictors.
How the results are validated
SELECTPRED is checked against SciPy on a 300-case file with missing values in seven configurations — the four categorical criteria, continuous predictors of a categorical target, mixed predictors and a scale target: scipy.stats.chi2_contingency (Pearson and log-likelihood, no continuity correction), f_oneway and pearsonr, with Cramér's V and Goodman-Kruskal lambda written out and the chapter's ranking keys. The ranking order and the number of removed cases are identical, statistics agree to 10 decimal places and p values to 1e-12; the automatic list-length rule is checked separately. Validation evidence
Questions
How many predictors are selected?
The number you type in Number of predictors to select, or, if blank, SPSS's automatic rule round(min(max(30, 2√L0), L0)) — so with fewer than 30 ranked predictors all of them are marked selected.
Which criterion should I use for categorical predictors?
The Pearson chi-square p-value is the default and works for mixed predictor types. Cramér's V ranks by effect size, which is fairer when sample sizes differ across predictors; lambda ranks by the reduction in prediction error. V and lambda apply only when all predictors are categorical.
Why was a variable screened out?
The Screened Predictors table gives the reason: more than the maximum percent missing, one category above the maximum percent, more categories than the allowed share of cases, a constant value, or a coefficient of variation below the minimum. Relax the threshold if the variable should stay.
Does a high rank mean the predictor is causal or will stay important in a model?
No. Each test looks at one predictor alone, so two highly correlated predictors can both rank high while only one adds information in a model. Use the list as a shortlist and let a multivariable model decide.
Is my data uploaded?
No. Screening and ranking run in your browser and the file stays on your device.
Related procedures
- Automatic Linear ModelingLINEAR
- Naive BayesNAIVEBAYES
- Binary Logistic RegressionLOGISTIC REGRESSION
- Decision Tree (CHAID, CRT)TREE
- Crosstabs StatisticsCROSSTABS /STATISTICS
- Bivariate CorrelationsCORRELATIONS
- Free SPSS alternative: full coverage list
Run Predictor Selection on your own data
Open a .sav, CSV or Excel file. Everything runs in your browser; the file never leaves your device.
SPSS is a trademark of IBM Corp. Crosstabs is not affiliated with or endorsed by IBM.