SPSS command: SPSS LINEAR

Automatic linear modeling (LINEAR) in your browser

Q: Which entry criterion should I use?

AICC, the default, balances fit against model size and suits most data. F statistics mimic classic stepwise regression with p-to-enter and p-to-remove. ASE holds out a random share of records and picks the model that predicts them best, which guards most directly against overfitting.

Q: Forward stepwise or best subsets?

Best subsets compares every combination (up to 20 effects) and can find a better model that stepwise misses, at more computation; it lists the ten best. Forward stepwise is faster and shows the order in which effects enter.

Q: What does automatic data preparation change?

It can recast measurement levels, trim outliers beyond 3 robust standard deviations, replace missing predictor values with the mean, median or mode, and merge categories with similar target means. Every action is listed per predictor; untick Automatically prepare data to fit the raw values.

Q: How is predictor importance defined?

For each effect in the final model, the increase in residual sum of squares when it is dropped (its Type III sum of squares), divided by the total over all effects so the importances sum to 1.

Q: Are the p-values after selection trustworthy?

They are computed as if the model had been fixed in advance, so they are optimistic. Treat the selected model as a hypothesis and confirm it on new data or with the overfit-prevention criterion.

Q: Is my data uploaded?

No. Preparation, selection and fitting run in a Web Worker in your browser.

Automatic linear modeling fits a linear regression after cleaning the predictors and choosing which ones to keep.

Run Automatic Linear Modeling in your browser All SPSS procedures

Free · No account · Files stay on your device · Updated 18 October 2026

Worked example result

Coefficients · wait_time

t = −9.405

p <.001

Model Term	Coefficient	Std. Error	t
Intercept	88.0881	2.0820	42.308
wait_time	−1.1696	0.1244	−9.405
staff_rating=1, 2	−8.7957	1.3008	−6.762
staff_rating=3, 4	0.0000	—	—
channel=phone	−12.8620	1.5927	−8.076

Coefficients

See the full worked example

What it tells you

Crosstabs runs SPSS LINEAR on your device: optional automatic data preparation (recasting, outlier trimming, missing-value replacement, merging of categories), forward stepwise or best-subsets selection by AICC, F, adjusted R² or an overfit-prevention error, and an effects table with predictor importance, coefficients and outliers.

Which predictors earn a place in the model, step by step or as the best subsets, and how the selection criterion changes along the way.
How much of the target's variance the selected model explains (adjusted R², shown as accuracy) and its AICC.
How important each effect is — its Type III sum of squares as a share of the total — plus coefficients with confidence intervals.
What data preparation changed: which predictors were recast, trimmed, filled in or had categories merged, and which records are influential outliers.

When to use it

Use it when

A scale outcome with many candidate predictors and messy data — outliers, scattered missing values, categorical predictors with many sparse levels.
A quick, defensible first model to see which drivers matter before a hand-built regression.
You want selection by an information criterion or a held-out error rather than p-values alone.

Use something else when

The model is fixed in advance by theory: use linear regression or GLM Univariate and report every planned term.
A binary, count or ordinal target: LINEAR needs a scale target; use logistic, generalized linear or ordinal regression.
You need boosting or bagging ensembles: use the separate Ensembles of linear models procedure.
Survey weights: sampling weights are not run; use the complex-samples procedures for design-based inference.

How to run it in Crosstabs

01
Open your data
Open the file in the workspace and switch to Statistics mode.
02
Choose Automatic linear modeling
In the Regression group, pick Automatic linear modeling.
03
Set target and predictors
Choose the Target, tick Continuous predictors and Categorical predictors (a categorical variable declared ordinal is treated as ordinal), and keep Automatically prepare data ticked.
04
Choose the selection
Under Model selection pick the Model selection method (Forward stepwise, Best subsets or Include all predictors) and the Criteria for entry/removal: Information criterion (AICC), F statistics (forward stepwise only, with the entry and removal p-values), Adjusted R-square or Overfit prevention criterion (ASE) with its Overfit prevention set (%) and Random seed.
05
Declare any weight and run
Set the Confidence level (%) for the coefficients. If the dataset has a weight, declare it as a frequency weight (rounded to whole cases) or a sampling weight (not run), then click Run.

Run Automatic Linear Modeling in your browser

Worked example

Computed by the Crosstabs engine from the synthetic data below when this page was built — not typed in.

Data

150 synthetic service visits. Satisfaction (0–100) falls with wait time, rises with a 1–4 staff rating and shifts by channel, with app and web built to have the same effect; age is pure noise. Two wait times were mistyped as 150 and 180 minutes and six ages are missing. Settings: Automatically prepare data on, Forward stepwise with the Information criterion (AICC). Fixed seed.

Automatic Data Preparation

Predictor	Measurement	Action taken
wait_time	Continuous	Trimmed 2 outlier(s) to 3 robust SDs
staff_rating	Ordinal	Recast from continuous to ordinal (fewer than 5 values); Merged categories to maximize association with the target: {1, 2} {3, 4}
age	Continuous	Replaced 6 missing value(s) with the mean
channel	Nominal	Merged categories to maximize association with the target: {phone} {store} {app, web}

Model Building Summary

Step	Action	Effect	Information criterion (AICC)
1	Entered	wait_time	705.8761
2	Entered	channel	662.0138
3	Entered	staff_rating	623.0419

The criterion value is for the model after each step.

Coefficients

Model Term	Coefficient	Std. Error	t	Sig.
Intercept	88.0881	2.0820	42.308	<.001
wait_time	−1.1696	0.1244	−9.405	<.001
staff_rating=1, 2	−8.7957	1.3008	−6.762	<.001
staff_rating=3, 4	0.0000	—	—	—
channel=phone	−12.8620	1.5927	−8.076	<.001
channel=store	−6.2033	1.5433	−4.020	<.001
channel=app, web	0.0000	—	—	—

A coefficient of 0 without a standard error marks a redundant parameter (the reference category).

How to read it

Forward stepwise by AICC entered wait_time, channel, staff_rating and stopped at AICC = 623.04; the final model explains adjusted R² = 0.523 of satisfaction across 150 records. Age, the noise variable, was left out. wait_time has the largest importance (0.441). Data preparation trimmed 2 wait-time outlier(s) (two values were mistyped), filled 6 missing age(s) so those records stay in the model, and merged channel into 3 groups: {phone} {store} {app, web}.

Output tables explained

Model Summary: Target, whether data preparation was on, the selection method and criterion, AICC, accuracy (adjusted R²), R², the overfit-prevention error when used, and the records used and excluded.
Automatic Data Preparation: Per predictor: its measurement level after preparation and the action taken — recast, outliers trimmed to 3 robust SDs, missing values replaced by mean, median or mode, categories merged, or excluded as constant.
Model Building Summary: Forward stepwise: the effect entered or removed at each step with the criterion value (or F and its significance).
Best Subsets: Best subsets: the ten best models by the selection criterion, ties going to fewer effects.
Effects: ANOVA for the corrected model and each effect (Type III sum of squares, F, significance) with importance, and a predictor importance chart.
Coefficients: Coefficient, standard error, t, significance and confidence interval per model term; a zero without a standard error marks the reference category.
Outliers: Records with Cook's distance above 4/(N − p), at most 25, next to a predicted-by-observed chart; residuals can be downloaded as CSV.

Assumptions and notes

The usual linear-regression assumptions apply to the final model: linear effects, independent records, roughly constant residual variance.
Preparation statistics come from the training records: continuous predictors with fewer than 5 values become ordinal, ordinal ones with more than 10 categories become continuous, and categories are merged when an F test (alpha 0.05) cannot separate their target means.
Records with a missing target are excluded; without data preparation, so are records with any missing predictor.
Best subsets searches all combinations for up to 20 effects and switches to the documented hybrid with forward stepwise beyond that.

How the results are validated

LINEAR is checked against a NumPy hand-run of the SPSS LINEAR and Automated Data Preparation chapters that fits every candidate model with numpy.linalg.lstsq on explicit design matrices (SciPy for the distributions). Eleven cases cover forward stepwise by AICC, F, adjusted R² and ASE, best subsets, include-all with frequency weights, removal steps and the hybrid search beyond 20 effects. Preparation actions, selected effects, step sequences and subset rankings are identical; criteria agree to a relative 1e-9, R², adjusted R², AICC and ASE to 1e-10, coefficients, standard errors, t and confidence limits to 1e-8, p values to 1e-10, effect sums of squares and F to 1e-8, and leverage and Cook's distance to 1e-10. Validation evidence

Questions

Which entry criterion should I use?

AICC, the default, balances fit against model size and suits most data. F statistics mimic classic stepwise regression with p-to-enter and p-to-remove. ASE holds out a random share of records and picks the model that predicts them best, which guards most directly against overfitting.

Forward stepwise or best subsets?

Best subsets compares every combination (up to 20 effects) and can find a better model that stepwise misses, at more computation; it lists the ten best. Forward stepwise is faster and shows the order in which effects enter.

What does automatic data preparation change?

It can recast measurement levels, trim outliers beyond 3 robust standard deviations, replace missing predictor values with the mean, median or mode, and merge categories with similar target means. Every action is listed per predictor; untick Automatically prepare data to fit the raw values.

How is predictor importance defined?

For each effect in the final model, the increase in residual sum of squares when it is dropped (its Type III sum of squares), divided by the total over all effects so the importances sum to 1.

Are the p-values after selection trustworthy?

They are computed as if the model had been fixed in advance, so they are optimistic. Treat the selected model as a hypothesis and confirm it on new data or with the overfit-prevention criterion.

Is my data uploaded?

No. Preparation, selection and fitting run in a Web Worker in your browser.

Run Automatic Linear Modeling on your own data

Open a .sav, CSV or Excel file. Everything runs in your browser; the file never leaves your device.

Open the workspace

SPSS is a trademark of IBM Corp. Crosstabs is not affiliated with or endorsed by IBM.