SPSS command: SPSS GENLOG

General loglinear and logit models for tables

Q: Poisson loglinear or multinomial logit?

Use Poisson when all variables play the same role and you study their associations. Use multinomial logit when one factor is the response; its parameters then read as log odds ratios, as in logistic regression, and the explanatory margins are fixed.

Q: Why are so many parameters zero?

With reference coding, every parameter involving the last category of a factor is redundant and set to zero. The non-zero parameters are contrasts with those last categories.

Q: How do GENLOG and HILOGLINEAR differ?

HILOGLINEAR searches for a hierarchical model with K-way tests and backward elimination. GENLOG fits one design you choose and gives parameter estimates and residuals for it, including logit models. A common workflow is to find the model with HILOGLINEAR and then describe it with GENLOG.

Q: Which cells should I look at first?

Sort by the absolute adjusted residual. Cells beyond about ±2 are where observed counts depart most from the model and often point to the interaction the design leaves out.

General loglinear analysis fits a chosen model to the cell counts of a contingency table and reports a parameter for every effect, so you can read the size and direction of associations — or, in the logit form, how explanatory factors shift the odds of a response category.

Try with example data Run on your own file

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

What it tells you

Crosstabs runs the SPSS GENLOG procedure on your device with Poisson loglinear and multinomial logit models, three design choices, goodness-of-fit tests and cell-level adjusted residuals.

Whether a given set of effects — main effects, all two-way associations or the saturated model — reproduces the table (likelihood-ratio and Pearson tests).
The estimate, standard error, Z test and confidence interval of every non-redundant parameter, with reference (last-category) coding.
In a multinomial logit model, log odds ratios that describe how each explanatory factor shifts the response.
Which individual cells the model misfits, via standardized, adjusted and deviance residuals.

When to use it

Use it when

You already know which associations to test and want parameter estimates rather than a search for a model.
One categorical variable is the response and the others explain it, but you want to stay within the table framework (multinomial logit).
Comparing nested designs, for example main effects against all two-way interactions, by the change in likelihood-ratio chi-square.

Use something else when

You want the data to suggest the model through backward elimination: use hierarchical loglinear analysis.
You need custom designs with selected interactions, cell structures or covariates: the panel offers saturated, all two-way and main-effects designs only.
A numeric response or continuous predictors: use logistic regression or generalized linear models.

How to run it in Crosstabs

01
Choose General loglinear analysis
Open your data in Statistics mode and, in the General linear model group, pick General loglinear analysis (GENLOG).
02
Pick the model type
Set Model type to Poisson loglinear for symmetric association analysis, or Multinomial logit when one factor is the response.
03
Set the factors
For Poisson tick two to six Factors. For a logit model choose the Dependent factor and tick the Independent factors.
04
Choose the design and run
Set Design to Saturated, All two-way interactions or Main effects only and click Run. In a logit model the response–factor associations enter with All two-way interactions; Main effects only fits the response margin alone.

Run General Loglinear Analysis in your browser

Worked example

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

Try with example data

Data

300 synthetic trial accounts cross-classified by renewed (Yes, No), onboarding (Guided, Self-serve) and size (Large, Small). Renewal is more likely after guided onboarding and for large companies, with no interaction between the two (fixed seed). Model type = Multinomial logit, Dependent factor = renewed, Independent factors = onboarding, size, Design = All two-way interactions.

Goodness-of-Fit Tests

Item	Value	df	Sig.
Likelihood Ratio	0.355	1	.551
Pearson Chi-Square	0.354	1	.552

Model: Multinomial Logit. Design: Constant + renewed + onboarding + size + renewed*onboarding + renewed*size + onboarding*size.

Parameter Estimates

Parameter	Estimate	Std. Error	Z	Sig.	Lower Bound	Upper Bound
[renewed = No]	0.668	0.204	3.274	.001	0.268	1.067
[renewed = No] * [onboarding = Guided]	−0.880	0.243	−3.624	<.001	−1.356	−0.404
[renewed = No] * [size = Large]	−0.871	0.251	−3.466	<.001	−1.364	−0.378

Constants for each explanatory-factor combination are not parameters under the multinomial assumption and are not shown. Redundant parameters fixed at zero are omitted.

Cell Counts and Residuals

renewed	onboarding	size	Observed Count	Expected Count	Adjusted Residual
No	Guided	Large	15	16.184	−0.595
		Small	45	43.816	0.595
	Self-serve	Large	25	23.816	0.595
		Small	55	56.184	−0.595
Yes	Guided	Large	49	47.816	0.595
		Small	53	54.184	−0.595
	Self-serve	Large	28	29.184	−0.595
		Small	30	28.816	0.595

How to read it

With Design = All two-way interactions, the logit model links renewed to onboarding and to size but has no three-way term, and it fits the 2×2×2 table: likelihood-ratio χ²(1) = 0.355, p = .551, and no adjusted residual exceeds 0.60 in absolute value. The association parameters with renewed carry the effects: [renewed = No] * [onboarding = Guided] = −0.880 (p < .001) and [renewed = No] * [size = Large] = −0.871 (p < .001); each is the log odds ratio of not renewing (Guided versus Self-serve, Large versus Small), so the negative values mean guided onboarding and large companies renew more often.

Output tables explained

Data Information: Valid, missing and weighted cases and the number of categories of each factor.
Goodness-of-Fit Tests: Likelihood-ratio and Pearson chi-square with df; footnotes state the model and the full design. A small p means the design misses associations.
Parameter Estimates: Estimate, standard error, Z, significance and confidence interval per parameter; last-category parameters are redundant and fixed at zero. In logit models the explanatory-factor constants are not shown.
Cell Counts and Residuals: Observed and expected counts and percentages with raw, standardized, adjusted and deviance residuals; adjusted residuals beyond about ±2 point to misfit.

Assumptions and notes

Counts are Poisson (loglinear) or product-multinomial given the explanatory factors (logit), with each case in one cell.
Expected counts are not too small; for saturated designs 0.5 is added to every cell so all parameters exist.
Tables are limited to six factors and 5,000 cells; missing values are excluded listwise and only frequency weights are accepted.

How the results are validated

GENLOG is compared with statsmodels Poisson GLM fits for a Poisson main-effects model, a Poisson all-two-way model and a multinomial logit two-way model on three factors: likelihood-ratio and Pearson chi-square agree within 1e-8 with identical df, estimates within 1e-7 (the Poisson constant within 1e-8), standard errors within 1e-6, expected counts within 1e-8 and adjusted residuals within 1e-6. Validation evidence

Questions

Poisson loglinear or multinomial logit?

Use Poisson when all variables play the same role and you study their associations. Use multinomial logit when one factor is the response; its parameters then read as log odds ratios, as in logistic regression, and the explanatory margins are fixed.

Why are so many parameters zero?

With reference coding, every parameter involving the last category of a factor is redundant and set to zero. The non-zero parameters are contrasts with those last categories.

How do GENLOG and HILOGLINEAR differ?

HILOGLINEAR searches for a hierarchical model with K-way tests and backward elimination. GENLOG fits one design you choose and gives parameter estimates and residuals for it, including logit models. A common workflow is to find the model with HILOGLINEAR and then describe it with GENLOG.

Which cells should I look at first?

Sort by the absolute adjusted residual. Cells beyond about ±2 are where observed counts depart most from the model and often point to the interaction the design leaves out.

Run General Loglinear Analysis 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.