15 SPSS Errors and Warnings Every Dissertation Student Hits, Decoded (2026)

SPSS messages are written for statisticians, not for the person trying to finish a results chapter at eleven at night. “This variable is not permitted in this list” tells you nothing about what to change, and the most dangerous problems produce no message at all.

This is a reference list of the fifteen that stop student analyses most often. Each entry gives what the message means, what actually caused it, and the fix. Bookmark it and jump to the one you are looking at.

1. “Variable not permitted in this list”

What it means: the variable’s type or measurement level does not match what the dialog accepts.

The real cause: almost always a numeric variable stored as string, usually because it was imported from Excel or a survey export with a stray space or a text header in the data rows.

The fix: in Variable View, check the Type column. If it says String, convert with Transform > Automatic Recode or re-import with the column typed correctly. Then set the Measure column — Scale, Ordinal or Nominal — because several procedures filter their variable lists on it.

2. “Text: (end of command expected)”

What it means: a syntax error, usually of punctuation.

The real cause: a missing full stop at the end of the previous command. Every SPSS command ends with one, and omitting it makes SPSS read the next command as a continuation.

The fix: look at the line before the one flagged. Also check for a variable name containing a space or a hyphen, both of which need care in syntax.

3. No message at all, and the result is wrong

What it means: the most expensive failure mode in SPSS is silence.

The real cause: a procedure ran on data that did not mean what you thought. Variables assigned to the wrong slots in a repeated-measures design, a status variable coded backwards, or a grouping variable with an undeclared missing code counted as a real group.

The fix: before interpreting anything, check that N matches your expectation and that group sizes match your recruitment. A number appearing in the output is not evidence that the analysis was correct.

4. “Only one case in the data” or an N far smaller than expected

What it means: listwise deletion has removed nearly everybody.

The real cause: a case is dropped if it is missing any variable in the analysis. Twelve variables each with a few missing values can jointly eliminate most of the sample.

The fix: run Analyze > Descriptive Statistics > Descriptives and compare valid N across variables to find the culprit. Then decide deliberately between deletion and an alternative using the guide to handling missing data.

5. A stray 99, 999 or -1 inflating your mean

What it means: nothing is wrong with SPSS. Your missing-value codes are being treated as data.

The real cause: the codes were never declared. A 99 meaning “prefer not to say” averaged into a five-point scale destroys the mean, and no warning appears.

The fix: declare them in the Missing column of Variable View before any analysis. Then check every mean against the possible range of its scale — a mean of 12.4 on a 1–5 item is the tell.

6. “This matrix is not positive definite”

What it means: your correlation or covariance matrix is mathematically impossible to invert.

The real cause: two variables are perfectly or near-perfectly correlated, often because one is a linear combination of others — including a total score entered alongside the items that compose it. It can also occur when you have fewer cases than variables.

The fix: examine the correlation matrix for values above about .95, and remove one of each offending pair. If you entered a subscale and its total, drop the total.

7. “Determinant = .000” in factor analysis

What it means: the same problem in the specific context of factor analysis: extreme multicollinearity among your items.

The real cause: two or more items are effectively duplicates, or you have far too few cases for the number of items.

The fix: inspect the correlation matrix, remove near-duplicate items, and confirm your case-to-item ratio is adequate before rerunning — the sequence is set out in the walkthrough on exploratory factor analysis in SPSS.

8. “Maximum number of iterations reached” (logistic regression)

What it means: the model did not converge, so the coefficients printed beneath are not trustworthy.

The real cause: usually complete or quasi-complete separation — a predictor that perfectly predicts the outcome, leaving an empty cell in the cross-tabulation. The tell is an enormous coefficient with a standard error in the thousands.

The fix: cross-tabulate every categorical predictor against the outcome and look for zero cells. Merge sparse categories or drop the predictor. The same diagnosis applies to the categorical models covered in the guide to ordinal versus multinomial logistic regression.

9. Enormous standard errors and odds ratios

What it means: the estimate is uninformative even though it printed.

The real cause: too few events per predictor, or the separation problem above. An odds ratio of 4.2 million with a confidence interval from zero to infinity is not a finding.

The fix: reduce the number of predictors to fit your event count, or collapse categories. Report the reduced model, not the unstable one.

10. Levene’s test is significant in a t-test

What it means: not an error at all — a signal telling you which row of the output to read.

The real cause: unequal variances between your two groups, which is common and often harmless.

The fix: read the second row, “Equal variances not assumed”, which reports the Welch correction with fractional degrees of freedom. Report those fractional degrees of freedom exactly as shown, and say you used the correction — the convention is covered in the independent samples t-test walkthrough.

11. Mauchly’s test is significant

What it means: sphericity has been violated in a repeated-measures ANOVA.

The real cause: the variances of the differences between your conditions are unequal, which is common with three or more time points.

The fix: read the Greenhouse-Geisser corrected row rather than the “Sphericity Assumed” row, and report the corrected degrees of freedom. The full procedure is in the walkthrough on repeated measures ANOVA.

12. “More than 20% of cells have expected count less than 5”

What it means: your chi-square result is unreliable.

The real cause: too small a sample spread across too many categories.

The fix: merge categories where it is substantively defensible, or use Fisher’s exact test, which the chi-square walkthrough covers. Do not simply report the chi-square with the warning suppressed.

13. “Tolerance = .000” or VIF above 10

What it means: severe multicollinearity in a regression.

The real cause: two predictors measuring nearly the same thing, or a variable and its own transformation entered together.

The fix: drop one of the pair, combine them into a single index, or centre the variables if the collinearity comes from an interaction term. The diagnostics are explained in the multiple regression walkthrough.

14. A negative Cronbach’s alpha

What it means: a value outside the possible range for a reliability coefficient, which is always a coding problem rather than a finding.

The real cause: reverse-worded items that were never reverse-scored, so some items correlate negatively with the rest of the scale.

The fix: recode the reverse-worded items before computing the scale, then rerun. The item-total correlation column identifies which ones — see the guide to Cronbach’s alpha.

15. Output that will not paste into Word

What it means: a formatting problem rather than a statistical one, but it costs hours at the worst possible moment.

The real cause: pasting an SPSS pivot table as an image, or as a table whose borders and decimal places do not match APA conventions.

The fix: double-click the table to open the pivot editor, apply an APA-style TableLook, set decimals consistently, then copy as a Word table rather than a picture — pictures cannot be edited later and print badly.

Three habits that prevent most of this

Set up Variable View before you analyse anything. Type, missing codes and measurement level for every variable. Ten minutes here removes entries 1, 5 and much of 3 from this list permanently.

Use Paste, not OK. Every dialog has a Paste button that writes the equivalent syntax to a file. That file is a record of exactly what you did, it can be rerun after you fix a coding error, and it is the single most useful thing you can hand a supervisor who asks how you got a number.

Check N and group sizes on every output. Most silent failures show up first as an unexpected sample size. If N is not what you expect, stop and find out why before reading anything else.

When the message is not the problem

Two categories of difficulty look like software errors and are not. The first is a procedure that runs perfectly on the wrong test for your design — a decision problem, addressed by the statistical test decision guide, not by anything in a dialog box. The second is data that were never usable, which is a screening problem covered in the guide to screening invalid survey responses.

And if the licence itself is the obstacle, the free alternatives run the same analyses — see the comparison of JASP, jamovi, SPSS and R.

Once the analysis runs

Fixing the error is the fast part. Turning the output into a results chapter that reports the right statistics, in the right order, with the assumption checks stated and the notation correct, is the part that takes the evening.

Tesify takes your SPSS output and drafts that section in APA 7 — including the corrected degrees of freedom, the assumption results and the hedged interpretation — so the time you just spent debugging is not also lost to formatting.