How to Interpret ANCOVA Output in SPSS (and Report It in APA 7)

An ANCOVA (analysis of covariance) compares group means on an outcome while holding constant a continuous covariate, such as a pre-test or baseline measurement. SPSS GLM Univariate prints six or seven tables, and the main result sits in the Tests of Between-Subjects Effects table, but you must also read the adjusted means, the covariate's effect, the pairwise comparisons, and the homogeneity-of-regression-slopes check. This guide reads every table row by row with a worked example (three study methods, pre-GPA as the covariate), flags the misreadings that cost marks, and ends with the APA 7 write-up. For the reporting format on its own, see the ANCOVA APA guide.

What SPSS gives you

Analyze > General Linear Model > Univariate, with the outcome in Dependent Variable, the grouping factor in Fixed Factor(s), the covariate in Covariate(s), Descriptive Statistics and Homogeneity Tests and Estimates of effect size ticked under Options, Display Means for... set to your factor with Compare main effects and Confidence interval adjustment: Bonferroni ticked under EM Means, produces:

1. Descriptive Statistics: N, mean, and SD per group (unadjusted).
2. Levene's Test of Equality of Error Variances: the homogeneity-of-variance test.
3. Tests of Between-Subjects Effects: the ANCOVA results for the factor, covariate, and error, with partial eta squared. This is the main result.
4. Estimated Marginal Means: the adjusted means and standard errors for each group after controlling for the covariate.
5. Pairwise Comparisons: Bonferroni-adjusted tests of differences between adjusted means.
6. Homogeneity of Regression Slopes (if you fit the interaction model): whether the covariate relates to the outcome the same way in each group.

The worked example

Seventy-five students took an exam after one of three study methods (Lecture, Video, or Interactive; 25 per group). Their pre-exam GPA was measured as a covariate. All tables below come from one dataset.

Table 1: Descriptive Statistics (unadjusted)

• N, Mean, Std. Deviation: the raw (unadjusted) numbers per group. Video had the highest mean (80.32), Lecture the lowest (73.12), but these will change after adjusting for pre-GPA.
• These summary statistics are the only descriptives SPSS GLM shows in the Descriptive Statistics table.

Table 2: Levene's Test of Equality of Error Variances

Levene's test checks whether the groups have similar variances, an ANCOVA assumption. Recent SPSS versions show additional rows (median-based tests); the "Based on Mean" row is shown here. Sig. = .752 is well above .05, so the equal-variance assumption holds. This Sig. is not your ANCOVA result.

Table 3: Tests of Between-Subjects Effects

• Corrected Model row: everything the model explains beyond the grand mean (factor and covariate together). With a covariate its sum of squares is not the sum of the rows below it, because the factor and the covariate overlap. The df = (groups − 1) + (covariates) = 2 + 1 = 3. Not usually reported.
• Intercept row: the grand mean adjusted for the covariate. Not reported.
• pre_gpa row (covariate): the covariate's own effect. F(1, 71) = 25.12, p < .001. This says pre-GPA significantly predicted exam scores. You must report this to justify the adjustment. The partial η² = .261 shows the covariate's effect size.
• Study Method row (factor): the adjusted effect of the grouping variable. This is the main result. F(2, 71) = 10.11, p < .001, partial η² = .22. Study method mattered even after holding pre-GPA constant. Partial eta squared = SS(effect) / (SS(effect) + SS(error)), so .22 is the study-method sum of squares as a share of itself plus the error sum of squares, with pre-GPA held constant.
• Error row: the variation unexplained after removing the factor and covariate effects. Its df = N − groups − covariates = 75 − 3 − 1 = 71.
• Total: the sum of every squared score, including the grand mean. Not reported.
• Corrected Total: all variation around the grand mean. df = N − 1 = 74. Not reported.

Table 4: Estimated Marginal Means

Covariates appearing in the model are evaluated at the following values: pre_gpa = 3.02.

• Mean (Adjusted or Estimated Marginal Mean): the predicted exam score for each group when pre-GPA is at its grand mean (3.02 in this example). These are the means that the ANCOVA actually tested, not the raw means in Table 1. Video still appears highest (80.30), but the gap to Lecture (73.25) is now adjusted for baseline GPA differences.
• Std. Error: the SE of each adjusted mean, computed from the model's covariance matrix. These SEs allow you to build the 95% CI.
• 95% Confidence Interval: the CI around each adjusted mean, computed with the t critical value (1.994 for df = 71).

Table 5: Pairwise Comparisons (Bonferroni adjustment)

The mean difference is significant at the 0.05 level.

• (I) and (J) Study Method: SPSS compares every group with every other group in both directions. The six rows show three unique comparisons. Lecture vs Video and Video vs Lecture are the same test with opposite signs.
• Mean Difference (I-J): the (I) adjusted mean minus the (J) adjusted mean. An asterisk marks p < .05 after Bonferroni correction.
• Sig.: the Bonferroni-adjusted p value. The adjustment is min(1, unadjusted p × 3) to correct for three pairwise comparisons. A value of .000 means p < .001 after adjustment. SPSS's default is LSD (no adjustment); Bonferroni is chosen under Confidence interval adjustment, and Sidak is the other option there.
• 95% Confidence Interval: Bonferroni-adjusted, excludes 0 exactly when Sig. is below .05.

Read once per unique pair: the Video group scored 7.05 points higher than Lecture (p < .001), and 4.30 points higher than Interactive (p = .025). Lecture and Interactive did not differ significantly (2.76 points apart, p = .257).

Table 6: Homogeneity of Regression Slopes

This check is not a separate SPSS table. Instead, you refit the model adding the factor × covariate interaction and look at the Tests of Between-Subjects Effects table. The rows below are from a model where Study Method × pre_gpa is included:

F(2, 69) = 0.24, p = .790 is well above .05, so the slopes are homogeneous. The covariate has the same linear relationship with the outcome across all three study methods, and ANCOVA is appropriate. If the interaction were significant (say, p < .05), ANCOVA would not be valid for your data; you would report separate slopes per group or use a different design.

Misreadings that cost marks

• Reporting the unadjusted means from Table 1. The ANCOVA tests the adjusted means (Estimated Marginal Means), not the raw group means. Always report and compare the adjusted means.
• Not reporting the covariate's effect. Your write-up must include the covariate's F, df, and p to justify the adjustment and show the covariate mattered.
• Claiming significance for pairs that are not. Use the Pairwise Comparisons table to see which pairs differ at p < .05. Do not claim significance based on the adjusted means alone.
• Forgetting the homogeneity-of-slopes check. If the interaction is significant, ANCOVA is not valid for your data.
• Writing p = .000. Write p < .001. That applies to the covariate, factor, and pairwise tests.
• Reporting only one df. Each F needs both: F(2, 71), never just F(2).
• Using Levene's p-value as the ANCOVA result. Levene tests variance equality, not group differences.

The APA 7 format template

After adjusting for [covariate], a one-way ANCOVA showed a significant effect of [independent variable] on [dependent variable], F(df1, df2) = X.XX, p = .XXX, partial η² = .XX. The covariate, [covariate], was significantly related to [dependent variable], F(1, df2) = X.XX, p = .XXX. Bonferroni-corrected pairwise comparisons of adjusted means showed [the significant pairs].

The write-up for this example

After adjusting for pre-exam GPA, a one-way ANCOVA showed a significant effect of study method on exam scores, F(2, 71) = 10.11, p < .001, partial η² = .22. The covariate, pre-exam GPA, was significantly related to exam scores, F(1, 71) = 25.12, p < .001. Homogeneity of regression slopes held, F(2, 69) = 0.24, p = .790. Bonferroni-corrected pairwise comparisons of adjusted means showed that the Video group (M = 80.30, SE = 1.12) scored significantly higher than the Lecture group (M = 73.25, SE = 1.12), p < .001, and significantly higher than the Interactive group (M = 76.01, SE = 1.12), p = .025. Lecture and Interactive did not differ significantly, p = .257.

Let KyroStat do the write-up for you

Assembling adjusted means, standard errors, the covariate effect, partial eta squared, pairwise comparisons with Bonferroni correction, and the homogeneity-of-slopes check is a lot to get right. KyroStat runs the ANCOVA on your data, checks homogeneity of slopes, and hands you the finished APA 7 sentences, a table of adjusted means with pairwise comparisons, diagnostic plots, and the underlying Python code. Upload your spreadsheet, and the report is done in seconds.

Also useful for the write-up (affiliate links): Proofreading for your thesis or paper (20% off with code PAP20): Paperpal

Frequently asked questions

Why report adjusted means and not the raw group means? The ANCOVA tests and adjusts the group means for the covariate's influence. The adjusted (estimated marginal) means are what the test actually evaluated. Reporting raw means would mislead readers about what was tested.

What is homogeneity of regression slopes? It means the relationship between the covariate and outcome (the slope) is the same in every group. If slopes differ, the covariate does not adjust fairly across groups, and ANCOVA is not appropriate. Test it by fitting the group × covariate interaction; if it is non-significant, the slopes are homogeneous.

What do I report if the homogeneity-of-slopes check fails? If the interaction is significant, consider fitting separate regression lines per group, or report the ANOVA without the covariate adjustment, or use a different design. ANCOVA assumes homogeneous slopes; violating that assumption can lead to misleading conclusions.

Should I report both the factor F and the covariate F? Yes. The covariate F justifies why you adjusted; the factor F is the main result. Both belong in your write-up.

Can I use a post-test as the covariate? No. The covariate should be a baseline measured before the treatment (such as a pre-test), so the adjustment does not remove part of the treatment effect.

Can I have more than one covariate? Yes. Add each covariate's F, df, and p to your write-up. Each covariate gets its own row in the Tests of Between-Subjects Effects table.

Why use Bonferroni instead of other adjustments? Bonferroni is the most common choice for these pairwise comparisons (SPSS's default is LSD, which does not adjust); it is conservative but straightforward. Sidak is another option. Note that Tukey post hoc tests are not available for models with a covariate (only for pure ANOVA).

What is partial eta squared for ANCOVA? Partial η² = SS(effect) / (SS(effect) + SS(error)). It is the proportion of residual variance explained by the effect after removing the covariate's influence. For the Study Method row, partial η² = .22, meaning the factor accounts for 22% of the variance left after the covariate.

Which related guides might I need? Often these: one-way ANOVA output (without a covariate), ANCOVA reporting for the exact APA format, partial eta squared, Levene's test, and two-way ANOVA for multiple factors.