Statistics for Research - Lesson 31 - One Way ANOVA - Theory and Practice in SPSS (v29)

One-way ANOVA is the go-to method for testing whether three or more independent groups have meaningfully different mean scores on a single outcome—so long as the outcome is measured on a continuous scale and the groups are comparable. When researchers previously used independent-samples t tests for two groups, the same “compare means” logic expands to analysis of variance (ANOVA) for multiple groups. The core goal is an overall check: do the group means differ enough to suggest they come from different populations, rather than just random sampling noise.

The assumptions drive which ANOVA variant and which follow-up comparisons are valid. The dependent variable must be continuous (interval or ratio). Observations must be independent—no participant can appear in more than one group. The outcome should be normally or near-normally distributed; if normality fails, bootstrapping is flagged as a later remedy. Data should avoid spurious outliers, and the spread across groups must be roughly homogeneous. In SPSS (or R), homogeneity of variance is assessed with Levene’s test. If Levene’s test is not significant, equal-variance ANOVA procedures can be used; if it is significant, equal-variance assumptions break down and Welch’s ANOVA becomes the safer choice.

Group size balance matters because unequal sizes can amplify problems with variance equality and normality. The method is described as a two-stage workflow. Stage one runs an overall hypothesis test across all groups; a significant result triggers stage two, where post hoc comparisons identify which specific group pairs differ. Post hoc tests function like multiple t tests but control the error rate across many comparisons. The transcript distinguishes the statistics: t tests focus on mean differences between two groups, while the F test in ANOVA evaluates whether variability among group means is large relative to within-group variability—capturing differences across multiple groups at once.

A practical SPSS walkthrough illustrates the decision rules. For “vision” communication scores across junior, middle, and senior job ranks, the means rise across groups (junior 4.84, middle 5.05, senior 5.31), but Levene’s test is insignificant and the ANOVA main effect is also non-significant (F ≈ 1.84, p ≈ 0.160). With no significant overall effect, post hoc pairwise tests are unnecessary. The transcript then shows a contrasting scenario using “development initiatives,” where Levene’s test is significant, so Welch’s ANOVA is used. Welch’s test is significant (W ≈ 4.76), and post hoc comparisons proceed with Games–Howell (appropriate for unequal variances). In that example, junior differs significantly from senior, while junior vs. middle and middle vs. senior comparisons are not significant.

Finally, reporting guidance ties results to the correct statistics: when equal variances hold, report the ANOVA F statistic and p value; when they don’t, report Welch’s statistic and p value, then report only the significant pairwise differences from the chosen post hoc test (Tukey for equal variances, Games–Howell or similar for unequal variances). The takeaway is procedural: check assumptions first, choose the correct ANOVA engine, and only run the post hoc comparisons that the assumption structure and overall significance justify.