Significant ANOVA, but no post-hoc comparison is significant

A real and legitimate result, not a mistake. It looks contradictory. It is not.

Why it happens

They test different hypotheses. The omnibus F-test asks whether means differ *anywhere*, pooling information across all groups. A pairwise test asks about two specific groups using only their data. An overall pattern can be detectable while no single pair is.

Post-hoc corrections cost power. Tukey, Bonferroni and Games-Howell raise the bar for each comparison to control the family-wise error rate. A difference that would be significant alone may not survive.

Your effect may be spread out. If means increase gradually across four groups, the overall trend is clear while no adjacent pair is far enough apart.

What to do

Do not hunt for an uncorrected test that gives you a significant pair. That is exactly what corrections exist to prevent, and reviewers spot it.

Report it honestly:

The one-way ANOVA indicated a significant effect of condition, F(3, 76) = 3.12, p = .031, η² = .11. However, no pairwise comparison reached significance after Tukey correction. The largest difference was between conditions A and D (mean difference = 2.14, 95% CI [-0.11, 4.39], p = .066).

Consider a trend test. If your groups are ordered, such as doses, a test for linear trend answers your real question with more power.

In Plotara

The summary reports the model-level verdict and every individual comparison separately, and names which is which.