Equal Variance in Post Hoc?

[kesm]

Using SPSS and doing an ANOVA with Post Hoc, but unsure if I should be assuming equal variance or not. The groups I'm looking at are each divided by length of relationship (<6 months, 6-12 months, 1-2 years, 2-3 years, 3-4 years, 5-10 years, 10+ years)

[juilletmercredi]

There's a test in the SPSS output that will tell you - it's called "Levene's test for heteroscedasticity" or "Levene's test for equality of variances". I think the SPSS output defaults to producing it, but you may have to check a box in the ANOVA command menu to get it to show up. If the test is not significant, then you read the line for equal variances assumed. If the test is significant (p < .05), then the variances are not equal and you need to read the line "equal variances not assumed." Basically what SPSS does in the latter case is adjust your degrees of freedom to account for the heteroscedasticity, with the net effect of making it more difficult to reject the null.

 

When in doubt, the "equal variances not assumed" estimate is the more conservative of the two.

[lewin]

If it helps calm your fears, ANOVA is usually very robust to violations of its assumptions. Also, based on those bins I would be surprised if it's normally distributed. In university students you'll get a positive skew and in grownups you'll probably get a negative skew.