How do I perform Turkey post-hoc test (for multiple comparisons) using linear mixed effect model with random (subjects) and fixed effects (3 interventions [groups], 3 repeated measures [time]) in SPSS?
SPSS has options for LSD, Bonferonni and Sidak only
The Tukey HSD test isn't defined for repeated measures and linear mixed models. However, with only three means being compared Fisher's LSD would be more powerful than Tukey's HSD and often the same Type I error protection.
The innovation in Tukey's HSD is that Type I error is controlled for partial null hypotheses (not just the omnibus null). This isn't necessary in the case of just three means because of logical constraints in the pairwise tests (so protection for omnibus null is sufficient).
For more than three means I'd recommend something like the Hochberg correction as its easy enough to do by hand (using just the p values from the uncorrected pairwise t tests).
Hochberg, Y. (1988). A sharper Bonferroni procedure for multiple tests of significance. Biometrika, 75, 800–803. doi: 10.2307/2336325.
SPSS should have an EMMEANS call option, that will probably be more appropriate for that kind of model. But I don't know if you need to access this by command syntax. Or how EMMEANS will handle this kind of model. Likely, you will want to consult the software documentation.
@Thom, thank you so very much for your clear explanation. The SPSS syntax command has only three options for multiple comparisons, LSD (default), Bonferonni and Sidak. For my primary outcomes, I used the Bonferonni adjustment whereas for my secondary outcomes, since they are exploratory, I used the LSD.
I think the Bonferonni adjustment is probably too conservative for pairwise comparisons with only three means. The Fisher's LSD will be more powerful (at least in cases where the omnibus F test is significant as that's the only mechanism for FW error control in the LSD case). Bonferonni and Sidak don't require that omnibus F test is significant (and doing so will make them even more conservative).
@Thom. Thank you and I think I will do as you suggested. So in my own case the LSD is adequate. I also learned that many authors don't report the type of post hoc analysis applied following LMM for fixed and random factors in SPSS although the LSD is the default analysis.
- Badges
- Science topic
- Similar topics
- Engineering
- Electrical Engineering
- Electrodes