Correct fitlme posthoc tests for multiple comparisons

This post helped me work out how to use coefTest to analyze single contrasts for linear mixed effects models.

Does anybody know how to perform a series of contrasts and obtain p-values that are corrected for the multiple comparisons (ideally via Tukey)?

Answer 1

[UDAYA PEDDIRAJU]

Hi Ken,

To perform post-hoc tests with Tukey's correction for multiple comparisons on a linear mixed effects model "fitlme" in MATLAB, follow these steps:

• Fit the Linear Mixed Effects Model:

lme = fitlme(data, 'response ~ fixedEffects + (1|randomEffects)');

• Perform Post-hoc Tests with Tukey's Correction:

[~,~,stats] = anova(lme);
results = multcompare(stats, 'CType', 'tukey-kramer');

• "CType", set to 'tukey-kramer', applies Tukey's correction for multiple comparisons.
• "results" will contain the comparison results with adjusted p-values.

This will provide you with Tukey-adjusted p-values for your multiple comparisons.

Let me know if this gives a workaround.

Documentation for reference: anova: https://www.mathworks.com/help/stats/anova.html and https://www.mathworks.com/help/stats/multcompare.html?searchHighlight=multcompare&s_tid=srchtitle_support_results_1_multcompare.

[Ken Campbell]

Hi Udaya,

Thanks for thinking about this. Unfortunately, anova only tests the main effects in linear mixed effects models. I want to compare categorical levels with a factor.

I can run individual comparisons via coefTest but only for one contrast at a time. If I run several tests in series, I need to correct them for the multiple comparison - hence the original question.

Essentially, I want the multcompare functionality of standard anova but with a linear mixed model design. Sadly, multcompare() does not take lme output.

Ken