Alternative statistical tests alpha diversity

Hi all,

I have a repeated measures design where six decedents were sampled across time, but they were sampled in two different seasons. I want to compare alpha diversity between seasons, individuals, and across time. While I looked at using a linear mixed effect model (q2-longitudinal), the plotted residuals were unusually high and low, making this test inappropriate. I have also read that kruskal-wallis should not be used for repeated measures, but maybe that is incorrect? Is there an alternative group comparison method in qiime2 that I am overlooking that makes little to no assumptions about the data structure? I have searched the forum for a similar question, but have not come across a similar post. If this has already been asked, I apologize.

Best,
Alex

Hi @emmo1,

This means that the model is a poor fit and can be improved, not necessarily that the test is inappropriate. There is quite possibly some factor that should be included in your model to capture the variance explained by that factor; e.g., you are including time and season as a fixed effect and individuals as a random effect, but maybe there is another factor (diet change between seasons?).

q2-longitudinal has an ANOVA action, but ANOVA sure makes assumptions about data structure (normality, etc), so is likely to have more problems than LME.

Iā€™d recommend fiddling with your model to see if you can improve the residuals. Maybe try separating by season and looking at temporal changes within each season separately. I wonder if season has some covariates that may be breaking your model.

Good luck!

1 Like

Hi @Nicholas_Bokulich,

Thank you for the recommendation. I will try to tweak the model and see if the residuals improve.

Best,
Alex

1 Like

This topic was automatically closed 31 days after the last reply. New replies are no longer allowed.