Alpha diversity -- ANOVA vs. LME -- how to choose which method

Hello QIIME2!

I have been reading through the forum and all tutorials on qiime2 but I am at a loss as how to proceed. My question is in regards to deciding if an ANOVA or LME is better fit for alpha-diversity analysis. I have run core-diversity and alpha-significance but I was wanting to run a model where I could evaluate multiple factors and their interactions at once. Thus, I was drawn to ANOVA and LME.

The set of of the experiment and question is this:
Microbial swabs were collected from two biological sites and 4 different time points on 20 animals (10 in 1 treatment group and 10 in another). I see this as a repeated-measures design in some regards and have run the following code for LME:

qiime longitudinal linear-mixed-effects --m-metadata-file MillsCompleteMetadata1.tsv core-metrics-results_merged_12000_year1/evenness_vector.qza --p-state-column Time --p-group-columns Sample_Type,treatment --p-individual-id-column Animal --p-metric pielou_evenness --o-visualization lme-evenness-type_treatment.qzv

but I also ran an ANOVA:
qiime longitudinal anova --m-metadata-file MillsCompleteMetadata1.tsv core-metrics-results_merged_12000_year1/evenness_vector.qza --p-formula pielou_evenness~TimeSample_Typetreatment --o-visualization evenness_anova_time_type_trt.qzv

The outputs are very similar but in the LME there is not effects of Time but in the ANOVA there is an effect of Time. When I look at the alpha-significance with Time there is a HUGE difference (as in very very very small p-value and q-values) thus I am inclined to think Time is doing something. However, the LME model would perhaps fit better and look to see if individual animal microbiome is changing over time in a better way?

I should mention that I did run ANOVA with shannon and a similar result to evenness was reported. Faith_pd showed no difference in Time in the LME or ANOVA.

Thank you!!


Hello Hannah,

Pay attention to the assumptions of the model - they will help you to choose the appropriate one. ANOVA has an assumption of independent data points. The samples taken from a single subject are dependent (repeated measurements taken from a single subject). Thus, ANOVA assumptions are violated under this study design.
The models in longitudinal should specifically account for this.



Thank you!! I will proceede with LME! I appreicate your help on this and confirming some hesitations on the ANOVA.

Thank you!