Linear Mixed Effect Models

[pumpkin]

Hi,
I am not clear about the linear mixed effect models. This is the code from the tutorial.

qiime longitudinal linear-mixed-effects \
  --m-metadata-file ecam-sample-metadata.tsv \
  --m-metadata-file shannon.qza \
  --p-metric shannon \
  --p-group-categories delivery,diet,sex \
  --p-state-column month \
  --p-individual-id-column studyid \
  --o-visualization linear-mixed-effects.qzv

Does it treat the studyid as the random effect? The delivery/sex is the fixed effects? In addition, the individual-id-column parameter is a metadata column that indicates the individual subject/site that was sampled repeatedly. So does it mean that the the software can know the relationship between the sampled ones and study id?
Thanks a lot.

[Nicholas_Bokulich]

Hi @pumpkin,
Thank you for posting!

@pumpkin

Does it treat the studyid as the random effect?

No. A random intercept is used as a random effect for each individual in the individual-id-column (i.e., studyid in this case), but individual-id-column is not used as a random effect. The current release (2017.10) does not have support for user-defined random effects, but the next release (end of this month) will add this feature.

@pumpkin

The delivery/sex is the fixed effects?

Correct. Delivery, diet, sex, and month are all fixed effects (state-column is a fixed effect but is defined separately from group-categories because we use that value for other things like plotting.)

@pumpkin

So does it mean that the the software can know the relationship between the sampled ones and study id?

Correct, the model takes individual differences into account. Currently this just takes the form of random intercepts for each individual, but random slopes (to account for different trajectories between individuals) will be supported in the next release.

I hope that helps clarify this method for you!