A similar question was raised here.

https://www.researchgate.net/post/Which-is-better-GLM-or-MLM-in-association-mapping

Join us

ICIT2023

http://icit.zuj.edu.jo/icit2023/Index.html

I'm not familiar with the application area, but it's straightforward in a sense because the GLM is in sense just a special case of the MLM. The MLM incorporates multiple random factors (rather than just the random factor implicit in the error term for a standard GLM). If you have model like this:

MLM:

y ~ 1 + x1 + x2 + (1|unit)

it will estimate a random intercept for unit with a variance for unit as well as a variance for the error term. If the variance for unit is zero then its equivalent to a single level GLM

y ~ 1 + x1 + x2

This is a trivial example but the principle applies to more complex designs. So if there is an expectation that units will vary other than because of sampling error (i.e., there are individual differences in units) then the MLM is always superior in the sense that its correctly capturing the structure of the variation in the data. The single level model in this case will be inferior in particular because the standard errors will be to small (leading to Type I error inflation for example).

Read the Answer

https://www.researchgate.net/post/Which-is-better-GLM-or-MLM-in-association-mapping