In his book Statistical Rethinking (2nd edition), Richard McElreath proposes that "multilevel regression deserves to be the default form of regression". He expands:

Papers that do not use multilevel models should have to justify not using a multilevel approach. Certainly some data and contexts do not need th emultilevel treatment. But most contemporary studies in the social and natural sciences, whether experimental or not, would benefit from it. Perhaps the most important reason is that even well-controled treatments interact with unmeasured aspects of the individuals, groups, or populations studied. (p.15)

The only shortcomings he mentions in the next paragraph are related to its practical implementation:

Fitting and interpreting multilevel models can be considerably harder than fitting and interpreting a traditional regression model

My question is, what would be technical arguments against the use of multilevel modeling by default?

  • 1
    $\begingroup$ This seems like a burden of proof fallacy, that is not based on practice or theory. $\endgroup$
    – AdamO
    Commented May 4, 2021 at 15:48
  • 1
    $\begingroup$ Notice that Richard McElreath writes in the context of Bayesian statistics, where every parameter is considered as a random variable, so the distinction to fixed and random effects does not make much sense stats.stackexchange.com/questions/348982/… also multilevel models are pretty straightforward to implement in modern Bayesian statistical software, so he's likely quite biased for this kind of models. $\endgroup$
    – Tim
    Commented May 4, 2021 at 16:41
  • $\begingroup$ @AdamO I am unsure If I follow your reasoning. There is a point stated that there is rarely an instance within a diversity of fields where there is no nested structure to data, and that the default of fixed-effects models should instead be the one to be exceptionally argued. Given that this is not the current modeling culture in many fields, my question asks about what are the technical reasons that one could argue of why changing such a default is a bad idea. $\endgroup$
    – Kuku
    Commented May 4, 2021 at 22:44


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.