I have a mediation model, where a level-2 independent variable influences a level-2 mediator and this then has an impact on a level-1 dependent variable. For argument sake, let's call level-2 the firm, and level-1 the worker, and let us assume that there are 100 firms and 5000 workers in total.
Can someone answer or point me to a resource that could answer the following: Why would a multilevel mediation analysis (2-2-1) be better than simply aggregating or averaging all the worker data at the firm level? Then, I can perform a simple regression or so at the firm level to test any hypothesis.