Question about mediation Analysis and two level of observation

[Michael Lee]

Hi everyone,
1.First of all I want to do mediation and moderated mediation analysis on a panel data, is there any instruction for that? I couldn't find anything straight.

2.In mediation analysis is it necessary to observe all the variables at same level? I mean for example if industry competition is a variable at industry level and company sales is a variable at firm level, can I estimate effect of industry competition on company's sale or I need to have variables at same level? what about moderation analysis?

Thanks a lot.

[Daniel Schaefer]

1. My understanding is that you typically want xtset and xtreg for panel data. You might also want to read through this document: https://www.stata.com/manuals/xt.pdf

2. You could do this as an HLM model with observations nested within industries and nested within years to capture the panel design. There's nothing that prevents you from having multi-level data and running a straightforward (non HLM) regression, but you are likely violating some model assumptions if you do. I do not know how robust the model will be to these assumption violations, but perhaps someone else here does. Another option might be to treat your higher level variables as dummies or as categorical variables - but this isn't without its own problems of course. Too many dummies and you run into multiple collinearity issues, and it will not be straightforward to treat a continuous variable as a factor.