I have multilevel, cross-sectional data with individuals nested in districts. I am running a three variable path analysis (with numerous controls). I have read Preacher et al.'s (2010) work on multilevel SEM and have estimated 1-1-1 mediation models and calculated the within and between indirect effects. Only the within indirect effects are statistically significant. Given this finding and there is no good theoretical reason to focus on the between indirect effects in the first place, I would prefer to treat the between-level simply as nuisance.
I believe there are two ways to do this: adjustment for clustering or a within-group analysis (group-demeaned or LSDV approaches). Would you recommend a within-group path analysis in this instance? If so, do you know of any references?