Hello, a common device used to separate Level 1 effects into Within and Between components is to group mean center the Level 1 predictor and introduce the cluster mean as a Level 2 predictor. In the context of mediation analysis, I suggested “unconflating” Level 1 effects in mediation models that include 1–1 linkages, a strategy called unconflated multilevel modeling (UMM). However, even though the resulting Between effect is separated from the Within effect, the Between effect is biased toward the Within effect to the extent that intraclass correlations2 (ICCs) are low and cluster sizes are small.

This bias, in turn, will also contribute to bias in any Between-level indirect effect. Unconflated Multilevel Modeling could be done by using multilevel structural equation modeling (MSEM), you could see this link for more explanation

centering variables.

Thanks Haitham. I was familiar with the papers of Preacher et al. (2011)., but was looking for a more practical interpretation of the literature. As far as I know now Mplus can do this job better than AMOS. The following syntax (page 3 in the PDF) makes a lot clear: https://www.statmodel.com/download/Preacher.pdf