stegmann at psych.uni-frankfurt.de
Thu Jun 24 12:30:47 CEST 2010
I am just conducting various mediation analyses with data from individual participants who are nested in teams. I found the “sobel.lme” function to be pretty helpful when doing so. (A big thank you to Paul Bliese for his multilevel-package and the awesome manual!)
Here’s the example from the Help-file:
#A small but significant indirect effect indicates leadership mediates
#the relationship between work hours and well-being.
The package estimates three regression models according to standard-mediation-analysis-operating-procedure. Each model has one intercept and one/two slopes.
Does “sobel.lme” estimate:
a) random intercept, fixed slope(s) → as in for example Krull & MacKinnon (2001) Multilevel Modeling of Individual and Group Level Mediated Effects
b) random intercept, random slope(s) → Does not necessarily involve a predictor for the slope on level 2 – as this would be phrased moderation rather then mediation (potential matter of dispute, though…).
c) fixed intercept, fixed slope(s) → wouldn’t make much sense given the purpose of the function…
d) fixed intercept, random slope(s) → wouldn’t make much sense either…
I know R is sometimes smarter than one would expect after having used other statistical software for years, but…
Multilevel Mediations can take in principle three forms (Krull & KacKinnon, 2001; numbers indicate levels):
Is “sobel.lme” really smart enough to “notice” that a variable is constant within groups and therefore assigns it to the group level accordingly?
If not, it would probably help to know which of the above three mediation-forms is implemented in the function.
Thanks for reading this lengthy mail. It would be really nice if anyone could tell me (and the help-archive) anything regarding these two questions.
Dipl.-Psych. Sebastian Stegmann
Goethe University, Frankfurt, Germany
More information about the R-help