Hello,
Based on my reading on structural-equation models, using them to test for mediation with a panel (xt) data set that
is both very unbalanced and has substantial serial
autocorrelation is too problematic to be useful. (To test my hypothesis, my model also includes lagged variables.)
But, I thought I would post here in case there are approaches that address/mitigate some of the issues with conducting such an analysis.
I believe that fixed effects model with robust clustered standard errors address some of the issues. But, when it comes to testing mediation, my understanding is fixed effects
models introduce a whole other set of issues. Alternatively, an analysis of first-differences also address some
of these issues. But, I have not seen this approach used for mediation analysis, so I am assuming it may be problematic.
The test of the mediation hypothesis is more exploratory and non-essential to the main goals of my overall analysis. So, I am starting to think that conducting a test of mediation with this particular data set is more problematic than useful. But, I would love to hear others’ thoughts on this, particularly if
there is something useful that I haven’t yet considered.
Thank you!
Based on my reading on structural-equation models, using them to test for mediation with a panel (xt) data set that
is both very unbalanced and has substantial serial
autocorrelation is too problematic to be useful. (To test my hypothesis, my model also includes lagged variables.)
But, I thought I would post here in case there are approaches that address/mitigate some of the issues with conducting such an analysis.
I believe that fixed effects model with robust clustered standard errors address some of the issues. But, when it comes to testing mediation, my understanding is fixed effects
models introduce a whole other set of issues. Alternatively, an analysis of first-differences also address some
of these issues. But, I have not seen this approach used for mediation analysis, so I am assuming it may be problematic.
The test of the mediation hypothesis is more exploratory and non-essential to the main goals of my overall analysis. So, I am starting to think that conducting a test of mediation with this particular data set is more problematic than useful. But, I would love to hear others’ thoughts on this, particularly if
there is something useful that I haven’t yet considered.
Thank you!
