daj posted on Tuesday, February 23, 2010 - 4:42 pm
I have a survey dataset, in which each respondent evaluates two agents. Each evaluation is treated as an individual observation in my dataset (so I have n x 2 observations, where n is the number of respondents). I'm treating each respondent as a cluster.
I'm trying to eliminate unobserved individual effects using the Fixed Effects method that is common in econometrics analysis - where we demean each observation with the cluster's average. My question is how can I do this in MPlus?
There are two approaches that I can think of: 1. Set up a multilevel analysis. But I don't really have cluster-level data (as we are focusing on the respondent-agent relationship), and does this help to eliminate unobserved individual effects?
2. Do a group-mean centering for all variables, and demean each observation with its group-mean. Run the analysis using the demean variables.
Does any of these sound right? Or is there a better way?
With only 2 agents within respondent it is like having twin data where seldom a multilevel approach is used but instead a multivariate approach. So you have n observations and you have 2 DVs (eval of agent 1 and eval of agent 2).
I don't know about the de-meaning approach. A paper talking about Fixed effects approaches and SEM is: