Tom posted on Wednesday, November 02, 2005 - 7:16 pm
I wonder if multilevel SEM could also be applied to data stemming from repeated cross-sectional designs.
Specifically, I have 15 annually repeated surveys with randomly sampled respondents (--> no panel design). Thus, samples are level 2, individuals are level 1.
The model I would like to compute would then comprise of:
- one dependent level 1 variable Y,
- one independent level 1 variable B1
- two independent level 2 variabls B2 & B3
While B1, B2 and B3 are all assumed to predict Y, also B2 is assumed to predict B3.
As I have never seen before neither such a level-2 mediation (b2 --> b3) nor an application of multilevel SEM for repeated cross-sectional surveys, I wonder if the model above would be feasible at all?
BMuthen posted on Thursday, November 03, 2005 - 12:42 am
Are you thinking that the 15 surveys (years) have characteristics that make the sample respondents of a given year be more like each other than the respondents in the other years? If not, there would be no intraclass correlation and therefore no need for twolevel modeling.
Tom posted on Thursday, November 03, 2005 - 7:57 am
The example above is still hypothetical. But according to some, considering individuals nested in repeated surveys as described might be considered to form a hierachical structure. If this is correct and under the assumption that there would be sufficient ICC in the dependent variable, would the model described in my first question be feasible? Best, Tom