Message/Author |
|
|
Hi Linda, I would like to know whether we can control for 'clustered standard errors' in the analysis of SEM and how to specify this in the model. The respondents are based in different locations and it is possible that some contextual factors could affect employee attitudes and outcomes in particular location. I have the outlet identification for each respondent. Thanks. |
|
|
Nonindependence of observations due to clustering can be handled by TYPE=COMPLEX or TYPE=TWOLEVEL. See the introduction to Chapter 9 for a discussion of these two methods. |
|
|
Hi Linda, Thanks for the advise. It is a one level analysis and so I used TYPE = COMPLEX and specified STRATIFICATION = outlet. I however got different warning messages from two different approaches in identifying the observations. Both seem to give the same results in both fit statistics and path coefficients. I'd like to know which one is more appropriate for this type of analysis. Please advise. Thanks. I got this warning message when I specified USEOBSERVATIONS to identify the group of respondents that I want to include in the analysis: When a subpopulation is analyzed with TYPE=COMPLEX, standard errors may be incorrect. Use the SUBPOPULATION option instead of the USEOBSERVATIONS option to obtain correct standard errors. When I changed the command from USEOBSERVATIONS to SUBPOPULATION, I got the following warning message: WARNING: THE VARIANCE CONTRIBUTION FROM A STRATUM WITH A SINGLE CLUSTER (PSU) IS BASED ON THE DIFFERENCE BETWEEN THE SINGLE CLUSTER VALUE AND THE OVERALL CLUSTER MEAN. |
|
|
You should use SUBPOPULATION. The warning simply gives you information about how a single cluster is handled. |
|
Back to top |