|
|
Large and varied design effects |
|
Message/Author |
|
Bryan Duff posted on Wednesday, March 29, 2006 - 5:56 am
|
|
|
Drs. Muthen, I am running a path analysis (with 1-minus-alpha correction for measurement error) with just over 1300 cases (students). These cases come from 13 schools (clusters), with cluster sizes ranging from about 25 to 180. All variables are measured at the individual level, so I am not running a multilevel model; I just wish to correct for the effects of clustering. The design effects for the variables in the path model range from a low of 2 to a high of about 30. Given these conditions, and given that I am unable to collect further school-level data that might enable a multilevel analysis, is it acceptable to run a type=complex analysis in Mplus? Thanks very much. |
|
|
Even with TYPE=COMPLEX you need a minimum of 20 to 30 clusters for good estimation. |
|
Back to top |
|
|