Are there any examples using Mplus with the Youth Risk Behavior Survey (YRBS). How do you include weight factor assigned to each student record (weight), primary sampling units (psu), and stratum (indicates the stratum the school the student was assigned). Are there any examples with this data?
A simpler way to work around this issue is to ignore the cross classification -- presumably the SE won't be underestimated that much. This is done by treating a cluster which is in two strata as two separate clusters.
It may be possible to run this model as a mixture/multiple group/known class model where the strata is the grouping variable. This will allow the cross classification.
I tried your initial suggestion -- treating psu with multiple strata as separate clusters and ran a simple regression model (rather than the LCA model that I'm working on, which takes about 1.5 hrs to estimate). Ignoring stratification and treating psu as the cluster variable (with weighting) produces the following: Estimates S.E. Est./S.E. Q_X ON Q_A 0.017 0.063 0.262 Q_B 0.058 0.044 1.309 Q_C 0.005 0.169 0.029 Q_D 0.045 0.086 0.528 Q_E 0.100 0.120 0.835 Q_F -0.188 0.107 -1.760 Q_G 0.065 0.152 0.430 Q_H 0.092 0.043 2.115
With a revised psu approach I get the following: Q_X ON Q_A 0.017 0.119 0.139 Q_B 0.058 0.065 0.893 Q_C 0.005 0.176 0.028 Q_D 0.045 0.111 0.410 Q_E 0.102 0.061 1.676 Q_F -0.188 0.068 -2.784 Q_G 0.065 0.210 0.310 Q_H 0.091 0.069 1.320
Primary difference is S.E. for Q_F & Q_H, resulting in differences in statistical significance. Would this suggest any concern for either approach (neither of which fully incorporates the strata variable)? Also, can you say more about your second option -- which would represent the group and which the known class?