Hi Friends,
I want to do a CFA for the latent variable of inclination for voting. I have the data about whether he or she voted in 04 and 08 presidential election and 05/06 local elections or 09-11 local elections. I held that these voting behaviors are highly correlated, so I chose sem rather than gsem even though all the variables are binary variables. To account for the nonnormality distribution of these variables and possibly even the latent variable of inclination for voting, I used the asymptotic distribution-free method to do the calculation.
When I ran the command, sem politic-> lelection56 lelection911 pelection04 pelection08,latent(politic) method(adf).
I suspect that the not full rank is caused by the high correlation between Pelection04 and lelection56.
Lelec~56 lele~911 pelec~04 pelec~08
lelection56 1.0000
lelection911 0.3029 1.0000
pelection04 0.8174 0.2888 1.0000
pelection08 0.3889 0.4789 0.4125 1.0000.
The above command worked fine when I changed the method to QML which only requires the normal distribution of latent variable. After checking the modification indexes, I finally settled on the below command:
sem politic-> lelection56 lelection911 pelection04 pelection08,latent(politic) vce(robust) cov(e.lelection56*e.pelection04 e.lelection911*e.pelection08)
and got
Is it my solution to the problem appropriate?.
I want to do a CFA for the latent variable of inclination for voting. I have the data about whether he or she voted in 04 and 08 presidential election and 05/06 local elections or 09-11 local elections. I held that these voting behaviors are highly correlated, so I chose sem rather than gsem even though all the variables are binary variables. To account for the nonnormality distribution of these variables and possibly even the latent variable of inclination for voting, I used the asymptotic distribution-free method to do the calculation.
When I ran the command, sem politic-> lelection56 lelection911 pelection04 pelection08,latent(politic) method(adf).
I suspect that the not full rank is caused by the high correlation between Pelection04 and lelection56.
Lelec~56 lele~911 pelec~04 pelec~08
lelection56 1.0000
lelection911 0.3029 1.0000
pelection04 0.8174 0.2888 1.0000
pelection08 0.3889 0.4789 0.4125 1.0000.
The above command worked fine when I changed the method to QML which only requires the normal distribution of latent variable. After checking the modification indexes, I finally settled on the below command:
sem politic-> lelection56 lelection911 pelection04 pelection08,latent(politic) vce(robust) cov(e.lelection56*e.pelection04 e.lelection911*e.pelection08)
and got
Is it my solution to the problem appropriate?.
Comment