|
|
Scaling mixed continuous/categorical ... |
|
Message/Author |
|
|
Hello, I am hoping you can help me. I am trying to run a multigroup SEM with 4 latent factors and 11 observed indicator variables (5, 2, 2, 2 per factor). All of the indicators are continuous, with the exception of a dichotomous variable that measures a two-indicator factor with a continuous covariate. I have noticed that there is, not surprisingly, a large difference between the size of the variances of the dichotomous variable (0.15), and the two continuous variables for this factor (41.8 for the endogenous variable, and 109.5 for the covariate). I was wondering whether it makes sense to rescale variances when a factor combines categorical and continuous variables. In my very limited understanding polyserial correlation matrices are substituted for covariance matrices in instances in which associations between dichotomous and continuous variables are estimated, so scale differences should be less of an issue? Thank you in advance for any guidance you can give on this. Jonathan |
|
|
It is not necessary to rescale the continuous variables unless their variances are very large. We recommend keeping the variances of continuous variables between one and ten. You can rescale using the DEFINE command to divide the variables by a constant if needed. |
|
|
Thank you Linda. From what I understand you are saying, it is not the ratio of variances that is important but their absolute size? Also, since the variances of the two continuous variables in my data are both above 10, are you suggesting that they should be rescaled? I should have mentioned in the previous post that I am using the WLSMV estimator, with a total sample size of 615. Thanks, Jonathan |
|
|
If they are not far above 10, you probably won't have problems. But if they are, I would rescale. It is especially important when you have a combination of categorical and continuous variables. |
|
Back to top |
|
|