I have a dilemma as far as the use of latent variables. Are there any rules as to when to use latent variables?
I have child and parent measures of the same constructs and ran separate models for the two informants. However, I was asked by a reviewer to use a latent variable approach by including both informants' measures under one variable.
The correlations between the measures between child and parent report are modest at best (highest is .22). Can low correlations be a reason for not using latent variables? My main interest is to get at the different perspectives that children and parents may have. Also, if there is a good reference to support this, I would appreciate it. Thank you!
My understanding is that the use of different informants often leads to discrepancies - sometimes those discrepancies are of primary interest. Forcing such measures into a single construct may not be the best approach. Perhaps using a latent variable construct for each informant would be useful. I would ask on SEMNET where fellow psychologists might have experiences with this.