In the slides for today, variance for latent factors was fixed to 1 to make them identified. On Hermine’s day, the twin model fixed coefficients to 1 instead of variances to 1.
In general, how do people decide what to fix to make a latent variable identified?
Replying to “In the slides for today, variance for latent facto…”:
99% of time they will give the same answer - 1% the variance component will go negative and you can’t see this if you fix the variance
Replying to “In the slides for today, variance for latent facto…”:
Oftentimes in genomic SEM, the magnitude of variance isn’t so important, and so fixing the latent factor variance to 1 allows you to interpret each of the factor loadings. But in twin models, where the variance of a factor can be often be concretely interpreted (as for example h2), you might prefer to fix a factor loading to 1 instead.