Likelihood ratio test of unconditional and conditional SEM

[Niamh Doherty]

I am trying to see which SEM model is the best model for my data

The first is unconditional:

sem (intercept@1 lin_slope@0 -> lifesat33) ///
(intercept@1 lin_slope@9 -> lifesat42) ///
(intercept@1 lin_slope@13 -> lifesat46) ///
(intercept@1 lin_slope@17 -> lifesat50), ///
noconstant latent(intercept lin_slope) mean(intercept lin_slope) ///
method(mlmv)

Where lifesat* is life satisfaction.

The second is conditional on gender:

sem (intercept@1 lin_slope@0 -> lifesat33) ///
(intercept@1 lin_slope@9 -> lifesat42) ///
(intercept@1 lin_slope@13 -> lifesat46) ///
(intercept@1 lin_slope@17 -> lifesat50) ///
(female _cons -> intercept lin_slope), ///
noconstant latent(intercept lin_slope) ///
method(mlmv)

I want to run a likelihood ratio test on them to see which most accurate but when I do there are different numbers of observations so I am unable. The unconditional model has 13,073 observations and the conditional model has 18558 observations.

Can someone explain why the conditional model has more observations? I assumed it would have been the other way round.
Is there anyway I can get them to the same number of observations to run an LR test?

I'm new to SEM and any help would be greatly appreciated!