Partial scalar invariance for ordinal data

[Arash]

Hi there, 

I aim to establish partial scalar invariance using ordinal data in a two-factor model and then compare latent means across groups. Following the method proposed by Wu and Estabrook (2016), I first established threshold invariance and then metric invariance. However, scalar invariance was not achieved, so I now seek to test partial scalar invariance.

I am not sure whether it is permissible to free one or more intercepts to establish partial scalar invariance, given that for ordinal data all intercepts are fixed to zero in all groups for identification so that latent means can be freely estimated. Can one intercept be fixed to zero in the reference group but freely estimated in the other groups? Would this violate the identification rules for ordinal data, or is it acceptable? I attempted to free one intercept, and the model converged successfully.

Thanks,

Arash