Notes PML Tutorial -- YR 19 April 2020
1) as explained in the tutorial, when thresholds are 'structured'
(eg group.equal includes "thresholds"), then ctr_pml_plrt2() should be
used instead of ctr_pml_plrt() to get the model test statistic.
in < 0.6-6, this was NOT done automatically (as reflected in the tuturial),
but from 0.6-6 onwards, this is done automatically
2) from 0.6-6, the estimates of the polychoric estimates differ somewhat
for some pairs (new implementation!); this affects mostly the PML robust
test statistic
3) on some machines, in particular for the multigroup example, you may get
the warning:
lavaan WARNING: Hessian is not fully symmetric.
Max diff = 8.45073372965089
(Max diag Hessian = 66496.3429486499)
(possibly with different numbers). This is nothing to worry about. It is
a new check (for YR) to evaluate the accuracy of Hessian matrices;
eventually, this warning will disappear again.