Creating new variable that represents the difference between two latent factors

28 views
Skip to first unread message

Jack Bailey

unread,
Mar 14, 2018, 11:30:59 AM3/14/18
to lavaan

The model above is from a published article and attempts to show the direction of causality between economic perceptions and government approval. The observed variables in the path diagram are computed by taking the difference between a variable at time T and time T-1. The problem is that the data are ordinal and there are other variables that we can rely on to help to reduce measurement error. I'd like to recreate the above model but using latent factors instead of observed variables to account for ordinal data and measurement error. The problem is that I can't think of a way to compute the difference between a factor at time T and time T-1. Is this possible?

Mauricio Garnier-Villarreal

unread,
Mar 14, 2018, 4:06:08 PM3/14/18
to lavaan
Yes, you can estimate your factors at each time point. Then you need to establish longitudinal weak and strong invariance. Then you can estimate a latent change score factor

You can see McArdle and Nesselroad longitudinal SEM book or Grimm and Ram growth model book for more details about change score models

For example this is simple model syntax to estimate a change score from observed variables pre and post.

mod_ch_tot <- '
tot_post ~ 1*tot_pre
change =~ 1*tot_post

change ~c(m1)*1
change ~~ c(v1)*change

tot_post ~~ 0*tot_post
tot_post ~0*1 

change ~~ tot_pre

### cohen d
cd1 := m1/sqrt(v1)
'

Jack Bailey

unread,
Mar 15, 2018, 6:21:37 AM3/15/18
to lavaan
Thanks very much, Mauricio.
Reply all
Reply to author
Forward
0 new messages