Handling plausible values in lavaan

179 views
Skip to first unread message

t bind

unread,
Oct 23, 2017, 10:38:44 PM10/23/17
to lavaan
Hello,
I've conducted an IRT analyis with multiple imputation to get some plausible values for 3 latent traits. Now I want to use lavaan for a linear regression, with the latent traits as independent variables and some Grades as a dependent variable. I have drawn 5 plausible values for each trait. If the corrosponding PVs are named the same for every drawing, can lavaan handle this automatically? In other word: If the PV name for the first latent trait is the same within als 5 drawings for this trait, will lavaan perform a mean out of five different regression analyses?

Thank you for your help
Torsten

Terrence Jorgensen

unread,
Oct 25, 2017, 5:05:51 AM10/25/17
to lavaan
It sounds like you have 5 copies of the same data set, where each copy of the data set has different estimated factor scores (plausible values of the latent trait).

You could think of this as a missing data problem.  You have not observed the latent traits, and your IRT analysis has "imputed" them multiple (5) times.  You could therefore save the 5 data sets in a list, and pass them as data to the runMI() function in semTools, which will pool results by averaging parameter estimates across the 5 data sets, as well as pool the average SE per data set with the variability of estimates across 5 data sets.  I would expect this to yield nominal Type I error rates, although it would be less efficient than running a single latent regression model that includes the measurement model for the latent traits and the regression model among those latent traits (and any other variables in your model).

To do the missing-data trick, make sure you have the latest development versions of lavaan and semTools:

install.packages("lavaan", repos="http://www.da.ugent.be", type="source")
# install.packages("devtools") # if necessary, to install semTools below
devtools
::install_github("simsem/semTools/semTools")


Terrence D. Jorgensen
Postdoctoral Researcher, Methods and Statistics
Research Institute for Child Development and Education, the University of Amsterdam

Reply all
Reply to author
Forward
0 new messages