Title: Applying Pairwise Likelihood within Structural Equation Modeling
with ordinal variables using lavaan

Author: Myrsini Katsikatsou

Affiliation: London School of Economics

Abstract: Pairwise likelihood (PL) estimation has been developed for
Structural Equation Models (SEMs) to overcome the computational problems
faced in the case of maximum likelihood and three-stage least squares
methods.  Pairwise likelihood keeps the computational complexity low
regardless of the model size (i.e. the number of observed or latent
variables) and at the same time shares some of the desired properties of
likelihood methods such as the derived estimator is asymptotically
unbiased, consistent and normally distributed.  Moreover, AIC and BIC model
selection criteria can be extended to the case of PL. After a brief
presentation of the background, it will be demonstrated how a SEM with
ordinal variables can be fitted using PL in the R package lavaan and what
inference tools accompany the method.  In particular, so far, the following
have been implemented in lavaan: Wald test for the parameter estimates,
pairwise likelihood ratio test (PLRT) for overall fit, nested models, and
equality constraints, and PL versions of AIC and BIC criteria.  Finally,
looking ahead, it will be discussed how PL can be employed in the case of
data with missing values.