Improved Latent Variable-based Outcomes for Subsequent Regression Analysis
 
Karen Bandeen-Roche and Janne Petersen
Professor of Biostatistics and Medicine and Ph.D. Candidate, Department of Biostatistics
Johns Hopkins Bloomberg School of Public Health and University of Copenhagen
 
Latent variable models have long been utilized by behavioral scientists to summarize 
constructs that are represented by multiple measured variables or are difficult to 
measure, such as health practices and psychiatric syndromes.  They have been regarded 
as particularly useful when variables that can be measured are highly imperfect 
surrogates for the construct of inferential interest, but among numerous criticisms, 
they are also criticized as being overly abstract and computationally intensive.  
My poster describes a new strategy for developing latent measurement model-based "indices" 
for subsequent use in regression modeling that, unlike most existing strategies, 
yields approximately unbiased estimators for regression parameters vis a vis full 
latent variable regression.  Small sample performance properties are evaluated.  
The methods are illustrated using data on vision and adverse functioning in older 
adults.  It is hoped that, by counter-balancing strengths and weaknesses of latent 
variable modeling, the findings will improve the utility of latent variable-based 
approaches for scientific investigations.