Computation of individual latent variable scores from data with multiple missingness patterns

Latent variable models are used in biological and social sciences to investigate characteristics that are not directly measurable. The generation of individual scores of latent variables can simplify subsequent analyses. However, missing measurements in real data complicate the calculation of scores. Missing observations also result in different latent variable scores having different degrees of accuracy which should be taken into account in subsequent analyses. This manuscript presents a publicly available software tool that addresses both these problems, using as an example a dataset consisting of multiple ratings for ADHD symptomatology in children. The program computes latent variable scores with accompanying accuracy indices, under a 'user-specified' structural equation model, in data with missing data patterns. Since structural equation models encompass factor models, it can also be used for calculating factor scores. The program, documentation and a tutorial, containing worked examples and specimen input and output files, is available at http://statgen.iop.kcl.ac.uk/lsc .