Modeling intraindividual change over time in the absence of a “Gold Standard”

Abstract

Looking at intra-individual change over time in a particular phenomenon may present some methodological challenges. The aim of this report was: 1. To show the effect of independent classification errors on the estimation of incidence and remission rates. 2. To show how a logitbased time-specific latent variables model can be used to model two distinct components of intraindividual change over time in the absence of a "gold standard", namely: (a) the continuity and discontinuity in the latent states over time; and (b) the strength of the association between the time-specific latent variables. 3. To illustrate this model using data on physical aggression from a representative sample of Canadian children assessed at 8-9 years of age and then again two years later at 10-11 years of age. The results showed that classification errors can yield either gross under or over estimates of the true incidence and remission rates. Furthermore, remission was far more sensitive than incidence to classification errors whereas incidence varied more drastically than remission depending on the amount of classification errors. We found that there was no association in the region off the main diagonal of the transition probability matrix beyond that expected by chance alone. In general, the stability of a 8-9 year-old child's latent physical aggression status (i.e., low-, medium- or high-aggressive) did not depend on its severity. Furthermore, the likelihood of changing from one latent physical aggression status to another was generally equal to the one of changing from the latter to the former.