The third advantage of the analysis of data using an IRT framework is that it accounts for measurement error. Unreliability suppresses the correlation between measurements (attenuation). Particularly when using a scale with only a few items, the correlations amongst sum score variables may be grossly attenuated. Clearly, this has important implications for the estimation of variance components in genetic research. In an IRT framework, the problem can be solved by, instead of focussing on sum scores, considering the correlations between latent variables (see, for instance, Béguin and Glas 2001; Fox and Glas 2003). These so-called latent correlations can be seen as estimates of correlations corrected for attenuation. A simulation study and an application of IRT to real data in a later section will show the possible extent of such attenuation effects on the estimation of heritability.
