When investigating dimensionality using Rasch analysis, a log-likelihood test may
be performed to determine statistically whether one or more dimensions in fact
characterize a dataset, thereby determining whether multi-dimensionality is
supported empirically. Within this test, so-called final deviance of each model is
compared. Final deviance is a measure indicating the likelihood of the observed
data fitting the assumptions of the estimated model. A smaller likelihood value indicates
a better fit. Comparing the efficacy of two models therefore requires comparing
their final deviances. Since the deviances, and, thus, their differences are χ2-
distributed, a comparison to a critical value in a χ2-distribution indicates if this
difference is in fact significant. Degrees of freedom are determined by the difference
of the number of parameters that are estimated. If such a log-likelihood test reveals
a more than uni-dimensional structure, correlations between those latent dimensions
can determine whether students’ abilities with respect to these dimensions are
parallel, antiparallel, or independent.