On Effect Size Measures for Nested Measurement Models.

A class of effect size indices are discussed that evaluate the degree to which two nested confirmatory factor analysis models differ from each other in terms of fit to a set of observed variables. These descriptive effect measures can be used to quantify the impact of parameter restrictions imposed in an initially considered model and are free from an explicit relationship to sample size. The described indices represent the extent to which respective linear combinations of the proportions of explained variance in the manifest variables are changed as a result of introducing the constraints. The indices reflect corresponding aspects of the impact of the restrictions and are independent of their statistical significance or lack thereof. The discussed effect size measures are readily point and interval estimated, using popular software, and their application is illustrated with numerical examples.

Bayesian constraint relaxation.

Prior information often takes the form of parameter constraints. Bayesian methods include such information through prior distributions having constrained support. By using posterior sampling algorithms, one can quantify uncertainty without relying on asymptotic approximations. However, sharply constrained priors are not necessary in some settings and tend to limit modelling scope to a narrow set of distributions that are tractable computationally. We propose to replace the sharp indicator function of the constraint with an exponential kernel, thereby creating a close-to-constrained neighbourhood within the Euclidean space in which the constrained subspace is embedded. This kernel decays with distance from the constrained space at a rate depending on a relaxation hyperparameter. By avoiding the sharp constraint, we enable use of off-the-shelf posterior sampling algorithms, such as Hamiltonian Monte Carlo, facilitating automatic computation in a broad range of models. We study the constrained and relaxed distributions under multiple settings and theoretically quantify their differences. Application of the method is illustrated through several novel modelling examples.

Minkowski-Weyl Priors for Models With Parameter Constraints: An Analysis of the BioCycle Study.

We propose a general framework for performing full Bayesian analysis under linear inequality parameter constraints. The proposal is motivated by the BioCycle Study, a large cohort study of hormone levels of healthy women where certain well-established linear inequality constraints on the log-hormone levels should be accounted for in the statistical inferential procedure. Based on the Minkowski-Weyl decomposition of polyhedral regions, we propose a class of priors that are fully supported on the parameter space with linear inequality constraints, and we fit a Bayesian linear mixed model to the BioCycle data using such a prior. We observe positive associations between estrogen and progesterone levels and F2-isoprostanes, a marker for oxidative stress. These findings are of particular interest to reproductive epidemiologists.