Testing For Individual Effects Research Articles

Consistency of Bayes factor estimates in Bayesian analysis of variance.

Factorial designs lend themselves to a variety of analyses with Bayesian methodology. The de facto standard is Bayesian analysis of variance (ANOVA) with Monte Carlo integration. Alternative, and readily available methods, are Bayesian ANOVA with Laplace approximation as well as Bayesian t tests for individual effects. This simulation study compared the three approaches regarding ordinal and metric agreement of the resulting Bayes factors for a 2 × 2 mixed design. Simulation results indicate remarkable disagreement of the three methods in certain cases, particularly when effect sizes are small and studies include small sample sizes. Findings further replicate and extend previous observations of substantial variability of ANOVAs with Monte Carlo integration across different runs of one and the same analysis. These observations showcase important limitations of current implementations of Bayesian ANOVA. Researchers should be mindful of these limitations when interpreting corresponding analyses, ideally applying multiple approaches to establish converging results. (PsycInfo Database Record (c) 2024 APA, all rights reserved).

Moment-based tests for individual and time effects in panel data models

This paper proposes two Hausman-type tests respectively for individual and time effects in a two-way error component regression model by comparing estimators of the variance of the idiosyncratic error at different robust levels. They are both robust to the presence of the other effect, and the test for the individual effect has a larger asymptotic power than the corresponding ANOVA F test when the effects are correlated with covariates. Tests jointly for both effects are also discussed. Monte Carlo evidence shows their good size properties and better power properties than competing tests, and the application to the crime rate study gives further support.

Testing serial correlation in semiparametric panel data models

We propose three test statistics for testing serial correlation in a semiparametric partially linear panel data model that could allow lagged dependent variables as explanatory variables. The first is for testing zero first-order serial correlation, the second for testing higher-order serial correlations and the third testing for individual effects. The test statistics are shown to have asymptotic normal or chi-square distributions under the null hypothesis of a martingale difference error process. We conduct some Monte Carlo experiments to examine the finite sample performances of the proposed tests. We also discuss the generalization to testing serial correlation in a nonparametric framework.

Testing for individual effects in autoregressive models

This note describes a test for the presence of individual effects in dynamic models. The test employs restrictions on sample moments implied by the absence of individual effects. Testing the assumption of unobserved heterogeneity is desirable because estimation methods which control for heterogeneity are typically more complicated and require more data than would be required without individual effects. The test is easily implemented as the calculation of statistics follows directly from a linear, instrumental variables estimation technique. The test is illustrated in the context of estimating a dynamic wage equation using data from the Panel Study of Income Dynamics.