DOI: 10.5176/2251-1911_CMCGS13.38
Authors: S. INTARAPAK, R. SUWANDECHOCHAI, T. SUPAPAKORN
Abstract:
The nested error regression model is widely used for the analysis of clustered data. In this paper, we show that the generalized least squares F-test is appropriate for the nested error regression model because it performs controlling the size of the test and the power of generalized least squares F-test is an increasing function in terms of the intra-cluster correlation. For small sample sizes, the generalized least squares F-test using Fuller–Battese transformation can control the type I error rate, but the generalized least squares F-test using Helmert transformation leads to the inflated size since the intra-cluster correlations are disregarded. This suggests that the transformation for generalized least squares method should not ignore the intra-cluster correlation. Furthermore, the modification F-test and the ordinary least square F-test are mentioned. Clearly, the latter does not perform well because of the assumption of the error term. The modification F-test works as well as the generalized least squares F-test using Fuller–Battese transformation but the power is the decreasing function of intra-cluster correlation.
Keywords: generalized least squares F-test; controlling the size; power of the test
Updating...