Some Large-Sample Tests for Nonnormality in the Linear Regression Model

摘要: Abstract This study provides conditions under which several well-known and easily computable statistics for testing nonnormality (√b 1, b 2, D, W′, W) can be modified large-sample use in the classical linear regression framework by replacing true stochastic error with least squares residual. Monte Carlo simulations indicate that perform acceptably well even n = 35. None of tests clearly dominates others, although version W′ performs moderate sample sizes, while D omnibus R test, based on joint √b 1 larger samples (n 50, 100). We illustrate normality using a recent empirical Lillard Willis.