Straight-line models in star-shaped mixtures

Normal-theory inferences are validated in part for straight-line models having star-contoured errors. Adverse effects for the intercept include inconsistent estimation and disturbances in levels of the standard tests. Tests for slope remain exact in level; they are unbiased, and their power through mixing typically dominates the standard Gaussian case. Bounds on level, and envelopes for power curves, are given for certain ensembles and mixtures of distributions, and these are evaluated numerically for selected cases. Effects of mixtures on model diagnostics are examined further.