Models should be specifically formulated to allow selection factors to vary in their effects over time. Extant models allow for unmeasured traits to affect selection, but they rest on the rather unyielding and excessively restrictive assumption that these traits and their effects are time-invariant. Fixed traits are surely important, but they may be only a small subset of the relevant factors for health and mortality differences. Since little is known about the magnitude of errors caused by deviations from the assumption of invariance, it is difficult to evaluate whether choosing an incorrect model for selection may lead to more serious problems than ignoring it altogether. More research is needed on the nature, advantages, and shortcomings of models with time-varying effects (Weitz and Fraser, 2001).