In a recently published article from this month's American Naturalist, Vincent Careau and colleagues (2010) propose a new "pace-of-life" hypothesis for the evolution of behavior / life-history relationships among breeds of dogs. This hypothesis relies on various among-breed correlations, including a strong negative relationship between "trainability" (measured as a combination of success in obedience training and ease of house-breaking) and mass-adjusted mortality (obtained, astonishingly, from a dataset of over 222,000 doggy life insurance policies - originally reported on by Bonnett et al. 1997). These data are shown (with a little post-production illustrative embellishment) above.
The authors speculate that the strong relationship between pace-of-life and longevity has resulted from antagonistic pleiotropy between artificially selected traits and life history; rather than from correlated artificial selection. This certainly makes sense in some cases. For example, it seems unlikely that dog breeders directly selected for high mortality in their lines. The ultimate source of several other among-breed correlations found by the authors is less clear, however. For instance, it is somewhat more plausible that humans may have intentionally or unintentionally selected for the observed among-breed negative correlation between body mass and activity level. In this case, the authors advance the possibility that highly active large dogs may have been selected against, because high activity would become increasingly undesirable (and destructive) in large dogs.
Whatever else we might learn from this article, it should dispel any doubt that the classic Billy Joel mantra of "only the good die young" evidently does not apply to our canine friends.