When math geeks and English majors mingle, you end up with linguists. We don't make anything clearer or better. Also, we don't try to shoehorn language into some "perfect" state, nor do we regard language change as any kind of loss of communicative capacity.
I think that the issue here is partially one of assuming that dictionaries are sources of complete usage information. In actuality most non-OED dictionaries simply aim to provide a word sense (and this is not a knock on lexicographers, who face a daunting task). While I don't think that "decline" or "deterioration" require any idea of return to a previous state, "regress" does have this requirement. In baseball terms, think of career trajectory. As a player gets older, his skills may decline or deteriorate, but that does not mean that they pass through former skill levels. Some skills or abilities are simply lost, and a new less-productive state is created. When a player regresses, however, this is understood as returning to a previous performance level. Something cannot regress to a state that has not existed. In the statistical sense, the state may not have been actualized, but is shown through analysis-- an empirically expected state. A decline in Eduardo Rodriguez's production is not expected based on any observable grounds. For some it may be expected based on some pessimistic subjective grounds (e.g. "young pitchers are never as good as they look in small rookie samples."). I think that the willingness to accept an uninformed (no offense) opinion as a possible state is the source of the offending usage of regress/regression. And if that is true, then smas' point about people's context being "what *I* think" is on point.
tl;dr: "Regression" entails a return to a previous or expected state. Traditionally, "expected" entails "based on empirical evidence". Some people accept their impressions of things as empirical evidence.