i made a goof. i incorrectly filtered out some plays on my last plot that skewed the results. here is a better set. for reference, wpa is contextual and considers every play, e.g. scoring a touchdown when you're tied increases your odds of winning more than scoring when you're down by 69. epa, however, does not care about context, e.g. scoring a touchdown always adds the same expected value. to reduce systematic error due to game situations, epa only considers plays where the score is close (within 2 scores) and time is not a factor (more than 2 minutes remaining).
in other words, epa is "how good are you at moving/stopping the handegg in typical situations," and wpa is "how good are you at improving your odds of winning handegg games."
why bother with two metrics? well, let's say patrick mahomes throws an interception to end a game against the bills (yikes!). wpa is sensitive to the fact that this interception ended the game, so it penalizes this pic more than one thrown in, say, the first quarter of the game. but is mahomes (or any quarterback) really more
likely to throw a pic in the 1st quarter than the 4th? if you think so, then wpa is your stat. if, however, you think player performance does not drastically change with respect to situation, then you wanna look to epa.
