FWIW, Pro-football-reference.com has Carter's defensive snaps at 48%... not 25%.
Link...
https://www.pro-football-reference.com/players/C/CartJa05.htm
You're correct, of course, and I've gone and edited
the original comment and added a note about the edit.
I'm
really embarrassed I posted that without checking the numbers first. That's shameful, especially for somebody with as much academic background as I have.
Carter has played about 48% of the Eagles' defensive snaps so far, and consistently so - his snap counts have not been below 45% or above 50% in any of his four NFL games so far.
My point still stands with the correct numbers, though. If they both stay healthy, Witherspoon can be expected to get on the field for a large fraction of the Seahawks' defensive snap counts (it's been 100% through three games, but he did miss a game), while Carter should be expected to participate in about half the Eagle's defensive snap counts. A player who's on the field twice as much has more opportunities to affect the outcome of the game.
This has gotten me thinking a lot about valuations of players in the draft. One factor that I ought to consider more in comparing different potential draft picks is how much a given player will actually be able to get on the field and contribute, and specifically, the part of that that comes from potential percentages of team snap counts. I suspect that if you were to use a statistical model ("machine learning" if you prefer a catchphrase for the same thing) to extract how different teams' front offices have been measuring the value of potential draft picks, there are probably teams that don't properly take into account this detail of positional value (or others), and there are very likely to be arbitrage opportunities (OK,
StatArb opportunities) that arise from that. There are good ways to work around that, by using expectation values of metrics that attempt to measure total on-the-field value, like DYAR, AV, or EPA, but I'm wondering if there might be ways to extract more details about exactly how a given team might be expected to undervalue or overvalue certain positions, and if there are teams that might not be giving enough weight to expected snap counts.