Recently, Bill Simmons wrote about Joe Flacco, a so-so regular season quarterback who played poorly the first seven playoff games of his career, then had a seven-game stretch of extraordinary playoff performances over several years, the last of which came right before the article was published.
“For most of the history of modern sports – say, 1950 through 2005 – we would have raved that Joe Flacco was ‘clutch’ and left it at that. Ten years later, it’s much riskier to spit out a blanket statement without offering additional evidence…We’re too sophisticated about sports, mainly because of a relatively recent infatuation with advanced numbers and unconventional metrics….[Now] you can’t just say ‘Joe Flacco is clutch’…Joe Flacco has played a series of games that have made it seem like he’s clutch, but that doesn’t mean he’s clutch.”
Simmons is questioning the skeptics, critiquing the critiques. Why can’t we say Flacco is clutch? If he’s played a series of games that make it seem like he’s clutch, how is that different from saying he’s clutch? Does “clutch” mean anything more than that a guy plays that way, an observable quality? Or is it a “thing in itself,” inaccessible through observation?
Of course if the stats crowd wants to say “clutchness” is by definition an essential quality of an athlete, a personality trait of the kind they always ridicule, not subject to empirical observation, then it is impossible to confirm by their methods- but then impossible to debunk, too. If on the other hand it is something subject to observation, then the difference between “being clutch” and “playing in a series of games that make it seem like he’s clutch” collapses. (A middle range option is those studies that show that clutchness doesn’t exist, because the variance in playoff performance is on average just normal variance. That might tell us clutchness doesn’t exist in the aggregate, but does it tell us Joe Flacco isn’t clutch?)
On the other hand, Simmons’ colleague Bill Barnwell raises the question of predictive value. Flacco’s first seven playoff performances told us nothing about what would happen in his next seven, and “I’m not sure I believe that Games 8-14 can tell us very much about what he’ll do over his next seven playoff contests, either.” The first seven games and the last seven even out, to where Flacco is around his regular season average. (Or maybe Flacco got more clutch over time?)
But Simmons’ argument doesn’t care about predictive value. He assumes what has actually happened, what we observe, is more real or matters more than what we can predict might happen, the underlying qualities we deduce behind it.