The morning after the Fairleigh-Dickinson Knights upset the Purdue Boilermakers as a 23-point underdog, it was commonly reported that the Knights had just a 1.6 percent chance (going into the game) of pulling the upset. And while that number wasn't wrong, per se (if one just looked at the universe of results for 23-point underdogs since 1990), it also severely over-estimated the odds, if one dug deeper into the numbers. And digging deeper is what we will do here.
First, let's take a look at the overall results since 1990.
Going into the Purdue/FDU game, there were 272 games with point spreads at exactly 23, of which four resulted in an upset. Simple math says that four divided by 272 equals 1.47 percent. So, in the big picture, one would be comfortable assigning the Knights roughly a 1.5 percent chance of knocking off #1-seeded Purdue.
But this "big picture" ignores a very important fact. This game occurred late in the season, and in a post-season tournament, to boot. And, from my research, it's significantly harder to pull off a massive upset later in the season. Let's dive into the numbers.
Since 1990, there have been 1,577 games with point spreads between 21.5 and 24 points. Of those, the underdog won outright in 28 of 1,577, or 1.77 percent.
However, 26 of the 28 upsets occurred earlier in the season, at Game 16 or less, while just two upsets occurred at Game 17 forward. So, if one breaks the College Basketball season into two parts, it becomes clear that pulling off a monstrous upset is easier earlier in the season.
Since 1990, there were 1148 games played where a team favored between 21.5 and 24 points was playing its 16th (or less) game of the season. And our favored teams lost 26 of those 1148 games, straight-up (2.26 percent). In contrast, when our favored team was playing its 17th (or greater) game, then there were 429 games. And the favorite (including Purdue vs. FDU) has gone 427-2 straight-up. Indeed, going into the Purdue/FDU game, only ONCE in 428 games (one-fifth of one percent) had a team, favored between 21.5 and 24 points, lost at Game 17 forward. And that occurred on February 19, 2003 when the Portland Pilots went into Gonzaga, and upset the Bulldogs, 72-68, as a 22-point underdog.
For this study, I chose to circumscribe the data set to a point spread range of 21.5 to 24 points, so as to not contaminate the data with the results of (even higher) point spreads that result in much less upsets. But we see the same early/late season dichotomy in games with point spreads north of 24 points. Since 1990, there have been 1,803 games with point spreads at 24.5 (or higher). Of those 1,803 games, there were eight upsets (two-fifths of one percent). And all eight upsets occurred up until Game 16. From Game 1 through Game 16, the record was 8-1492 (one-half of one percent), while at Game 17 forward, it was 0-303 (zero percent).
I believe the primary reason this happens is that the numbers for the oddsmakers/bettors are less sharp earlier in the season, due to the paucity of data. So, teams could be over-valued, or under-valued, as the case may be. And there's a much greater chance for a big underdog to pull off a shocking upset (say, if a 22-point dog should have been a 19.5-point dog). But later in the year, when teams have 15, 25, or (in the case of Purdue and FDU) 35 games under their belt, the teams are pretty much who we think they are. So, you won't have the types of mis-priced games that might happen earlier in the season.
When viewed in this context, Fairleigh-Dickinson's upset of Purdue was the BIGGEST upset in College Basketball history (or at least since 1990, the year my database starts). No other team had ever won as an underdog of more than 22 points, at Game 17 forward!
Good luck, as always,