Inspired by a comment discussion in my post about the effectiveness of head coaches, I decided to resurrect my old research on when upsets happen in the SEC.
I began working on upsets in SEC play years ago based on the hypothesis that clearly better teams were more likely to lose games they shouldn't in the noon hour. Call it the curse of Jefferson Pilot. The last time I reported on this was in 2013, but I decided to redo it based on the "peers" definition I've been using for various things this offseason.
The definition in short: S&P+ rates teams based on how many points better or worse than a perfectly average team is, and if the year-end S&P+ ratings said a team was within a touchdown of another team, I call them peers because they're roughly on the same level.
To be clear: this research on upsets only includes regular season games in SEC play from 2005-15, and "upset" refers to when a team more than a touchdown worse than its opponent wins. This has nothing to do with betting lines.
Now, commenter nottherealmccoy brought up the issue of home field advantage in the head coach effectiveness post, and I explained in a response why I didn't feel a strong need to include an HFA adjustment back then. This time, though, I will go ahead and address HFA.
As an illustration, let's go over the winning percentage for home teams. And, so you know, I made sure not to include the Florida-Georgia and Arkansas-Texas A&M neutral site games in these figures since those didn't have true home and road teams.
Home teams overall, regardless of quality, had a win percentage of .543 in these games.
With no HFA adjustment, home teams had a win percentage of .593 in tossups. Home favorites had a .905 win percentage, and road favorites had a .891 win percentage. This method found that home teams fared better than road teams, as home favorites outpaced road favorites and home teams won the majority of tossups.
I tried again with a blanket HFA adjustment of three points for the home team, as that is the typical rule of thumb that people use. Having done that, the win percentage for home favorites fell to .855 and the win percentage of home teams in tossups fell to .474. Road favorites saw their win percentage rise to .913.
Let's think about what this adjustment did for a second. It moved games where the home team was 4.1 to 7.0 points better from tossups to mismatches, and it moved mismatch games where the road team was 7.1 to 10.0 points better from mismatches to tossups. In that light, all of the moves make sense. However, the figures don't make intuitive sense. Why would road teams win the majority of tossups? Or road favorites fare better than home favorites? This HFA adjustment ironically ends up making road SEC teams look better than home SEC teams.
As I mentioned in the aforementioned comment conversation, I believe home field advantage isn't actually just three points for every team. Vandy's home field advantage is not the same as LSU's is, for instance.
Fortunately, the good folks at Prediction Machine took a stab at computing a specific HFA value for each FBS team last summer. You can click that link if you want details, but it seems about as solid as these things get, and their data covers 2000-14. It's going to be hard to find something better than that.
Using the PMHFA (Prediction Machine home field advantage), I ended up with home teams in tossups having a .481 win percentage, home favorites having a .860 win percentage, and road favorites having a .924 win percentage. Road teams fared worse against expectations in tossups, but they did better against expectations in the mismatches.
Anyway, that's your introduction to the three kinds of methods I used for this upset research. I'll let you decide which method you think is the most accurate. Let's dive in.
The unadjusted numbers found 38 upsets in 341 mismatches, good for an upset rate of 11.1%. That's about one upset in every nine mismatches. It was about evenly split with 16 home favorites losing and 18 road favorites losing.
The HFA-adjusted numbers found 48 upsets in 361 mismatches, good for an upset rate of 13.3%. That's about one upset in every 7.5 mismatches. Home favorites were the disproportionate victims, with them losing 33 times versus 11 road favorites losing.
The PMHFA numbers found 46 upsets in 368 mismatches, good for an upset rate of 12.5%. That's one upset in every eight mismatches. Again, home favorites took many more falls at 32 versus just ten for road favorites.
All three methods thought that the 2007 and 2014 Cocktail Parties and the 2014-15 Arkansas-A&M games ended up as upsets among neutral site contests.
The first grouping of games is the early set where kickoff was between noon and 2:30 Eastern. For simplicity's sake, I'm only going to refer to times in Eastern.
The unadjusted numbers found 15 upsets in 107 mismatches, good for an upset rate of 14.0%. They were about evenly split with eight home favorites losing and seven road teams losing.
The HFA-adjusted numbers found 18 upsets in 109 mismatches, good for an upset rate of 16.5%. Home favorites took double the losses (12) as road ones did (six).
The PMHFA numbers found 16 upsets in 113 mismatches, good for an upset rate of 14.2%. Home favorites again took it on the chin, losing 11 games versus five for road favorites.
With all three methods, the upset rate in midday games was higher than the overall rate. This is the first piece of evidence for the Jefferson Pilot Curse theory.
The line between mid-afternoon and what comes before and after is a subjective one, but for this one I took games that kicked off at 3:00 or later but before 6:00. The vast majority of these are 3:30 kicks, as you would expect. Three of the neutral site upsets came in this time slot, with only '15 Arkansas-A&M being a night contest.
The unadjusted numbers found ten upsets in 81 mismatches, good for a rate of 12.4%. Home favorites lost just one game, while road favorites lost six.
The HFA-adjusted numbers found 11 upsets in 91 mismatches, good for a rate of 12.1%. Home favorites lost six games this time, with road favorites falling twice.
The PMHFA numbers found 11 upsets in 90 mismatches, good for a rate of 12.2%. Again, home favorites lost six games, and road favorites lost two.
In each case, the upset rate is lower in mid-afternoon games than in midday games. The HFA-adjusted set has the largest drop, while the unadjusted set has the smallest drop.
These are the games that kicked off at 6:00 Eastern or later. As mentioned before, one of the neutral site upsets happened during this time slot.
The unadjusted numbers found 11 upsets in 136 mismatches, good for a rate of 8.1%. Home favorites lost six games, while road favorites lost four.
The HFA-adjusted numbers found 15 upsets in 173 mismatches, good for a rate of 8.7%. Home favorites lost 12 times to just two times for road favorites.
The PMHFA numbers found 15 upsets in 144 mismatches, good for a rate of 10.4%. Home favorites lost 12 times, and road favorites lost twice.
In all three cases, the upset rate is lower at night than in the mid-afternoon and much lower than during midday games. The curse of Jefferson Pilot is looking stronger.
I pulled out games on Thursday, Friday, and Monday (the makeup for '05 Tennessee-LSU) into their own category because the main point of this is to look at times of the day on Saturday. The unadjusted numbers found two upsets in 17 mismatches (11.8%), while the two HFA-adjusted methods each found four upsets in 21 mismatches (19.1%).
Does Time of Day Matter?
Each of the three methods agree that upsets are more frequent in midday games than night games:
- Unadjusted: 14.0% at midday vs. 8.1% at night
- HFA-adjusted: 16.5% at midday vs. 8.7% at night
- PMHFA: 14.2% at midday vs. 10.4% at night
That said, the methods disagree about the mid-afternoon games. All three put the rate of upsets in mid-afternoon games between 12.1% and 12.4%, the most agreement of any time slot. However, it shows up differently across the three points of view.
For unadjusted numbers, 12.4% is a slight decrease from 14.0% before the big plunge to 8.1% at night. For HFA numbers, its the midpoint in a steady decline throughout the day from 16 .5% to 12.1% to 8.7%. For the PMHFA figures, it's a midpoint in a less steep curve from 14.2% to 12.2% to 10.4%.
So it does look like yes, time of day matters. Upsets are more common at noon than at night. The effect on mid-afternoon games is less conclusive, though.
What Is Happening Here?
I see three strong possibilities here:
1. Lack of preparedness in midday games.
My working hypothesis has always been that upsets are often—not always, but often—the result of a strong team overlooking a weaker team. Psychologically, it probably would be easier for that to happen earlier in the day because the stronger team had less time to get hyped up for the game. Plus the 3:30 CBS Game of the Week has had a bit of prestige to it, and basically every program in the SEC makes a big deal out of night games.
All three methods report a higher upset rate in cross-division games than contests within the divisions (this is for all games, not just midday). The two methods with HFA adjustments even report higher upset rates among the unfamiliar cross-division contests—for this I threw out games between the five permanent rivalries that haven't changed recently plus Mizzou-A&M and Arkansas-South Carolina games—with the unadjusted numbers having them being the same either way. This is more evidence that a lack of preparedness may play into upset frequency.
Further, some SEC teams are more likely to be overlooked than others. Everyone gives Alabama their best shot, but getting up for Vanderbilt is less easy. Well, all three methods found Mississippi State and Vanderbilt as having a greater rate of winning upsets in midday games than overall, and two of them (not the PMHFA) had Ole Miss winning upsets at a greater rate during the midday period as well. Those three teams plus Kentucky were the only teams to appear as underdogs in midday games more than ten times according to all three methods, and all are from the conference's historical bottom tier. The fact that three of the four won a disproportionate number of upsets at midday lends more credence to noon being the time when favorites are most likely to overlook weaker opponents.
2. Crowds are less into the noon games.
All three methods found that home favorites did worse in midday games than in mid-afternoon or night games. It's a slight difference between midday and night in the HFA-adjusted methods, but it's there in all three.
Why does this matter specifically? The book Scorecasting found that much of home field advantage can be explained by the effect that rowdy and loud fans have on referees rather than on the teams. Sure, we've all seen situations in college football where players appear to be bothered by crowd noise or have to burn timeouts because they couldn't communicate verbally. Those instances are pretty rare, though, and their effect is overwhelmed by the fact that the refs' mindset matters on every single play.
Fans have less time to prepare their minds and livers for noon (and especially 11 am Central) kicks than they do for later ones, so a diminished crowd effect at early games may also help explain the greater upset rate.
3. Nothing at all.
Even though I'm looking at 11 seasons' worth of games, the data set is small because each conference member only plays eight SEC games a year. I have 560 total games to work with, and the system that found the most total mismatches (PMHFA) only identified 368 of them. It found 113 at midday, 90 in the mid-afternoon, and 140 at night. In those first two time slots, a single game being an upset or not will swing the rate by about a percentage point.
Over 70% of the upsets found by each method were decided by seven points or less, while 39.5% of the overall set ended up one-score finishes. I know off the top of my head that two of the unanimous upsets came in overtime—Texas A&M's neutral site wins over Arkansas in 2014 and 2015—and two more came in triple overtime—2007 LSU's two losses. There probably are more that finished in overtime or came down to the last drive, and in those such games, there is room for randomness to play a major part in how the game finishes.
It's entirely possible that the trends I found aren't trends at all and are instead statistical noise. It may also be that my definition for what constitutes a tossup or a mismatch is off and is giving bad results.
But while this information can only go so far, it's not a bad idea for fans of strong teams to brace themselves and fans of weaker teams to hold a little extra hope when noon games arrive on the schedule.