Turnover Difference (part I)
Ah, turnovers. There is little doubt that much of the difficulty in predicting the outcomes of Football games comes from the intervention of the rascally fumble and interception comrades. Every week the gridiron coaches across the land spend a great deal of time trying to impress upon their players the utmost importance of not turning the ball over in the upcoming contest, while at the same time poring over defensive gameplans in an attempt to figure out ways of inducing their foes to do just that. Despite these impassioned pleas, the players will go out and regardless of whether it’s a pee-wee, high school, college or pro game, will often turn the ball over at a critical juncture. The coaches rant and rave, the fallible player/culprit quickly receives "goat status" and the fans and followers go home either despondent or jubilant depending on which side they were rooting for.
Memorable turnovers linger for years in the minds of football fanatics. Perhaps the most notorious "blundering" in NFL history, one any fan must have seen replays of several times over the years and Exhibit A in our later discourse on the "Horrible Defeat Syndrome" is a game matching the New York Giants with the Philadelphia Eagles that took place back in the seventies. The Giants found themselves ahead, with the ball and perhaps forty seconds left on the clock. In those days the kneel-down play was considered somewhat "unsporting" and so the Giants elected to run the basic sure as sure can be hand-off up the middle. Even the biggest longshot lover would not have bothered to ask the line on the Eagles winning at that point, since everyone knew that there was no possible way the Giants could lose. Except for what happened -- which is that there was a screw-up in the transfer of the football, such that it bounced out of the halfback's hands and was picked up off the ground by an Eagle defender, who literally mustered no more than a jog in coasting into the end zone for the score! Anyone watching was left speechless.
It would be unfair to focus solely on the Giants’ misfortunes though, for many clubs have felt the harsh sting of the play that went awry. The Cleveland Browns were stung not only by John Elway and "The Drive", but also by "The Fumble" (Earnest Byner deep in Denver’s red zone). On the other hand, it used to be that anytime you watched a 49er game in their glory days at some point in the broadcast they’d throw up a stat saying that when San Francisco had fewer turnovers than their opponent they had won something like fifty straight. The question posed to us though is how to make good use of these lightning strikes that change the outcome of games?
Well first let’s start with the basics: there are two types of turnovers, fumbles and interceptions. In the NFL typically 3% of passes are intercepted. Fumbles meanwhile can occur on any play. Runs, passes, kickoffs, punts, field goal attempts, even a passing play that is intercepted, run back a distance and then promptly fumbled right back to the offense.
There is no shortage of "blundering" consequently in the NFL and in the most recent season, the average team was turning the ball over 1.9 times per game (1.1 from interceptions, 0.8 by fumbles). What this means is that in a typical NFL contest you’ll see close to four turnovers.
All right you say, but how often do turnovers really matter? Certainly they can be very memorable, since they are clearly in the "big play" domain, but don’t good teams overcome these lapses and go on to win anyway? Well, to try and answer this I set the old Cray computer whirring with a little research project to determine which statistic in a box score is most predictive in telling you who won the game. Obviously if you look at the points scored by each team you will have 100% accuracy, since by definition the team that scores the most points wins. Excluding points I elected to look at six other supposedly key areas that are widely believed to be the most telling -- total yards, first downs, third down conversion rates, time of possession, rushing yards, and turnovers. By using a ten-year period in NFL history we’ll get a large sample and to add an extra twist to the analysis we’ll also track how often the "advantaged" team covered against the spread. Here are the results and winning percentages --
|More Rushing Yards
|Greater Time of Possession
|More Total Yards
|Higher 3rd Down Conversion
|More First Downs
So, turnovers prove worthy to the challenge and come out as far and away the most predictive of the six stats put to the test here. While the team with the most first downs wins 66% of the time, and the yardage champ hits 68%, the turnover winner is also the scoreboard winner in 78% of the games. Of course one less turnover is far more significant than one more yard gained or one more first down, but even comparing the overall turnover stat to breakdowns like "cases where one team had 50+ more total yards" or "cases where one team had 7+ more first downs" the turnovers still are the superior indicator. In more than three out of four games, the team with fewer turnovers will win, regardless of other considerations.
For a bigger shock check out the spread numbers -- Aha! you say, all I need to do is only play teams who'll have less turnovers and I'll be rich! Well yes, but of course it's not always particularly easy to know which side will wind up looking like an efficient Super Bowl caliber squad while their opponents show a passing resemblance to Mr. Magoo...if you knew the turnover numbers in advance, you'd be a 75% wizard against the line.
One other interesting point that came to light was that home field advantage is so significant that even when the visitor outplays the home squad in some domain, the expected win percentage is still less than when a home side has a similar advantage. For instance, when a home side winds up with more total yards on the day there’s a 75% chance the home team won, whereas an away side with more yards will typically only win 60% of the time.
In Part II we’ll look at how to use turnovers to handicap an upcoming game.
Turnover Difference (part II)
Last week we looked at how often a team won when it had an edge in various statistical areas. Surprisingly turnovers were the most predictive of the numbers we looked at, both in terms of who won and also in terms of who covered the spread.
Having established then that turnovers are a vital consideration in any analysis as to which side is likely to come out on top, we need to turn to a point that is fundamentally a matter of belief. Is being good in net turnovers a sign of talent, expert coaching, precision timing, aggressive defensive techniques or is it largely a matter of blind dumb luck?
In 1995 the St. Louis Rams got off to a surprising 4-0 start, and looking at the turnovers you'd see that while the Rams had fourteen takeaways, they had zero giveaways in their first four outings. This charmed life quickly came to an end and they finished the season 7-9 and needless to say, out of the postseason dance party. Score one for dumb luck (their final turnover numbers were 39 giveaways and 36 takeaways for a -3 net).
Of course, you’ll often find the well-coached teams with nifty turnover records and the less together franchises stumbling through year after year of bumbling. Marty Schottenheimer’s Kansas City Chiefs typically ran a conservative offensive scheme in conjunction with a tough defense that was skillful at forcing mistakes. As a result the Chiefs ranked 3rd, 11th, 1st, 2nd, 5th and 1st in net turnovers over a span of six seasons. It would be hard to advocate that such a consistent pattern is merely because of the turnover gods smiling on Arrowhead stadium. However another team worth looking at is the Buffalo Bills during their storied four straight Super Bowl seasons. The Bills ranked 4th in net turnovers in 1990, slipped to 10th in ‘91, dropped again to 18th in ‘92 (with a -3 net), and then were back in the high life as the 2nd best NFL turnover team in their last dance. Here we see a range of #2 to #18 with a consistent coaching staff and base of players. It would be difficult to take the stance that the Bills were a "butterfinger" group in 1992 but then overnight became "surehand Lukes" in 1993, but it’s easy to adopt the contrarian’s mantra that bad things can happen to anybody at anytime.
The bottom line is that getting the best of it in turnovers can make a team look good even when they're not, while some sloppy offensive play or tough luck can quickly turn a coach from "genius descendant of Bill Walsh" into "goat boy." If you're intent on beating the spread, take a good look at the turnover history so far and believe in the Due Theory.
Having dealt with the theory, let's turn to the practice. Since our primary aim is to figure out ways to beat the spread, the question becomes how to formulate a workable strategy for utilizing the turnover data. Fortunately I have a few suggestions! While the raw totals are revealing, the better way to look at the turnover story is on an "average per game" basis. So four games into the season we might find a match-up with Team A being +4 overall in turnovers and Team B slouching along at -6 overall. By comparing the numbers for each team and dividing by the number of games played, you create the magical Net Turnover Difference --the difference between the above teams is 10 (+4 vs -6), which divided by the four games played works out to be two-and-a-half. Expressing this from Team B’s standpoint, we’d say they are -2.5 net turnovers per game worse than Team A.
The obvious approach you would think at this point would be to back the team with the better performance thus far, yet surprisingly this is the wrong strategy. If you decided to play the team with the better net turnover mark coming into the game over the years, you would have been crushed from a betting standpoint. On the other hand, those who played the bad turnover team, would have been rewarded for their optimism with much success in beating the point spread.
Let’s lay down some ground rules for the "Turnover Difference Theory": you want to look for a difference in net turnovers between the two adversaries in a particular clash of more than a half a turnover per game (e.g. after five games, Team A is +4, team B is +1, which translates to a 0.6 difference per game), and look fondly on the poorer team as being due for an imminent turnaround. Do not consider this method for the first four weeks of the season since you need a certain amount of data to base your assumptions on. Heading back to the computer, here’s how you would have done playing this strategy from 1991 to 2000 --
||VS THE SPREAD|
Comparing this to the standards of successful handicapping, where 52.4% is break even, 55% is solid, 57.5% is excellent, and 60%+ is world class, this falls right in the "solid" range. When you consider the simplicity of the strategy, it is remarkable indeed that one factor can be so strong! We are furthermore not dealing with rare events as more than half of the games played will qualify as having one team with net turnovers significantly worse than the other. Without knowledge of won-lost records, yards gained, first downs made, head to head history, injuries, weather or even the line on the game itself you could have outperformed many skilled handicappers over a ten year period! Consistency on a year to year basis is unfortunately not ideal, and following the T/O method blindly from 1991 to 2000 would have produced these year to year efforts: 53%, 62%, 64%, 55%, 67%, 51%, 54%, 55%, 54%, 46%. Still this rates as only two losing season in ten tries. Of course, the troubling thing is that recent years have been less successful, with the 2000 season being horrible. There is a strong possibility that the linemakers and bettors are now adjusting their lines for the turnover effect, and so the 2001 season marks a line in the sand: either the T/O Difference theory bounces back with a great year, or we may need to write it off as a strategy of the past.
The next step people almost invariably take when they stumble upon something good is to try by adding further "qualifiers" to make it even better. This is generally a bad idea since false patterns can be found in any sample through backfitting, but here are a few worthy pointers: Home teams are better plays than visitors, and favorites are likewise a higher percentage than underdogs. The former of these observations seems plausible as horrendously bad turnover-prone teams wouldn’t be expected to fare well on the road -- the unfriendly confines and boisterous crowd do nothing to calm the nerves of an already jittery quarterback, whereas at home there’s a certain familiarity and comfort. The observation that favorites do well also seems of merit as a team that qualifies by the turnover difference method and is favored must have been playing well in spite of committing more than the normal number of blunders. Finally one last tip is that the results in games expected to be close (no line great than 4 ½ points) have been very strong.
All in all, on the scale of predictive value, turnovers rate as Serious Handicapping. While it would be foolish to blindly follow the strategy, it’s very worthwhile as a piece of the puzzle. You may groan as week after week it suggests you back some of the least attractive franchises in the league, the poor huddled masses of the NFL, but take heart in having the percentages on your side. Developing an informed eye towards teams’ turnover statistics should lead to smart wagering decisions for a long time to come.