Section 3: Variance (Part 3) and Results-Oriented Thinking
When I see a job listing that says they are looking for “results oriented individuals,” I smile to myself, because in the gambling world being “results oriented” is an unattractive quality. In sports betting and poker (not to mention in stocks, real estate, currency or any other kind of investing), the singular goal is to continually put your money ‘in good’ day after day, and to earn a positive return in the very long run. In doing so, the bettor must avoid the ex-post facto rationalization of events that surrounds him and focus only on the season or multi season goals on hand, and not on the game-to-game variance of wins and losses.
Day in and day out, talking heads try to attribute causality to things that are simply random. For example, they may declare that a certain team “just didn’t want it enough” when they fumbled on the final drive, or that another team “was just more prepared to win” because a breeze carried their game-winning kick a foot inside the upright. Never mind that it’s been proven that fumble recoveries are almost completely random, or that that very same writer probably would have lamented that the team “just didn’t come to play today” if the breeze had carried the kick a few feet in the other direction.
Sports writers and announcers (I will refrain from calling them ‘analysts’) make their living by spouting results-oriented drivel. They second guess clearly +EV (positive expected value) coaching decisions that didn’t happen to work out, praise -EV decisions that get lucky (“Les Miles, the riverboat gambler does it again!” ) and generally ignore really boneheaded decisions if they don’t end up affecting the game, like when Mike Leach elected to kick an extra point instead of going for two after scoring a touchdown to go up by 5 against Texas with 1 second left, and facing two penalties on the following kickoff.
Do not let the outcome of one game fool you – there is so much luck involved in an individual game that trying to draw conclusions from it would be a fruitless exercise. For example, suppose a team that’s favored by 7 ends up getting really unlucky and losing by 25. Now, an outcome like that is rare, perhaps it would happen only once out of fifty games or so. Is an outcome that unusual enough to throw all of our prior analysis out the window? Of course not! There are about fifty NCAA football games per week, so you would expect around one outcome this unusual every single Saturday!
Now this certainly does not mean that I don’t reevaluate my data constantly as soon as I get new information – what it does mean is that I bet on so many games, I don’t overreact when once or twice per week a team radically exceeds or falls short of its expectation. If a junior running back who is a career 4.1 ypc player (adjusted) comes out in his first start and runs for 8.8 ypc against an average defense, I won’t automatically assume that he’s turned into an All-American in the off-season. Instead, I would examine him in the context of dozens of other running backs in similar situations (backup to starter, returning linemen, returning QB, same or different coach etc.) in past years, see how likely it is that he’s become radically more productive in off-season workouts, or that he just had a very above-expectation game. I would probably bump up his anticipated production very slightly, and then continue to evaluate him weekly over the rest of the season to see whether he experiences any regression to the mean before drawing any firm conclusions.
For example, Vanderbilt started the 2008 season 5-0 and vaulted to a top-15 ranking in both the polls after two close SEC wins over Ole’ Miss and Auburn by a total of 7 points. Results-oriented thinkers proclaimed that “Vanderbilt has learned how to win close games,” or “Vanderbilt is one of the top 15 teams in the country because they come through in the clutch.” Yet analysts who understood statistics (not to mention Vanderbilt’s underwhelming history) pointed to the fact that despite playing a very mediocre slate of opponents, Vandy had actually been outgained 5.06 yards per play to 4.59 yppl in those 5 victories and was about even or worse in every predictive statistic. They understood that if Vanderbilt played those five teams a thousand times, they would probably win slightly less than half of their games, yet in this particular 5 game sample they ran incredibly well on turnovers (15 TOs by opponents vs. 4 TOs by Vandy) and went 5-0. Intelligent analysts realized that while Vanderbilt was probably a little better than they had initially been projected, they were nowhere near as good as their 5-0 record made them look. What happened? Luck and turnovers were about even for the next seven games, and Vanderbilt went 1-6 straight up and 2-5 against the spread to close out the regular season. I take advantage of overreactions to short term results that are based on variance by thoroughly examining how every team is actually performing rather than simply looking at scores and wins and losses.
In sports betting, results matter – long term results, with season after season of data and proven mathematical methods for choosing winners. Short term results, the score of a particular game, and the subject matter of the hundreds of hours of programming that ESPN must fill between Saturdays are of vastly overstated importance, and I encourage you to avoid being fooled by randomness.