Predicting NBA Scores
by Herb Ilk

Last fall I decided to put my computer programming skills together with the NBA statistical data that I had and come up with a formula to predict the final score of an NBA game. I wasn't necessarily interested in seeing how close I could come to the exact game score, but instead I wanted to see how often I could pick the correct winner against the spread. Let me first start by stating that I am not much of a gambler, and the only times I have placed any bets on an NBA game were my few trips to Las Vegas. But I am an avid sports fan and using a statistical analysis to beat the sports books has always intrigued me.

After sizing up the major sports I realized that NBA games would provide the best opportunity to apply a statistical formula for predicting a final score. Basketball games are more repetitive than any other sport and this factor alone makes them a prime candidate for analysis. Each team has possession of the ball over fifty times a game, so if team A's offense is superior than team B's defense then they have fifty chances to prove it. Its like calculating the percentage of times a flipped coin will come up heads. If you only flip the coin ten times then you aren't going to get as accurate a result than if you flipped it fifty times. I believe that this is also the main reason why there never seem to be as many big upsets in the NBA playoffs like there are in baseball and football.

You're probably thinking to yourself "If its that easy to predict a final score, then why don't the sportsbooks use a statistical method and always get the correct line" The answer is that maybe they do, but the sportsbooks aren't trying to guess the actual point spread of the game. Instead, they are trying to set a line so that they will have half of the money wagered on one team and half of the money wagered on the other team. That way they come out ahead no matter which team wins. Using a statistical formula won't allow you to bet on every game, but it will give you a clear understanding of which games are good bets.

Anyway, the predictions are posted every morning on Coverwire.com. I've also posted some additional analysis on that page along with the predictions for every game this season. After seeing the results for one season I can see ways to improve the formula, but even now it is winning 54% of the time overall and almost 60% of the time when the predicted final score is +/- 4.5 points different from the spread. This summer I'll make some changes to it and run the new version side by side with the old version during the 2003-04 season.