There is an incredible amount of data that exists in sports. Data comes from individual player performance, coaching, in-game events, betting lines, officiating and weather.

We can extract useful information from that data to make predictions about future outcomes. However, one of the most common mistakes made by bettors is making definitive claims from small samples of data.

Small Sample Size

You’ve probably heard someone cite a stat and then another person say it doesn’t matter because of the sample size. It is pretty easy to understand what small sample size means.

For example, say the Patriots have covered in 9 of their last 10 games as touchdown favorites. Does New England going 9-1 against-the-spread tell us anything about what will happen in the future? No. A sample of ten games is not predictive of future events.

The size of the sample, the total number of bets, helps determine the level of confidence we have in our systems. In other words, the larger the sample the more a system can be trusted.

How Big a Sample Do We Need?

When are the samples considered big enough and at what point does the data become meaningful to make predictions?

Using math we can determine the appropriate sample size to consider a betting system worthy of wagering. Below is a graph that shows the winning percentage needed to prove statistical significance versus sample size. For the formulas to calculate sample size and to learn more about confidence intervals and confidence levels click here.

screen-shot-2016-12-09-at-3-25-26-pm

Graph courtesy of SportsInsights.com

If you have a betting system with a win rate of 57% and there are 2,000 or more games in the sample, then we can say with 95% confidence that the results are true. This means the system will perform better than 55% in the long-run.

A system with a 200 game sample would need a win rate in the low 60% range in order to have confidence that it would be better than 55% in the future.

This analysis is meant to be a guide when building betting systems but the general rule is the more results (larger sample) a winning system has the more likely it is to succeed long term.

For more tips on system building including how to test for overfitting, click here.


Got an idea for a winning betting system?

Try Bet Labs with the Starter Package, just $49 for a month