When handicapping sports it's essential to look at past results against the spread. A large part of handicapping involves looking for profitable, but undiscovered, situations. But a winning record may not necessarily indicate a profitable betting system. As we all know, a series of 50-50 events does not always provide results exactly as we would expect - luck can still play a significant role, especially over small samples. Flip a coin 10 times; you may get 5 heads, but you are more likely to get a different number, like 4 heads, or even 9 heads.
Now flip that coin 100 times, and you will still probably not see precisely 50 heads. And flip it 1000 times and you will be close to a 50-50 distribution, but you still are unlikely to hit exactly 50%. Likewise, you can have your dog pick a team for you for Monday Night Football every week, and it wouldn't be all that shocking to see man's best friend hit a 65% 11-6 record over a season. Does that mean you've lucked into ownership of some money-making superdog? Of course not, it just means that short term winning and losing streaks happen randomly.
We can use statistics to figure out how likely a given record is to be due solely to chance. While many people's eyes start to glaze over when math is involved, it's actually a simple process to determine whether a result is due to randomness. You first need to determine the standard deviation of your data; this is just the square root of the sample size. If you have a system that went 60-40 over the past 5 years, the standard deviation is the square root of 100, or 10. Now, determine how many standard deviations your record is from 50%. You can do this by taking the net wins in your system (wins minus losses) and dividing by the standard deviation. The higher this number, the better. Your 60-40 system is two standard deviations above the expected results (60 minus 40 is 20, divided by the standard deviation of 10). That is good, but it means there is still a 5% chance the results were just due to dumb luck.
You should shoot for three standard deviations above the mean, which means only a 1% probability of the results being due to randomness alone. For a 100-game system, that translates to a record of 65-35. For a 20-game sample, that same level of significance requires a 17-3 record, demonstrating how much of an influence randomness can have on short-term results.
This method takes into account not just the winning percentage, but also the sample size. We all know that most short-term trends or angles are worthless, merely blips in long-term 50-50 systems. Determining the statistical significance of a set of data ensures that the underlying system has proven itself over time. Of course, one problem in sports handicapping is finding a large enough sample of data. You may think that home underdogs of 4 or more points in the NBA playoffs are a good bet, but you only have a handful of games to go on. Short-term luck will have played a large role in whatever conclusions you reach from the data.
There are two ways to deal with sample size issues. One is to acquire more data by looking further into the past to analyze more games. The danger in this is that results from a decade ago may not accurately reflect conditions today or in the future. The other is to expand your criteria by looking at, for example, all home dogs in the NBA playoffs, or all underdogs of 4 or more points including regular season games. This gives you more data to work with, but expanding your criteria dilutes your original system and may cause you to miss valuable opportunities.
Evaluating significance applies to handicappers as well. How many gambling messiahs have we watched arrive on the scene with a 60% season, only to bat .500 the rest of their tout career while slowly losing their clients' money? If a handicapper hit 60% in the NBA last year, does that mean he is actually a 60% handicapper? Perhaps he is a 50% handicapper who just got lucky over the course of a season. Evaluating his record for significance is an important first step in determining his separating the men from the boys.
Finding a significantly winning system (or, even rarer, a winning handicapper) is only half the battle. As our friends on Wall Street like to say, "Past performance is no guarantee of future results". Changes in betting markets, sports rules, or other conditions can make past results unreliable for current situations. It's important to not only consider significance, but also the limitations of the data and adjustments in spreads that occur over time as winning factors are teased out of past results by linesmakers and handicappers.
"The Complete Square's Guide to Sports Wagering" is a recurring series aimed at educating novice sports bettors. The next article will examine correlated parlays.
If you would like to make or read comments about this article, you may do so by visiting the Mess Hall forum at MajorWager where a thread has been started. Please click HERE