Testing Sports Hypotheses

[kimlee]

This is a useful link for testing sports hypotheses with the chi-square test. It works when you want to test a hypothesis with multiple outcomes. For example, suppose you believe an angle might be useful on either straight bets or teasers. There are three possibilities: teaser loses, teaser leg wins and straight side loses, or both win. Based on your no-vig betting odds, the respective probabilities of these events should be roughly 29%, 21%, and 50%.

Suppose you have a sample of 100 games where teaser legs hit 80% and straight bets hit 54%. Then the realized frequencies of events were 20, 26, and 46. The calculator says:

Chi squared equals 3.982 with 2 degrees of freedom.
The two-tailed P value equals 0.1365
By conventional criteria, this difference is considered to be not statistically significant.

Chi-Square Test (www.graphpad.com/quickcalcs/chisquared1.cfm)

[PerpetualCzech]

I should have commented on this earlier, sorry. This stuff is gold and shouldn't be left to waste as a single-post thread.

I know only very basic stats concepts. Again and again I hit the same wall in my analysis, a sample size that feels right (i.e. that it may be big enough) but I'm just not sure. I think my feel is good for the most part, but from what I understand the Chi-Squared Test puts that feel into a nice quantifiable result. Someday I gotta get of my ass and learn this stuff.

[2w2p2s]

This is a useful link for testing sports hypotheses with the chi-square test. It works when you want to test a hypothesis with multiple outcomes. For example, suppose you believe an angle might be useful on either straight bets or teasers. There are three possibilities: teaser loses, teaser leg wins and straight side loses, or both win. Based on your no-vig betting odds, the respective probabilities of these events should be roughly 29%, 21%, and 50%.

Suppose you have a sample of 100 games where teaser legs hit 80% and straight bets hit 54%. Then the realized frequencies of events were 20, 26, and 46. The calculator says:

Chi squared equals 3.982 with 2 degrees of freedom.
The two-tailed P value equals 0.1365
By conventional criteria, this difference is considered to be not statistically significant.

Chi-Square Test (www.graphpad.com/quickcalcs/chisquared1.cfm)


Nevermind.............TY

[ComptrBob]

Nevermind.............TY

You had the right idea about one of the numbers being wrong, but the correct frequencies are 20, 26, 54 not 20, 26, 46. (T lose %, T win % - S win %, S win %).

One of the real dangers of using the Chi-squared statistical test is that if it is applied to a dataset that has been heavily data mined using multiple hypotheses, it will not give statistically valid results.

[2w2p2s]

You had the right idea about one of the numbers being wrong, but the correct frequencies are 20, 26, 54 not 20, 26, 46. (T lose %, T win % - S win %, S win %).

One of the real dangers of using the Chi-squared statistical test is that if it is applied to a dataset that has been heavily data mined using multiple hypotheses, it will not give statistically valid results.


Got it. TY Bob, should have seen that's the # that was wrong.