What would the Null Hypothesis be for predicting sports outcomes?


New Member
So I made a program to predict the outcome of baseball games, as either a W or a L for the home team. I want to know if my program and algorithm are actually having an effect or is it just luck. I thought at first that it would be the likelihood of seeing how many times I was correct and seeing how likely that was. For that, I thought to do .5^n where n is the number of correct predictions I got. But I feel like that is not the correct way to do it as teams do not always have 50/50 odds to win. Also, that felt incorrect because they may not all be in a row. So I thought that maybe I should do the number of times my odds lined up with betting odds. But I dont know what the probability of me getting withing 5% of the professional odds are.

My question is what would the null hypothesis be? Would it be based on # correct or # of times my odds lined up with the odds for that game?
Last edited:
Would it make sense to test the one sample proportion of correct outcomes against nullhypothesis H0: po=0.5, using a normal distribution with one tail?


x:correct predictions
n:number of observations

for the one tailed hypothesis test p > p0, the resullt is significant if zobs > z(1−alpha) alpha: significance level.

The outcomes of the games might not be rated as 50/50, but from what i understand, the p0=0.5 here tests the accuracy of the algorithm, which will approach being correct 50% of the time if it doesnt work, given enough samples.

interesting project, will you let me know the result of the hypothesis test if you decide to do it that way?
Last edited:
a common sigificance level is usually alpha=0.05
so if your zobs>z(1-alpha), at alpha=0.05, it means that you are 95% certain that it is not just due to chance.

z(1-0.05) = 1.644854, this is your critical value for the one tailed hypothesis test. if zobs is larger than that, the result is significant.

i welcome others to chime in if they spot something here that doesnt make sense :)