Determine probability of better player given indirect measurements

71 Views Asked by At

Say you have a two-player game and three, fixed-but-different skill players: A, B and C. Player A only plays against C and player B also only plays against C. You want to to determine if player A is better than player B.

Say A wins 60 of 400 games with C while B wins 120 of 400 games with C. Clearly, in this case, one can say B is better than A.

But if the results are closer, say: A wins 150 games out of 400 with player C. B wins 165 of 400. With what confidence can we still say that B is likely to actually be better than A?

I think a binomial test would be appropriate if the two players I'm interested about were directly playing against each other. If A wins 175 out of 400 versus B, we can be >95% certain that B is better than A (98.6% certain). But I'm not sure how to apply that if the players can only face a common opponent but not each other.

1

There are 1 best solutions below

0
On BEST ANSWER

Use Proportion Differences Hypothesis Testing

p1 = Proportion of games A wins against C

p2 = Proportion of games B wins against C

H0 : p1-p2<=0

H1 : p1-p2>0

Pooled Sample proportion = (p1*n1+p2*n2)/(n1+n2)

p1 = 150/400 = .375

n1 = 400

p2=165/400= .4125

n2 = 400

p = (150+165)/800 = .39375

SE $= \sqrt{p*(1-p)*\left(\frac{1}{n_1}+\frac{1}{n_2}\right)}$

$ = \sqrt{.39375*.60625*\left(\frac{1}{400}+\frac{1}{400}\right)} = .03455$

Z statistic = (p1-p2)/SE = (.375-.4125)/.03455 = -1.085

level of significance is$\alpha = 0.05$

P(z<=-1.085) = 0.13889

p-value = 0.13889

Since the P-value (0.13889) is greater than the significance level (0.05), we cannot reject the null hypothesis and conclude that player A is less talented than player B.

In other words we can be 86.11% certain that player B is better than player A Thanks

Satish