In the book Computer Age Statistical Inference the James-Stein estimator is introduced. Brad Efron runs through an example where batting averages are estimated from each players 90 at-bats.

$$p_i\sim N(P_i,\sigma_0^2)$$

where $\sigma_0^2$ is the binomial variance


Here $P_i$ is the true average (which is averages from about 300 additional at-bats). $\bar{p}$ is the average of the $p_i$.

All of the above is very clear, but I am working with a situation where the value I have for each player is not 90. In fact, they are never equal. For one player I have an 1080 at-bats, for another I have 1223, and for another 1700, etc.

What is my denominator in $\sigma_0^2=\bar{p}(1-\bar{p})/90$ when I have unequal numbers of at-bats? Is the average a good value?


Technically, your example has unequal variances, which is a special, more complex case of the Stein estimation problem. See Stein Estimator with unequal variances or Stein Estimator with unequal variances - Part 2 for more details.

However, given that n is rather large in your example and the underlying variances should not be so different, using the averages may be a practical although not entirely correct path.

| cite | improve this answer | |

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.