I have multiple student graduation rates from the same school. Each year, there are variations in the student population and each year there are different graduation rates. Is there a way to compare the graduation rate from year to year to see the following:

  1. Graduation rate is not changing due to random chance?
  2. The population levels are not effecting the scores?

To expand on #2, I mean, lets say in year 1, I have 50 children, then in year 2, I have 55 children. This means the graduation rate in year 2, each child weighs a bit less than year 1.

Thank you,


Take $p$ as the probability of a single student graduating (Bernoulli distribution), then the number of students graduating would be a binomial distribution, $Binom(N, p)$; so basically you are testing if $p$ is constant from one year to another. One way to test this is likelihood ratio test.

The reduced-model (or the null model) is the simpler model where $p$ is constant. You can easily find the estimate $\hat{p}$ under this estimation by MLE or method of moments, knowing that the mean of distribution is $p \times N$, and calculate the likelihood value using this estimate of $p$.

There would be multiple ways to define the full-model (or the alternative model); one would be to divide the sample points into halves, and take $p_0$ as the probability of graduation for the first half, and $p_1$ as the probability of graduation for the second half. Estimate each $p_i$ separately, and calculate the likelihood as the product of the likelihood of the first half with $\hat{p}_0$ and the second half with $\hat{p}_1$. (or, alternatively one can say log likelihood is the summation of log likelihood of the first half and the second half).

Finally, you can calculate the log-likelihood ratio of the two models as shown here and make inference on the full vs reduced model.

| cite | improve this answer | |
  • $\begingroup$ Thanks @behzad.nouri, im going to have to do some research on these methods to get them working. I'll post my results. $\endgroup$ – code base 5000 Sep 19 '14 at 9:07

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.