1
$\begingroup$

I have 1000 computers and 2 have failed with a specific code bug. They all are exposed to the bug, because they all have the same code running. Just using those numbers, is the probability of the failure event 2/1000 ?

$\endgroup$
  • 2
    $\begingroup$ Why do you see the occurence of a bug as a random thing? If you rerun the code on all computers, will the same computers bug out? $\endgroup$ – Roland Jan 28 '14 at 12:08
  • 1
    $\begingroup$ Yes, same code is running on all, so it is somewhat random; dependent on many variables, some of which can not be easily determined. $\endgroup$ – user38092 Jan 28 '14 at 22:50
3
$\begingroup$

Assuming that the computers are identical and independent, and at any computer, the bug occurs randomly with an (unkown) probability of p, we can describe your question with a Binomial Distribution where the number of computers is n=1000.

What you are doing is to estimate the unknown p of the model by your empirical probablility $\hat p = \frac{2}{1000}$. This doesn't tell you the true value of p. Also, it doesn't tell you how good your estimation is, i.e. how close it is to the true value, but this kind of questions can be answered with the help of confidence intervals.

$\endgroup$
  • 1
    $\begingroup$ I would just add that if you want to be fairly conservative in your coverage, the best CI method to use might be the Blaker interval, which is not mentioned in the Wikipedia article @Roland linked to, but is implemented in R with the package BlakerCI. $\endgroup$ – Unwisdom Jan 28 '14 at 14:30
  • $\begingroup$ Thanks much for the feedback.I will explore the CI next and see what that looks like. $\endgroup$ – user38092 Jan 28 '14 at 22:51

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.