I need to compare two alternative methods of manufacturing a product, say A and B. I wanted first to dermine the sample size of the measurings, and found this post in a sister site. There, it says that the number of samples is a function of the Z-value for a desired confidence, the standard deviation and the expected difference between alternatives. At first, it seemed to solve my problem, but then it sparked more questions than I had before. That's why I come here for help (I think the subject belongs more to the specialized CV site that to the general Mathematics, that's why I came with it here instead of there):

  1. What should I use for standard deviation? The post contains σ which is normally used to designate the population's standard deviation. Obviously I don't have it, so I assume I should use the sample's instead (see here, for example). The problem is, those two are different for small sample sizes, so there comes a "recursive" what sample size should I take?
  2. There's only one standard deviation, and I want to compare two alternatives. Does it mean that I should determine a sample size for each alternative, depending on its sd? or I should define the sample size as the max of the two estimated numbers.
  3. Is there any comparisson (preferably numerical) to determine that indeed there's one alternative that is better than the other? If so, what test should I run?


====== EDIT TO ADD =======

At a factory there are two operators, A and B who perform the same assembly operation. The sequence each of them uses is slightly different (I call them "methods" A and B), but take about the same time. I want to find the fastest of the two methods, and mathematically support my decision.

Method A takes 4:34 minutes in a 4-sample average, and the standard deviation is of about 1:50 minutes (for the same 4 registries). Method B takes 5:01 in average, with a standard deviation of 0:13 (over 3 measurements).

I'd like to know which of the methods could be considered "faster" with a confidence of, say, 90%. That's why I landed reading about sample sizes, because intuition tells me that my 3 and 4 trials aren't enough to make a call. I hope this provide further details about what I need to achieve.

  • 2
    $\begingroup$ I don't think we we are really clear on what it is you are trying to do. Before making references to external sites, can you tell us what it is you are trying to find and how your variables are measured? Only then will you likely get some useful comments/responses. Why did you want to find the sample sizes? Has the data already been collected? Or are you trying to plan an experiment in the future to compare A and B? $\endgroup$ Aug 1, 2015 at 0:33
  • $\begingroup$ If knowing $\sigma$ would make using a Z suitable, replacing it with a $t$ and using sample standard deviation $s_{n-1}$ should usually be adequate for the case where you don't know $\sigma$. $\endgroup$
    – Glen_b
    Aug 1, 2015 at 2:16


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Browse other questions tagged or ask your own question.