0

I am looking for a suitable test for a set of binary data. Basically, we are verifying a piece of automation for a molecular biology process. We are testing samples that are either positive or negative, and comparing the results for automation, with the results for doing the same samples manually, and we need a test to determine if one is significantly better than the other. I will end up with two columns, one for "manual" one for "automation" and the response will be positive / negative, or 1/0.

Can anyone suggest a suitable statistical test? My only thoughts are either fishers or a two-sample proportion.

Am I right in thinking that normality testing wouldn't be appropriate for binary data like this?

kjetil b halvorsen
  • 63,378
  • 26
  • 142
  • 467
Daniel Grove
  • 1
  • 1
    Please give us more contextual information, as the sample size ... of course, for binary data which cannot be normal, – kjetil b halvorsen Sep 09 '21 at 02:40
  • What does "better" mean? Is the manual process always right but slow? Would you be willing to accept a small error for the automatic processes if it means throughput increases ten fold? – Demetri Pananos Sep 09 '21 at 02:47
  • Also, see [McNemar's Test](https://stats.stackexchange.com/questions/341812/one-sided-mcnemars-test) – Demetri Pananos Sep 09 '21 at 03:00
  • Both appear to be accurate, the automation is simply to speed things up, so what we need to determine is whether automation is equally accurate to manual handling. We have 85 tests for manual handling, and 85 for automation. I should add that the samples we are testing are samples where we already know whether they should provide a positive or negative result, so we can look at the results we are getting and know whether or not they are correct. – Daniel Grove Sep 09 '21 at 08:47

0 Answers0