0
$\begingroup$

I got a question regarding the binomial distribution and testing for significance in general.

I need to divide a dataset into a treatment group (with phone call) and a control group (no phone call). My boss told me I need to divide it based on significance tests. However, I told him that's not possible if no one has been called yet, since I have no data to work with now (what is the success rate? what is the number of successes?).

The question is: First, am I wrong for thinking that this request with the signficance tests is not possible? Second, how would I divide the control and treatment group in a logical way? I was thinking about 90/10 since I have a limited sample size, but I have no way of supporting my choice other than this.

$\endgroup$
0
$\begingroup$

I think your boss is saying that when you define your control and test groups through a random assignment, you should test that there are no existing significant differences between the 2 groups prior to intervention. If there were, then the post-intervention analysis would be skewed.

$\endgroup$
  • $\begingroup$ Thanks! Since my data is heavily skewed to the right do I need to transform it first by exponentiation (I then get a uniform distribution) and then test for no significant differences? $\endgroup$ – Stephen Theunissen Feb 15 '18 at 16:24
  • $\begingroup$ No need to transform here, as the Binomial test is exact. $\endgroup$ – user64106 Feb 15 '18 at 16:27
  • $\begingroup$ @StephenTheunissen: If you found the answer helpful, please upvote and/or accept. $\endgroup$ – user64106 Feb 19 '18 at 10:04

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.