"I think I am correct in using the Wilcoxon signed rank test as the data is not normally distributed" - No, not necessarily. It mainly depends what kind of hypothesis you want to test. If you decided for a hypothesis to test you can find/identify a procedure that enables you to test that hypothesis.

From your description it is not clear to me what the response is. But you write that your "review can only reduce the number of medications". If this is the response (change in number of medications), and if this can only become less, then your test makes no sense (it's meaningless at least).

Thank you for your response. To clarify, my null hypothesis would be that the review makes no difference to the number of medicines. The review can either do nothing (number of medicines will not change) or reduce the number of medicines (they cannot increase). I hope that makes sense.

Ok, then your test is meaningless. You know that the change can only be negative, so there is no (new) information you get from the test. It might make more sense to test if the data are sufficient to conclude that the decrease is relevant (whatever you deem is a relevant decrease).

If the data are a count of the reduced medicines have you considered a count model such as Poisson or negative Binomial regression or even a zero-inflated count model.

Actually there is a form of the latter known as a hurdle model that would model the probability of the number of medicines being reduced (vs. not) and the number being reduced simultaneously. It is arguably overkill if you have no other predictors but makes more sense if there are other predictors of interest. 50 data points is on the low side for this so it might be that a simpler model would be sufficient.

Hi thank you for your advice. Unfortunately I have no experience with count models. I will look into this.

Jochen thank you for your directness. My hope from a statistical test is to prove there is a significant reduction in the amount of medicines each patient is on. So null hypothesis there is no change in the amount of medicines and alternative hypothesis is there is a significant reduction in number of medicines. From my limited statistics teaching I thought that the best test I could use was Wilcoxon signed rank but maybe the answer is something I have not learnt about...

I think that you can carried out data transformation for satisfying normality assumption and then proceed for z test then it could gives more accurate results

Dear Emma McDowell , you can test if your data warrant to claim a significant reduction - you just need to test an appropriate null hypothesis. Testing "no change" as null hypothesis is meaningless in your case, but testing "a reduction of at least ... % " where "... %" is a (relative) reduction considered significant/relevant would make sense and would be meaningful. Such a hypothesis cannot be tested with Wilcoxon, but a count model as suggested by Thom would allow you testing such kind of hypotheses.

Oh that makes sense, thank you! I only have limited statistics teaching behind me so this has been a struggle.