0

I have two groups:

enter image description here Group 1, I ask them how willin to start once monthly pill and mark their response between 1-10.

Group 2, I first ask them how willin once daily pill, then I ask them once monthly.

I want to compare: How willin would subjects be to start a once monthly pill for hypertension between the two groups. I think those I ask once daily first would be more willin for once monthly.

What test would you use for my data. I think independent t-test? I have 200 total responders.

TheFermat
  • 135
  • 1
  • 7
  • Do you want to include age, race and Bender information in that comparison? Are the people in both groups the same people? – Bernhard Nov 19 '16 at 06:03
  • People in both groups are different. I wanted to first do normal analysis then do subgroup analyses and capture age, race, and gender. – TheFermat Nov 19 '16 at 06:04
  • And my other question? – Bernhard Nov 19 '16 at 06:04
  • People in both groups are different. I wanted to first do normal analysis then do subgroup analyses and capture age, race, and gender. – TheFermat Nov 19 '16 at 06:05

1 Answers1

1

In a strict sence, the t-test is for metric data and your data (or residuals) will never meet normality assumptions. One could therefore argue, that a rank sum test was more appropriate and with that n, it should have good power.

On the other hand, the t-test is pretty robust at n=200 and it is better known in some audiences.

It is a borderline case, in which both possibilities can be defended. If you might have to defend it against someone very strict, a rank sum test may be advantageous.

Bernhard
  • 7,419
  • 14
  • 36
  • Thank you for the response! I had a couple questions for my own general curiosity. 1) For the rank sum test, to confirm, my data would still not meet assumptions as the data isn't paired and doesn't come from the same population? 2) By rank sum test, in this case, the Mann–Whitney U test is more appropriate than the Wilcoxon signed-rank test? 3) At n=100, would the t-test still be considered robust? 4) Using subgroup analyses with age if n=50 (25 in each group), would that be acceptable or is the sample size too small for subgroup analyses? – TheFermat Nov 19 '16 at 06:27
  • 1) yes, it's a test for independent data 2)They come to identical results/are interchangeable. Name it as you want. 3) Matter of taste, usually yes. Some say >30, some >50, just stay reasonable. 4) subgroup analyses will lead to multiple testing and alpha inflation, which is probably the bigger problem to consider. Personally, I'd prefer the nonparametric alternative. – Bernhard Nov 19 '16 at 18:32