2

Is there a statistical program to calculate power for data generated by assigning a score (not necessarily linear ie score of 2 may not be equal to twice a score of 1)

Laura Goldsmith
  • 21
  • 1
  • Could you describe how the score is assigned and what test(s) you are running that needs a power analysis? – Jonathan Thiele Jun 19 '12 at 18:08
  • If by statistical program you mean method and by power you mean the power of the statistical test that is the method the answer is yes. Assuming this I will provide an answer. If I am wrong in my interpretation I will delete it. – Michael R. Chernick Jun 19 '12 at 18:19
  • 1
    What do you mean by power? What type of analysis are you even using? – AdamO Jun 19 '12 at 18:31
  • I think Laura meant the power of the statistical test she would be using to compare two groups. Admittedly I did a little educated guessing or mind reading here. – Michael R. Chernick Jun 19 '12 at 20:23
  • As your outcome is ordinal, I would consider [ordeinal logistic regression](http://en.wikipedia.org/wiki/Ordered_logit). Power analyses are often tricky. Greg Snow provides a good example of power analysis by simulation [here](http://stats.stackexchange.com/questions/22406/power-analysis-for-ordinal-logistic-regression). If you provide some more information on the population you are studying (e.g. how strong do you expect the relationship to be between exposures and outcome, what is the expected distribution of age and hormonal status), someone might be willing to provide more focused help. – jthetzel Jun 19 '12 at 20:47

1 Answers1

1

I assume that your score is ordinal but not interval in that ratios cannot be equated. Then you can apply rank tests to the data to compare results for two different groups. The statistical test that is appropriate for comparing unpaired data is the Wilcoxon rank sum test. If you specify a difference in medians that you would like to be able to detect and the sample size that you have, there are programs available to determine the power of the test. nQuery Advisor is one computer package that will do this. In SAS the power procedure can do this.

Michael R. Chernick
  • 39,640
  • 28
  • 74
  • 143