Wilcoxon Signed Rank Test

[chumley]

Hello all,

I have a situation where I have repeated measures pre and post treatment. Visually pre-treat measure is sig greater then the post-treat measure. All the pre-test measures are higher then the post-test. To determine if a paired t-test could be used I tested the normality of the difference in the measures.

The difference is not normally distributed, so I used the Wilcoxon Signed Rank Test. Signifigance between the two measures, as expected, was found. However, does the fact that all the diff between measures are positive and none are negative effect the accuracy of the test?

If so, what other statistic could be used?

Thanks for any input.

[kobylkinks]

If your question is about your data: why the diffs are positive and there is no negative observation then this probably means that your data is censored or your missing data mechanism is not random.
If not so then any test will give you significant result.
Normalizing (log x, 1/x, 1/sqrt(x), sqrt(x)) your data and using paired test will give confidence intervals for the mean diff. I mean that it will help you in measuring your differences.