Hi again
1) Parametric tests are more sensitive when parametric assumptions are
met. If the assumptions are not, for example the data is highly
skewed, nonparametric tests may have more statistical power.
2) The rationale would be that the assumptions required for ANOVA were
not met and you suspected that the violation was likely to bias the
results so you used non-parametric tests. I guess for the other times
you would have to say that either the assumptions of ANOVA were met,
or they weren't but you felt the violation was relatively minor and
unlikely to impact on the ANOVA. Evidence of a sensitivity analysis
(i.e. you ran both methods and they gave the same/similar results)
would be helpful.
Good luck in Friday
Sam
On 24 October 2010 15:30, Helen Mann <[log in to unmask]> wrote:
> Me again....Two things this time....
>
> 1) Can someone tell me why when I did a Wilcoxon Signed Ranks test
> (non-para) I found signficant differences between a) time 1-time 2 b) time
> 1-time 3 c) time 1 and time 3
> BUT when I did the parametric alternative (repeated measures ANOVA) I got
> significant differences on a) time 1-time 2 b) time 1-time 3 but c) time 1
> and time 3 was not significant....
>
> I thought that as non-para tests are not as sensitive as para tests that it
> would've been the other way round, that the para test would show something
> significant that the non-para wouldn't find.....
>
> Any ideas?
>
> 2) Given a data set of 155 participants on before, pre- and post- test (time
> 1, time 2 and time 3) what reason(s) would you choose to do a Friedman and
> Wilcoxon Signed Ranks instead of the repeated measures ANOVA when all your
> other tests have been parametric?
>
> Last minute panic before the viva on Friday!!
>
> Helen x
>
--
Sam Norton
Centre for Lifespan & Chronic Illness Research
University of Hertfordshire
www.go.herts.ac.uk/samnorton
|