- Mark as New
- Bookmark
- Subscribe
- Mute
- RSS Feed
- Permalink
- Report Inappropriate Content
I wish to test for significance among paired pre-test & a single posttest score (dependent variables) and categorical age (Independent variable). The dependent variables are non-normally distributed. I created a new dependent variable 'MEANDIFF' [posttest score - pre-test score] and used a nonparametric method:
proc npar1way data=have wilcoxon;
class agegrp;
var MEANDIFF;
run;
Factor |
Pretest Score |
Posttest Score |
p-value |
|
|
|
p-value here |
<18 |
54.2 |
74.3 |
|
18-34 |
58.6 |
75.8 |
|
35-54 |
57.1 |
80.2 |
|
I get an overall p-value but am not sure if it's statistically sound to do this. Should I just report medians and (IQRs) and not report a p-value?
Any insight?
Accepted Solutions
- Mark as New
- Bookmark
- Subscribe
- Mute
- RSS Feed
- Permalink
- Report Inappropriate Content
You should be testing MEANDIFF for location, i.e. is it significantly greater or less than zero, with the sign or signed rank test in proc univariate. See example
- Mark as New
- Bookmark
- Subscribe
- Mute
- RSS Feed
- Permalink
- Report Inappropriate Content
You should be testing MEANDIFF for location, i.e. is it significantly greater or less than zero, with the sign or signed rank test in proc univariate. See example