- Mark as New
- Bookmark
- Subscribe
- Mute
- RSS Feed
- Permalink
- Report Inappropriate Content
For a given experiment, How to verify that the observations are independent? How to verify the errors are normally distributed?
Accepted Solutions
- Mark as New
- Bookmark
- Subscribe
- Mute
- RSS Feed
- Permalink
- Report Inappropriate Content
Independence is determined based on knowledge of the experiment, ie measurements on siblings are not independent or multiple measurements om the same individual.
Normal an distribution can be verified by looking at a histogram - proc univariate - and normality tests also available via proc univariate.
- Mark as New
- Bookmark
- Subscribe
- Mute
- RSS Feed
- Permalink
- Report Inappropriate Content
Independence is determined based on knowledge of the experiment, ie measurements on siblings are not independent or multiple measurements om the same individual.
Normal an distribution can be verified by looking at a histogram - proc univariate - and normality tests also available via proc univariate.
- Mark as New
- Bookmark
- Subscribe
- Mute
- RSS Feed
- Permalink
- Report Inappropriate Content
What other methods are availble to test the normality?
- Mark as New
- Bookmark
- Subscribe
- Mute
- RSS Feed
- Permalink
- Report Inappropriate Content
Shapiro-Wilk test, two dead Russians test (Kolmogorov-Smirnov), QQ-plot.
The tests all suffer from the same kind of thing--if you have enough data to actually do the test, even miniscule differences from normality seem to trigger rejection of the null hypothesis.
Thus, I think the consensus these days is to look at the QQ plot, and see if there are noticeable shifts away from the diagonal.
See Rick Wicklin's blog. Here is a good start:
http://blogs.sas.com/content/iml/2011/10/28/modeling-the-distribution-of-data-create-a-qq-plot.html
Steve Denham