turn on suggestions

Auto-suggest helps you quickly narrow down your search results by suggesting possible matches as you type.

Showing results for

- JMP User Community
- :
- Discussions
- :
- Discussions
- :
- Multiple measures ANOVA or Survival probability An...

Topic Options

- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Printer Friendly Page

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Permalink
- Email to a Friend
- Report Inappropriate Content

Dec 7, 2015 6:23 AM
(851 views)

Hello,

I have some data that I am not quite sure how to analyze.

We had 20 specimens that were cut in 4 parts, one part left untreated and the other 3 received different treatments. We fatigued the samples recording the lifetimes. My ultimate objective is to figure out if any of these treatments are better than no treatment at all or better than the other treatments. Does not sound much complicated but the data is very skewed so my questions are:

Should I ran ANOVA over the (treated-control) differences? ignoring the skewness of the data

Should I take a log transformation and then calculate the differences? (the end result would be more of a comparison of ratios which sounds a bit weird to me)

Is some sort of survival probability analysis a better option to do these comparisons and if so, how do I account for the "pairing" of the data.

I really appreciate any comments, ideas, advice, I have been imagining so many ways to analyze the data that now I dont know what would give me the fairest assessment.

A

1 REPLY

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Permalink
- Email to a Friend
- Report Inappropriate Content

Dec 8, 2015 9:06 AM
(791 views)

You may ignore the skewness of the data up to an absolute value of skewness<=1 and kurtosis<=3.

Use the log-transformation if it generates approximately normally distributed data, otherwise use non-parametric methods like Kruskal-Wallis Test.

If you have repeated measures on certain units of observation make these units a random effect in parametric modelling.