Posted on Wednesday, 28th March 2012
Please read Sections 13.1.5-13.4 of Chapter13-MAR28.pdf and post a comment.
Please also take a quick look at the end of Section 13.1.2, where there are some new remarks about the F-test as a likelihood ratio test.
Posted in Class | Comments (13)
Leave a Reply
You must be logged in to post a comment.
March 28th, 2012 at 3:17 pm
At the end of 13.3.2, you mention that permutation or bootstraps for two-way ANOVA is more complicated but the concept is the same. Can you explain the proper structure for that? Do we still merge the data from all the different cells, or do we instead merge across one variable while retaining the grouping from the other variable?
March 28th, 2012 at 7:38 pm
I understand the motivation behind the rank-sum test, but the exact procedure for obtaining the p-value from the test still seems a bit unclear to me. Could you review how to get the p-value from this test?
March 28th, 2012 at 9:26 pm
I found the examples in 13.3 very helpful, but I’m confused by the use of Table 13.8 in the calculations- was it just to scale down the calculation for the rank-sum test or did it serve a bigger purpose?
March 28th, 2012 at 10:59 pm
After replacing distribution-free data by ranks to perform the rank-based ANOVA, can multiple comparison test be used (with the rankings) as they were for the parametric ANOVA.
March 28th, 2012 at 11:04 pm
sec 13.2.1: Is the notation X^T the trace of matrix X? How does that work when X is an nx1 vector of 1s? I am having difficulty with the notation on this page (421).
March 29th, 2012 at 4:04 am
You write that the procedures of non-parametric ANOVA methods continue to make the assumptions of additivity and independence of errors. How do these assumptions limit what can be said about non-parametric ANOVA?
March 29th, 2012 at 4:22 am
You mention a couple options for matrices of indicator variables. Is one generally preferable to the other?
March 29th, 2012 at 6:38 am
Can you provide some more detail on the Kruskal-Wallis Test?
March 29th, 2012 at 7:00 am
In 13.4, the usual F test on log trasformed data and rank based test have similar results. Do they usually show similar results? Should we consider transformation before using the rank test?
For rank based procedure and permutation/ bootstrap, is there a preference choosing one over another under some conditions?
March 29th, 2012 at 7:51 am
Is there a difference in models between 2-way between subjects (where there is still only 1 subject per condition) and 2-way within subjects (where one subject experiences all conditions)?
March 29th, 2012 at 8:12 am
What are the main drawbacks to the rank-sum test? It seems to help minimize the effects of what appear as outliers but it seems like it could be hiding potentially important data.
March 29th, 2012 at 8:20 am
I don’t quite get the intuitive leap from ANOVA interactions to the multiplication of x1 and x2 in the linear regression case. Is x1*x2 just a special case? Can we look at other combinations of x1 and x2?
Is there a form of rank-based ANOVA that preserves information about the absolute differences between consecutive ranks? Or is it really not that important? What assumptions, if any, are necessary for rank-based procedures?
March 29th, 2012 at 8:29 am
The details of the rank-sum test are still a little hazy to me. Can we go over it in class?