Re: Hypothesis testing with proportions

"Alan J. Salmoni" <salmoni@xxxxxxxxx> wrote in message
news:1122410561.097267.70020@xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
> t-tests and not been questioned at conferences. Having said that, I've
> seen simple Likert-type scales analysed with t-tests get past peer
> review..!

.... and why not. If the residuals are approximately normal, the variances
approximately equal (for the two sample test) and the independence
assumption tenable the t test will give you pretty sensible results for a
Likert scale, proportions etc. In these cases the alternatives
(transformation etc.) should give you very similar results. the main reason
to prefer the t test (or a CI based on t) is probably for clarity of
presentation.

Thom


.