Re: how to test this kind of hypothesis?

From: Glen (glenbarnett_at_geocities.com)
Date: 11/11/04 

Date: 10 Nov 2004 22:00:46 -0800

Richard Ulrich <Rich.Ulrich@comcast.net> wrote in message news:<7q73p09ur8b9n76l0bmdr03jc5hb9uju24@4ax.com>...
> On Tue, 09 Nov 2004 18:45:18 -0500, Jerry
> <greenmountain05401-forum@yahoo.com> wrote:
>
> > Hi,
> >
> > My question is in the nonparametric context.
> >
> > A pharmaceutical company is interested in studying the effectiveness of
> > 3 doses of a drug on lowering cholesterol. A total of 4 patients are
> > randomly assigned to the doses: 2 to the low dose and I each to the
> > medium and high doses, see data table below. The table entries are the
> > reduction in cholesterol.
> >
> > low medium high
> > -----------------------------
> > 2 4 2
> > 3
> >
> > I know I can use Kruskal-Wallis Test to test the
>
> Don't use KW for ordered alternatives, hardly ever.

Agreed - it's a big waste of power if you have an ordered alternative;
essentially you're spending 5/6 of your power (equivalently, wasting 5/6 of
your data) on alternatives you're not interested in.

There are nonparametric approaches for ordered alternatives.

Jonkheere-Terpstra is one such for the ordered alternative one-way design.

Glen