Help? Calculating sample size for nonparametric data?

[sanddevil]

Hi, all!

Wondering if I could get a bit of help. I've just jumped onto a research project looking at patient satisfaction with a particular surgical procedure. My PI previously used a 10-point Likert scale to look at one group of people's satisfaction with a similar procedure, and now he's changing procedures (slightly) and patient populations.

Here's the deal. When he first asked the question of satisfaction, he reported a mean of 2.1 (1=total satisfaction, 10=total dissatisfaction). He also ascertained a minimally important clinical difference of 5 points.

He wants to conduct a study powered to detect an MICD of 3 points, (alpha = 0.05, power = 80%).

Here's my problem:
1. Can you REALLY report a mean on ordinal values? Granted, he wasn't happy when I first brought that up.

More importantly:
2. I'm struggling and struggling and struggling to get him a good sample size calculation that I feel confident in. None of my textbooks help. I've looked at BMJ 1995;311:1145-1148 and J Am Statist Assoc 1987;82:645-647, but they both want PROPORTIONS, and I can't, for the life of me, figure out how to convert an MICD of 3 to Pr(Y<X) or to some sort of odds ratio.

Anyone? Help me get him a sample size?

(PS...to add one wrinkle, he talked to one of his "friends" who told him that he only needed 9 patients per arm in his study. I think he's dead wrong, but I can't seem to convince myself of that).

[sanddevil]

Can anyone help? I'm seriously dying here.