Re: Valid study design
From: Richard Ulrich (Rich.Ulrich_at_comcast.net)
Date: 02/28/05
- Next message: Richard Ulrich: "Re: Skuls: Diff. of Means by Monte Carlo"
- Previous message: picogoethe: "help with difficult problem?"
- In reply to: Jim McGivney: "Valid study design"
- Next in thread: Bruce Weaver: "Re: Valid study design"
- Messages sorted by: [ date ] [ thread ]
Date: Mon, 28 Feb 2005 18:12:28 -0500
On Sun, 27 Feb 2005 22:41:16 GMT, "Jim McGivney"
<mcgivney@nospam.winid.com> wrote:
> I take a photo of a plaster model of a dental arch (upper or lower teeth).
> From the photo I can construct an overlay. There are two methods of
> constructing overlays. Two different dentists will construct the overlays.
> Each examiner will construct 40 overlays using each of the two methods.
>
> There is a method to determine a numerical measurement of how much a
> specific overlay deviates from the dental model. I have the measurements
> for the 160 overlays.
>
> I would like to present and publish this data. I have been told that
> editors will insist on a p-value. I can obtain a mean and standard
> deviation of the results of each examiner and of each method.
>
> Does this sound like a valid study design ?
As I understand it -
There are 40 photos.
Two dentists each constructed 2 overlays from each photo,
using the two different methods.
Questions for the design --
Were the dentists well-experienced in both methods at
the start? or could there be learning during the experiment?
(That would damage good inference.)
Was the "order" of the methods randomized, or otherwise
done systematically? Did a dentist do both methods for a
photo and then move on? - or, all 40 by method A, then 40
by method B?
(Inference is also limited, if someone could counter-argue
that the one that was better was always done after practicing
first on the other one.)
>
> How do I calculate the p-value ?
Well, is everything totally in favor of one method?
Is that the hope of the study, so show superiority?
The simple demonstration might be by using the sign
test: A beats B for doctor-1, and A beats B for doctor-2.
(Test by the binomial, same as for coin tosses, where ties
are ignored. Compare A>B(heads) to B>A (tails). Is it
more extreme than chance?)
Does A always *beat* B or do at least as well, for both dentists?
When B wins by one dentist, WHY is that? Is there some
feature peculiar to the picture, so that BOTH dentists have
B winning at the same time, even though A usually wins?
Your statement gets tougher to make if it is not directly
shown for each of two, but you would combine the two.
and have a statistician help you write the excuse, based
on what else can be said about comparability and
fairness of the trial.
-- Rich Ulrich, wpilib@pitt.edu http://www.pitt.edu/~wpilib/index.html
- Next message: Richard Ulrich: "Re: Skuls: Diff. of Means by Monte Carlo"
- Previous message: picogoethe: "help with difficult problem?"
- In reply to: Jim McGivney: "Valid study design"
- Next in thread: Bruce Weaver: "Re: Valid study design"
- Messages sorted by: [ date ] [ thread ]
Relevant Pages
|
