Re: Are graded clinical signs more reliable than dichotomized?

Roland wrote:
My question is a bit different. I am talking about clinical signs that
are not measured on a continous scale but usually regarded as
qualitative, ie either present or not. However in reality most clinical
signs present with varying intensity and there is always a grayzone.
Different examiners have different "cutoff" for when they decide that
tenderness is present or not making such clinical examinations
unreliable.

For patients subjective experience of pain we are used to the visual
analog scale to grade the pain.

In my research I have used graded scales of clinical signs like
tenderness and assume that this will result in more reliable
assessments, but I wonder if there is any previous research which have
examined the reliability between examiners for qualitative data which
are graded rather than dichotome?

Roland Andersson
Surgeon
Sweden

Anything is better than a binary variable. The more levels the better. This isn't exactly the reference you need, but from a statistical standpoint there are many advantages of having at least 5 ordered levels to a variable. See http://biostat.mc.vanderbilt.edu/twiki/pub/Main/RmS/fehbib.html#whi93sam

Frank Harrell


Ray Koopman wrote:

Bruce Weaver wrote:

Here is one article you might find interesting from a recent issue of
Statistics in Medicine.

-------------------------------------------------------------------
STATISTICS IN MEDICINE
Statist. Med. 2006; 25:127-141
Published online 11 October 2005 in Wiley InterScience
(www.interscience.wiley.com)

Dichotomizing continuous predictors in multiple regression: a bad idea

Patrick Royston; Douglas G. Altman and Willi Sauerbrei
[...]

I think that methodologists are pretty much unanimous that heavily
quantizing -- especially dichotomizing -- a pre-existing continuous or
quasi-continuous measure is almost never a good idea. However, I think
the OP's question is a little different: How close to continuous
should a subjective rating scale be? When will a dichotomy suffice?
If a more finely graded scale is needed, how much finer should it be?

I don't think there is, or can be, a simple general answer to this
question. It will depend on too many details of the situation, some of
which (such as loss functions) may be subjective. Nevertheless, I seem
to remember encountering from time to time (and mostly many years ago)
convention presentations, MA and BA theses, etc, with titles such as
"A comparison of different rating scales for assessing XYZ", so
they're out there somewhere.


.

Relevant Pages

  • Re: four-point Likert scale items
    ... scale correlate. ... colleagues insists that we are still talking ordinal data. ... ordered values as the measure of the construct with a Likert SCALE. ... learned their statistics in psychology classes, ...
    (sci.stat.consult)
  • Re: four-point Likert scale items
    ... learned their statistics in psychology classes, ... describes analysis of ordinal data with parametric tests as the first ... implicitly seems to disapprove of averages of Likert items. ... for a short scale with conventional ordinal labels. ...
    (sci.stat.consult)
  • Re: Are graded clinical signs more reliable than dichotomized?
    ... analog scale to grade the pain. ... In my research I have used graded scales of clinical signs like ... are graded rather than dichotome? ... Statistics in Medicine. ...
    (sci.stat.consult)
  • Re: Cholesterol Scale of Measurement?
    ... would you be so kind to tell me what is the Scale of Measurement of ... is that an interval or ratio or whatever scale? ... i found this question in a quiz about statistics. ... mg per liter or some such? ...
    (sci.math)
  • Re: Pentatonic Scales
    ... Roland wrote: ... I was taught that the first scale is the "Blues" Am scale and the ... as long as you interpret the first line as corresponding ... to the low E string, the second to the B string, etc. ...
    (alt.guitar.beginner)