🚀 go-pugleaf

On Tue, 25 Feb 1992 10:50:29 EDT you said:
>From:  Jacqueline Cater
>Suppose you are handed a data set with a variable number of multiple obs
>per subject.  Short of eliminating data on a random basis so that each
>obervation is a unique person (which the researcher does not want to do)
>one can perform the logistic regression with the explicit caveat that the data
>do not represent independent observations and that any parameter estimates
>and concomitant estimates of relative risk for the antecendent factors
>are biased.  Why are multiple "fields per patient" taken as the unit of
>observation in many published studies?
>
Well, I can't speak for 'many published studies', as there _could_ be
instances where the individual observations were legitimately used as
the unit of study.

From my consulting experiences, it seems like those who came to me with
1000 observations on 1 subject (Ain't electronic measuring and electronic
storage of results just _FANTASTIC_!?) didn't understand basic principles
of a) sampling, b) research design, c) generalizeability, and d) statistical
inference.  (DISCLAIMER:  I've only been at this institution for 6 months,
so I'm _NOT_ speaking of experiences here.  (Hmmmm - when do I apply for
promotion & tenure.....?))

They seemed to believe that because the p value was vanishingly small that
'proved' (I threaten students in my stats courses with failure if they _ever_
use 'prove' within the context of statistical inference!) there was a
difference between conditions that was generalizeable beyond that 1 subject
and the specific conditions in the case study.  (What? - how can this be a
CASE STUDY with all those nice values and that small p value?!)

The above was just my experience, and may have been a consequence of the
statistical level of some researchers at my previous place of employment.

Perhaps some other thoughts could expand these limited horizons.

- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
Michael T. Weaver, RN Ph.D.
U. of Alabama at Birmingham       Internet: nurs036@uabdpo.dpo.uab.edu
School of Nursing                 Bitnet  : nurs036@uabdpo
UAB Station                       Radio   : kg8h
Birmingham, AL  35294-1210        Phone   : (205) 934-6913
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -