Interquartile Overlap

[zeeshan qadri]

Dear Group member,

 

I had analyzed a clinical data in which I found interquartile ranges overlap e.g [ Median  (IQR) 62 (51--105) & 75 (50--90) ] & one more question if my confidence interval are very narrow like 1.01 (1.00-1.01) Please comment on this findings.

 

Thanks

 

Zeeshan

[Steve Simon, P.Mean Consulting]

On 3/14/2011 7:56 AM, zeeshan qadri wrote:

> I had analyzed a clinical data in which I found interquartile ranges
> overlap e.g [ Median (IQR) 62 (51--105) & 75 (50--90) ] & one more
> question if my confidence interval are very narrow like 1.01
> (1.00-1.01) Please comment on this findings.

I'm not sure what your questions are. If two IQRs overlap, you can take
that as an indication that the two groups have reasonably similar
values. It may be worthwhile to look at other measures of the
similarities of the two groups. What other measures might be appropriate
depend on your original research hypothesis and the nature of the data
you are collecting. "Clinical data" is a very broad term. You can share
some more details about your data set here. I promise that no one will
try to steal your research idea.

That confidence interval is rather unusual, but it does arise quite
frequently in a logistic regression model where the units of measurement
are very small. In this case, you might want to convert to a different
(larger) unit of measurement and re-analyze your data. For example,
suppose you are examining the relationship between birth weight and
mortality. If you measure birth weight in grams, the odds ratio measures
the increase in the odds of dying when birth weight increases by only
one gram. That is such as small increase in risk that the odds ratio
cannot be easily displayed using the default options in most statistical
packages. Your odds ratio is probably something like 1.005853 with a
confidence interval that goes from 1.00124 to 1.01127. If you recoded
the data in terms of kilograms, you would get a much more sensible
number that is easily interpreted. In theory, you could take your odds
ratio and raise it to the 1000 power, but that's a rather perilous thing
to do, as even a bit of rounding can cause very bad results. It's much
better to transform the units and then re-run the model.

I hope this makes sense. Please feel free to write back with more
details if you want additional help from me and the others on this list.

Steve Simon, n...@pmean.com, Standard Disclaimer.
Sign up for the Monthly Mean, the newsletter that
dares to call itself average at www.pmean.com/news