On Sat, 24 Aug 2002 06:52:45 -0400
Shayna Padgett <shayna@knology.net> wrote:
> Hi. I am a Masters student preparing to defend my thesis. I have come up with
> a problem I am not sure how to address. I am performing logistic regression
> on my data based on a dichotomous disease outcome (yes/no). I am worried that
> there might be a correlation between two of my variables (one is nights awake
> per month, and the other is use of a supplemental hormome). I did a Point
> Beserial Correlation and got and R of 0.15, that was signficant. This was
> sucha small number, I assumed the correlation was weak. However, it was then
> suggested to me to do a simple t-test comparing the mean of nights for the
> two yes/no groups and doing so, I get significant differences, with one group
> having a mean twice that of the other, which to me implies correlation. Do
> you know the best way to sort this out? I am unsure which is most correct.
> Thanks.
>
Let the multivariable logistic regression do its job and estimate the partial
effects (odds ratios or log odds ratios) of interest. You need not worry about
correlations among the predictors unless there is strong collinearity. I
suggest working through a logistic regression text. You do not need to be
looking at point biserial correlations and t-tests in my opinion.
--
Frank E Harrell Jr Prof. of Biostatistics & Statistics
Div. of Biostatistics & Epidem. Dept. of Health Evaluation Sciences
U. Virginia School of Medicine http://hesweb1.med.virginia.edu/biostat
|