[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]
st: RE: continuous to categorical and determination of monotonic categories
I'll take the motivation here as given,
although clearly a separate discussion
on merits and demerits is possible.
There may be formal machinery to do this, but
I'd tend to proceed in an exploratory manner.
One key is to look at the conditional distributions
and see whether their properties change as expected.
Among other tools, you could look at a bundle
of cumulative distribution curves or kernel
estimates of density functions. -distplot- or -qplot- from
SSC have easy handles to draw graphs -by()-.
> The question that follows is part STATA and part statistical
> technique. Suppose I have a continuous independent variable in
> a logistic (or cox) model and I wish to reconfigure the
> variable as a categorical (read ordinal) variable so that the
> odds ratio (hazard ratio) will be more meaningful, but before I
> categorize the variable and insert it into the model I'd like
> to determine whether the categories (e.g hospital volume: low,
> medium, high) are monotonic relative to the outcome (e.g. shunt
> survival as measured in fractions of years) so as to establish
> some justification of linearity. If "monotonicity" cannot be
> determined, then I understand that indicator variables must be
> created and odds ratios (hazard ratios) returned are relative
> to a referent group. My question is two-fold: is this an
> acceptable technique and if so, is there a straightforward and
> acceptable way to proceed using STATA??
* For searches and help try: