Re: st: Dropping highly correlated variables

[Sergiy Radyakin <serjradyakin@gmail.com>]

George, there is no single Stata command that would do exactly what
you describe. So you will have to program it yourself. Given that you
know how to "proceed manually" just put those commands in a do file
and loop by variables. The more difficult part is whether there is a
unique solution to your problem even the number of variables
remaining. And just in case, if you are thinking about reducing
dimensionality, consider principal components analysis. Best, Sergiy.

On Mon, Jul 22, 2013 at 5:07 AM, George Murray
<george.murray16@gmail.com> wrote:
> Dear Statalist,
>
> I am using a large dataset with 1000 or so variables. Let’s call them
> v1,v2,…,v1000. My question is, if two variables have a correlation
> coefficient greater than 0.99, how do I delete the variable with the
> smallest number of observations (or choose to drop one arbitrarily in
> the case that the both have the same number of observations)?
>
> For example, if v123 and v456 have a 0.995 (>0.99) correlation
> coefficient, and v456 has more observations than v123, how would I
> drop v123?
>
> I know that one may use –correlate v1-v1000-, and then proceed
> manually (which is what I will have to do if no solution is found),
> but I am hoping that someone knows of a quicker way. Incidentally,
> this is not for any ‘statistical’ reason (e.g. multicollinearity), I
> am just cleaning up the dataset.
>
> Thank you,
>
> George.
>
> *
> *   For searches and help try:
> *   http://www.stata.com/help.cgi?search
> *   http://www.stata.com/support/faqs/resources/statalist-faq/
> *   http://www.ats.ucla.edu/stat/stata/

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/