Not asking the question you really have is a pretty inefficient way to
proceed.
That aside, the mechanics of how to do this have been thoroughly
ventilated, but its meaning has not been.
This looks like a lot of work put into throwing away information -- and
also one that prejudges the nature of any interactions. Is there any
situation in physical/biological/social science where the mechanism or
generating process is a step function with step at the 10th percentile?
If you want to capture some kind of sting in the tail a transformation
of some kind is more likely to yield an effective functional form in my
view.
Nick
[email protected]
Pancho Villa
Thank you, Nick and Martin. I abbreviated my problem for simplicity.
So much that I'm afraid the feedback I got from you is not helping me
much.
Actually, my variable y is not binary, but categorical with more than
2 possible values. My y depends on the 10th percentile values of x
and z, respectively. (In fact, it depends on the values of other
variables, but that's not problematic. This is the simplest
description of my problem that can get me helpful feedback from you.)
So, it's more like this:
y = 0 if x <= 10th percentile of x & z <= 10th percentile of z
y = 1 if x <= 10th percentile of x & z > 10th percentile of z
y = 2 if x > 10th percentile of x & z <= 10th percentile of z
y = 2 if x > 10th percentile of x & z > 10th percentile of z
I found that I can do this:
centile x z, centile(10)
which gives me the 10th percentiles of x and z in the same output.
But I don't know how to invoke the result for each variable in my next
lines. Truly basic.
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
