JAYESH KUMAR
> I am tring to use some local varibales in a regreession model.
> For example,
>
> su yvar,detail
> local a1=r(p1)
> local a2=r(p99)
> reg yvar xvar if yvar>`a1' & yvar<`a99' *Eq.1
>
> su yvar, detail
> reg yvar xvar if yvar>r(p1) & yvar<r(p99) *Eq.2
>
> Now the problem is when I am estimating Eq.1, it uses the criteria
> yvar>=`a1'&yvar<=`a99' rather than expected.It uses the
> 5119 observations,
> while it should be 5118.
>
> Eq.2 is not estimated, it gives the error of no observation.
> But the
>
> su yvar,detail
> su yvar if yvar>r(p1)&yvar<r(p99)
>
> shows that I have 5118 observations.
> and su yvar if yvar>`a1' & yvar<`a99' shows that I have
> 5119 observations.
>
> Is there any problem with declaration of local variable?
> What are the
> otherways in wihich I can restrict my sample to be between
> 1% and 99% of yvar?
My guess is that this is at root a precision
problem. Your assignment of r(p1)
and r(p99) to local macros isn't
guaranteed to keep all the digits
in the originals. After all, it's
in essence both a numeric to string conversion
and a change from binary representation to
decimal representation all in one. I'd do it like this,
cutting out all the locals:
su yvar, detail
gen insample = yvar > r(p1) & yvar < r(p99)
reg yvar xvar if insample
(Actually, if this were my problem, I would
use -qreg-, as I dislike all outlier-dropping
tricks like this, unless the outliers
are definitely wrong in some strong sense.)
Nick
[email protected]
P.S. "local variable" is your own terminology.
"local macro" or "local" is Stata's. Best to
use Stata's.
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
