st: e(wexp) versus e(wexp): different routines return different things

[daniel klein <klein.daniel.81@googlemail.com>]

Mark,

following your approach you might code

loc wts = subinstr("`e(wexp)'", "=", "", .)
if "`e(wexp)'" != "" {
			collapse (mean) `y_sample' `xb_sample' [fw = `wts']
		}

Since <wts> holds a string you cannot use a -scalar- here (this is
Stata not Mata).

Another way to do it would be to add an -if- statement like in

if "`e(wexp)'" != "" {
                        if strmatch("`e(wexp)'", "*=*") collapse
(mean) `y_sample' `xb_sample' [fw `e(wexp)']
			else collapse (mean) `y_sample' `xb_sample' [fw = `e(wexp)']
		}

No need for regular expression here.

However I see another problem with your program. Note that runnig

reg y high low cows [aw=mycount]

with your code above will, in most cases, still not work. Here is why.
You are using <mycount> as analytic weights in the -regress- command
and try to use them as frequency weights in your code (which seems
weird anyway, be sure you really want to do that). Frequency weighs
however cannot be noninteger while analytic weights, as well as
sampling weights, contain nonintegers most of the time and we
typically use the latter two with estimation commands. Note that e.g.
-newey2- (which is a user written program from SSS, plase see
http://www.stata.com/support/faqs/res/statalist.html#stata) does not
even allow frequency weights.

Best
Daniel

---
Mark wrote

Greetings,

I have a program which executes after I run an estimation procedure,
and it does a collapse where weighting is an option.
[...]
Here is the kicker: reg returns e(wexp) as "= mycount", but newey2
returns e(wexp) as "mycount" (without the equals sign!)

The reg version properly executes the program above, but newey2 gives
an error about an inability to weight, since the syntax requires an
equals sign.
[...]
*
*   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/