Jun Xu wrote
> This email might be kind of long, but to explain this
> problem clearly, could
> you bear with me for a second :( Thanks a lot.
>
> Thanks a lot for your help, and I have managed to write an
> ado file that
> seems to work fine till I found out that it won't work with
> a varlist of
> more than 8. I have attached our previous Q&A emails at the end.
>
> Notes:
>
> `imax': here is a macro that I have grabbed using 2^`nvars'-1.
> `nvars': is a macro containing number of variables in the varlist.
>
> part of the ado file that's relevant to my question
> ************************************************************
> ************
> ......
> ......
> ......
> *create this response pattern variabl and replace the value
> in future
> quietly gen `pat02'=.
> format `pat02' %0`nvars'.0f
>
> *create other tempvars
>
> forval i = 1 / `imax' {
> qui inbase 2 `i'
>
>
> local which : di %0`nvars'.0f `r(base)'
>
> *local a="`which'"
> *di "`which'"
>
> forval j = 1 / `nvars' {
> local char = substr("`which'",`j',1)
>
> *I only invoke `which' once within this loop
> .......
> .......
> .......
>
> }
>
>
> quietly replace `pat02'=`which' if _n==`i'
>
> *I don't see what's wrong here and I have used the
> following three *lines
> kind of check the instantaneous value of `which' and it is
> a *binary code
> everytime, but `pat02'[`i'] becomes not binary after the
> *binary code
> 100000000, so I strongly suspect either I misused the
> *replace statement, or
> there is something unique to "local which : di
> %*0`nvars'.0f `r(base)'" the
> command that Nick taught me in his email to *my question,
> which I might not
> understand well. The puzzle here is *that no problems with
> 8 variables and
> less, but not 9 variables and *above as you will see in the
> output, where
> the binary codes will have *other values than 0 and 1.
In your code `pat02' is created as a -float-. That
is, you didn't specify a variable type, so Stata created `pat02'
as a -float- (unless, as seems unlikely, you earlier -set type
double-).
There aren't enough bits in a float value to ensure that your
larger "binary" numbers are held absolutely exactly, which you
need for this problem.
The heart of the difficulty is shown by this example:
. set obs 1
. gen myvar = 11111111
. d myvar
storage display value
variable name type format label variable label
----------------------------------------------------------------------
---------
myvar float %9.0g
. format myvar %9.0f
. l
+----------+
| myvar |
|----------|
1. | 11111111 |
+----------+
. replace myvar = 111111111
(1 real change made)
. l
+-----------+
| myvar |
|-----------|
1. | 111111112 |
+-----------+
That is, despite your -format- some inaccuracy is creeping
once you have figures about as big as 10^9. (Of course,
Stata does not know that you are privately thinking of these
decimals as binary numbers.)
I think you would be better off using a -long- or
a -string- representation.
Here is some code which I think is close to what
you want.
*! NJC 16 February 2003
program countnonmissing
version 7
syntax varlist
tokenize `varlist'
local nvars : word count `varlist'
local imax = 2^`nvars' - 1
qui {
forval i = `imax'(-1)1 {
inbase 2 `i'
local which : di %0`nvars'.0f `r(base)'
noi di "{txt}`which' " _c
local vars
forval j = 1 / `nvars' {
local char = substr("`which'",`j',1)
if `char' {
local vars "`vars'``j'' "
}
}
local nv : word count `vars'
local nvm1 = `nv' - 1
local W
forval j = 1 / `nvm1' {
local wj : word `j' of `vars'
local W "`W'`wj',"
}
local wj : word `nv' of `vars'
local W "`W'`wj'"
count if !missing(`W')
noi di %6.0f r(N)
}
}
end
e.g.
countnonmissing *
countnonmissing make mpg rep78
Nick
[email protected]
*
* 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/
