[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

Re: st: -collapse- and -merge-

From   "Austin Nichols" <>
Subject   Re: st: -collapse- and -merge-
Date   Sat, 3 Nov 2007 12:18:59 -0400

Philipp Rehm <>:

This seems to be the desirable outcome.  If you specify a merge
matched on a variable with missing values, you expect the missing
values to be matched.  If you specify uniqusing in your example, it
should not change the behavior since there is only one missing value
in the using file.  If you want missing values not to be merged, and
you have only one type of missing in both files, you can redefine one
or both of them so they no longer match, e.g.
 replace id=.a if id==.
or drop the obs with missing ids, but this is a choice you should
make, not -merge-.

On 11/3/07, Philipp Rehm <> wrote:
> I am regularly puzzled by a particular feature of -merge-, namely to
> match missing observations with each other. Here is an example:
> sysuse auto, clear
> sort price
> keep in 1/15
> replace foreign=. in 1/5
> preserve
>         collapse (mean) PRICE=price, by(foreign)
>         sort foreign
>         list
>         tempfile m
>         save `m'
> restore
> sort foreign
> merge foreign using `m'
> list foreign PRICE
> I can avoid this problem in various ways (a "drop if foreign==." after
> the -collapse- would be one option). I also understand that Stata reads
> missing values as very large numbers (i.e.: all nonmissing numbers < . <
> .a < .b < ... < .z). I do not understand, however, why it matches
> missing values with each other. Moreover, the same behavior persists
> when I specify the -merge- option "uniqusing".
> Let me add that this behavior does not seem as strange in the example
> above. However, I usually -merge- data from totally different
> data-sources. There is no logical pattern to the missing values, and no
> reason to match them.
> Am I missing something? Clarifications are appreciated.
> Thanks,
> Philipp
> *
> *   For searches and help try:
> *
> *
> *
*   For searches and help try:

© Copyright 1996–2017 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index