Filter variables based on middle and last characters of the var name

[Wendy Garrard]

I need to filter for variables scattered throughout a large-ish dataset based on a combination of characters in the middle of the var names. Although this wildcard notation (awo*b~C) is not valid, it represents conceptually what I would like to filter. For example, I need to select all the variables from all waves (the wave # is the * in "awo*b~C) for scale "b" that have been recoded as "correct" indicated by the letter "C" added to the end of the original var name. I need to repeat variations on this for many different tasks.

This seems like a simple thing to do, but so far I see no obvious way to do this efficiently.

• Tried various wildcards (like * and ? and ~; in the variables manager and in command lines) .
• Search archives such as Stata Tips (like inlist and inrange).
• Also, the rather unsystematic construction of the var names does not lend itself to the ways I normally use loop functions like foreach or forvalues

Can someone please advise how to do this? ~~Thanks!

~~~Using ~ Stata14 ~ Windows~~~

[Nick Cox]

 
forval i = 0/9 { 
       capture confirm varlist awo`i'b*C 
       if _rc == 0  local mylist `mylist'  awo`i'b*C 
}

di "`mylist'"

[Wendy Garrard]

Thanks Nick. I tried this and did not get an error message. But neither did I get a display of the filtered variables or anything else). Maybe I am not understanding what to do with the "mylist " ?

Also, where the second wildcard is located in the var name (the ~) there are differing numbers of alphanumeric char between the "b" and the "C". I thought the ~ wildcard was a placeholder for >=1 char, however so far I have not been successful in using it.

[daniel klein]

The tilde (~) character is the same as the asterisk (*) and means 0 or more characters, but adds the restriction that only one variable name is allowed to match. The documentation is misleading, because it states that both, ~ and * mean 1 or more characters.

I would go with rename here. Try

rename (awo#b*C) (foo=) , dryrun r
return list
local myvars `r(oldnames)'
display "`myvars'"

Best
Daniel

[Nick Cox]

Sorry; syntax was wrong and the capture was eating the error.

clear 
set obs 1 
gen awo1bfrogC = 1 
gen awo7btoadC = 1
gen newt = 1 

forval i = 0/9 { 
       capture unab this : awo`i'b*C 
       if _rc == 0  local mylist `mylist'  `this' 
}

di "`mylist'"

awo1bfrogC awo7btoadC

[Wendy Garrard]

Nick , I see the logic of the syntax, and it would be a great time saver. Unfortunately, I am not able to make this work. All I can find changed in my data are the three vars (*frog *toad *newt) w all cases=1. Perhaps I just don't understand how to correctly use this approach.

• When I run this with my data, it does display the two new var names, but after the di "`mylist'" command. But when I try to use it as a varlist name with anything other than di (e.g., describe "`mylist'") it produces nothing, and return the error "invalid name"
• FYI -- I had t change the * to two ** since all my vars have at least 2 chars between the "b" and the "C".

Thanks for the suggestion though - it gives me ideas that I can try for other tasks.

[Nick Cox]

The approach gives you a complete set of wildcard varlists which you then can use in some command. Using display is just to show what is produced.

I don't understand your comment about ** at all: manifestly in my example * is quite sufficient to catch "frog" and "toad" which are multiple characters. It is not the case that ** has a different interpretation from * in wildcards

[Wendy Garrard]

My comment about the ** vs a single * is based on my experience with Stata recently.

• If I wanted to use a wildcard to filter for varariables named varX3name varZ5name varQ7name etc, I had to use var**name because it would not find all cases if I only used 1 * (like this var*name). It worked fine once I tried using an * for each placeholder I wanted to use. Maybe it is an anomalie, since the manual says * should represent >=1 char just like ~ . But since I could not get ~ to work, I stumbled upon using multiple **, and it did filter as I wanted.

Much thanks. I will use this approach ALOT! ;-).

[Nick Cox]

Thanks for the appreciation: I am not trying to be stubborn, but you have not given a reproducible example where ** works and * doesn't. I certainly can't reproduce that behavour:

 
. clear

. set obs 1 
number of observations (_N) was 0, now 1

. foreach v in varX3name varZ5name varQ7name { 
  2.         gen `v' = 42
  3. } 

. 
. describe 

Contains data
  obs:             1                          
 vars:             3                          
 size:            12                          
------------------------------------------------------------------------------
              storage   display    value
variable name   type    format     label      variable label
------------------------------------------------------------------------------
varX3name       float   %9.0g                 
varZ5name       float   %9.0g                 
varQ7name       float   %9.0g                 
------------------------------------------------------------------------------
Sorted by: 
     Note: Dataset has changed since last saved.

. 
. d var*name 

              storage   display    value
variable name   type    format     label      variable label
------------------------------------------------------------------------------
varX3name       float   %9.0g                 
varZ5name       float   %9.0g                 
varQ7name       float   %9.0g