RE: st: keeping dates which are n periods apart

["Rajesh Tharyan" <R.Tharyan@exeter.ac.uk>]

Hi nick,

Thanks very much for that, what you suggest is exactly what I needed. I
think it must been my wording of what I wanted to achieve that prompted 
the complicated approach.

This was the original statement of the problem... In an earlier post

1. Start with the first date.

2. See if the next date is more than 730 days from it. if its less than
73o days away from the first date drop it.

3. If it is more than 730 days away keep it and 

4. Now this date becomes the reference date and the difference between
this date and the next is calculated and so on

However, Scotts approach to use mod and your final suggestion was all that
was needed.

Thank you

Regards
Rajesh



-----Original Message-----
From: owner-statalist@hsphsun2.harvard.edu
[mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of n j cox
Sent: 19 May 2008 15:37
To: statalist@hsphsun2.harvard.edu
Subject: Re: st: keeping dates which are n periods apart

The word description of your problem -- concisely given in your title --
suggests a simple solution based on Scott Merryman's excellent 
suggestion of -mod(,)- as a basis for your problem.

keep if mod(year, 3) == mod(1987, 3)

gives the solution that you specify for the first example. The choice of 
1987 reflects the fact that 1987 is the start of your first example.

Note that this

1. Does not entail any attention to the panel structure of your data.

2. Ensures that the same years are selected for all panels when available.

3. Can be transferred to other periods by changing 3 to what is desired.

4. Can be modified easily for aspects of the problem not specified. For 
example, you could calculate mod(year, 3) and choose a start based on 
what kept most observations. Or, very irregular panels might be sampled by

bysort id (year) : keep if mod(year, 3) == mod(year[1], 3)

which for the second example below (different from the first) selects 
1986(3)1995.

Conversely, I don't understand why much more complicated solutions have 
been offered for this kind of problem. What am I missing?

Nick
n.j.cox@durham.ac.uk

Rajesh Tharyan

I have this ( a subset of the original dataset, the original dataset has
about 6000 ids with an  average of 6 years each)

     id   year
      4   1987
      4   1988
      4   1989
      4   1990
      4   1992
      4   1993
      4   1994
      9   1987
      9   1988
      9   1989
      9   1990
      9   1992
      9   1993
      9   1994

I need to keep years if they are more than 2 years apart by company. In an
earlier post

http://www.stata.com/statalist/archive/2006-02/msg00952.html

David Harrison suggested the following which works fine with one id


local refdate = year[1]
gen byte dropflag = 0
forvalues i = 2/`=_N' {
	if year[`i']-`refdate'<=2 {
		replace dropflag = 1 in `i'
	}
	else {
		local refdate = year[`i']
	}
}
drop if dropflag
drop dropflag

However I am not able to do it by id. what I would want is this..

     id   year
      4   1987
      4   1990
      4   1993
      9   1987
      9   1990
      9   1993

I tried using the above code with levelsof and foreach as shown below but no
luck.

*******************
input id year

          4   1987
          4   1988
          4   1989
          4   1990
          4   1992
          4   1993
          4   1994
          9   1986
          9   1987
          9   1988
          9   1989
          9   1990
          9   1992
          9   1993
          9   1994
          9   1995
          9   1996
          9   1997
end

levelsof id, local(levels)
foreach l of local levels {
qui sum year

local refdate = year[1]
gen byte dropflag = 0
forvalues i = 2/`=_N' {
	if year[`i']-`refdate'<=2 {
		replace dropflag = 1 in `i'
	}
	else {
		local refdate = year[`i']
	}
}
drop if dropflag
drop dropflag
}
list, table clean noobs

*******************

which gives me

     id  year
     4   1987
     4   1990
     4   1993
     9   1996

I tried a variation of the above as

**Snip**

levelsof id, local(levels)
foreach l of local levels {

qui sum year if id==`l'

local refdate = r(min)
local m= r(N)
gen byte dropflag = 0
forvalues i = 2/`m'{
	if year[`i']-`refdate'<=2 {
		replace dropflag = 1 in `i'
	}
	else {
		local refdate = year[`i']
	}
}
drop if dropflag & id==`l'
drop dropflag
}

**Snip**

That gives me

     id   year
      4   1987
      4   1990
      4   1993
      9   1995
      9   1996
      9   1997

Any pointers much appreciated

*
*   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/

*
*   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/