Re: st: running sum restarting after missing value

[n j cox <n.j.cox@durham.ac.uk>]

Just as a follow-up: Given that you have -tsset- your data

.  tsset id year
       panel variable:  id, 1 to 1
        time variable:  year, 1984 to 1989

and installed -tsspell- from SSC, then spells are
defined by -var1- being non-missing:

. tsspell , cond(!missing(var1))

This automatically produces three variables, by default
_seq, _spell, _end.

. l

     +---------------------------------------------------+
     | id   year   var1   sum_var   _seq   _spell   _end |
     |---------------------------------------------------|
  1. |  1   1984      1         1      1        1      0 |
  2. |  1   1985      1         2      2        1      0 |
  3. |  1   1986      1         3      3        1      1 |
  4. |  1   1987      .         .      0        0      0 |
  5. |  1   1988      1         1      1        2      0 |
     |---------------------------------------------------|
  6. |  1   1989      1         2      2        2      1 |
     +---------------------------------------------------+

You could then type

. bysort id spell : gen sum_var1 = sum(var1)

n j cox wrote:

> This is confusing. You say you want a running sum, but your example
> does not show one. I am going to trust your initial statement and
> ignore your example.
>
> -sum()- is trained to ignore missings. Usually this is s feature,
> but not for your purposes.
>
> One take on this is that you evidently regard runs of non-missing values
> as distinct spells. Identifying such spells explicitly would enable
> you to do something like this:
>
> by id spell : gen sum_var1 = sum(var1)
>
> and with one bound you are then home free.
>
> -tsspell- on SSC is one user-written tool for working with spells. I 
> took a step back from that to explain, in excruciating detail, how to do 
> it all (and more) from first principles in
>
> Cox, N.J. 2007. Identifying spells. Stata Journal 7(2): 249-265.
>
> In this particular problem there are various direct ways of
> restarting, without identifying spells of non-missing values as
> distinct spells. Here's one:
>
> gen sum_var1 = .
>
> bysort id (year) :
>     replace sum_var1 =
>         cond(missing(var1[_n-1]),
>             var1,
>             var1 + sum_var1[_n-1]))
>
> For sources on information on -cond()-, -search cond()-.
>
> In words, if the previous one is missing, the sum becomes the present 
> value; otherwise add the present value to the sum so far. Missings
> will map to missings on this rule.
>
> As usual, note that -replace- entails a previous -generate-.
>
> The connections between this work on "identifying spells" and work of 
> any other past or present students of Hogwarts Academy is allusive, 
> elusive and illusory.
>
> By the way, it seems that you have yet to read to the end of the
> Statalist FAQ:
>
> http://www.stata.com/support/faqs/res/statalist.html#spell
>
> Nick
> n.j.cox@durham.ac.uk
>
> Erasmo Giambona
>
> I am trying to create a running sum using the sum function by group.
> My problem is that I would like STATA to restart summing again after
> each missing value and match the total with all previous observations.
> For example:
>
> id yeara var1
> 1 1984 1
> 1 1985 1
> 1 1986 1
> 1 1987 .
> 1 1988 1
> 1 1989 1
>
> My output should look like
>
> id year var1 sum_var1
> 1 1984 1 3
> 1 1985 1 3
> 1 1986 1 3
> 1 1987 . .
> 1 1988 1 2
> 1 1989 1 2
>
> I tried the following, but it doesn't get me what I need.
>
> by id: gen sum_var1=sum(var1) if var1!=.
*
*   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/