st: Re: transforming panel data

[Christopher F Baum <baum@bc.edu>]

--On Saturday, September 28, 2002 2:33 -0400 Paulo wrote:

> Hi Stata users,
>
> Please, I need help to change the dataset below format. Note that the
> reshape command is not enough to do this transformation.
>
>             Id          period      others variables
>   1.         1          1996
>   2.         1          1996
>   3.         1          1996
>   4.         1          1997
>   5.         1          1997
>   6.         1          1997
>   7.         1          1998
>   8.         1          1998
>   8.         1          1999
>   9.         1          1999
>  10.         1          2000
>  11.         1          2000
>  12          1          2000
>  13          1          2001
>  14          1          2001
>  15          2          1996
>  16.         2          1996
>  17.         2          1996
>  18.         2          1997
>  19          2          1997
>  20          2          1997
>  21          2          1998
>  22          2          1998
>  23          .          1999
>  24          .          1999
>  25          2          2000
>  26          2          2000
>  27          2          2000
>  28          2          2001
>  29          2          2001
>  30          3          1996
>  31          3          1996
>  32          3          1996
>  33          3          1997
>  34          3          1997
>  35          3          1997
>  36          3          1998
>  37          3          1998
>  38          3          1999
>  39          3          1999
>  40          3          2000
>  41          3          2000
>  42          3          2000
>  43          .          2001
>  44          .          2001
>  45          4          1996
>  46          4          1996
>  47          4          1996
>  48          4          1997
>  49          4          1997
>  50          4          1997
>  51          4          1998
>  52          4          1998
>  53          4          1999
>  54          4          1999
>  55          4          2000
>  56          4          2000
>  57          4          2000
>  58          4          2001
>  59          4          2001
>
>
> ***************************************************************
>  ***** I need the data in this format (with the year controlling the time
> series operator) ********************
>
>             id         period      others variables
>   1.         1          1996
>   2.         1          1997
>   3.         1          1998
>   4.         1          1999
>   5.         1          2000
>   6.         1          2001
>   7.         1          1996
>   8.         1          1997
>   9.         1          1998
>  10.         1          1999
>  11.         1          2000
>  12.         1          2001
>  13.         1          1996
>  14.         1          1997
>  18          .          1998
>  19          .          1999
>  20          1          2000
>  21          .          2001
THere is no way of knowing why the multiple observations for 1996 in the 
original data should be placed in the order of the transformed data. E.g. 
if those observations were

           id         period  class ...
 1.         1          1996    a
 2.         1          1996    b
 3.         1          1996    c
 4.         1          1997    a
 5.         1          1997    b
 6.         1          1997    c

Then the desired result would appear from 'sort id class period'. If this 
is the case (and it must be, since there are multiple periods per id) then 
to 'tsset' these data, and use the time series operator, your panel 
variable is not merely id -- it is the combination of id and class. You 
must generate a new integer indicator that uniquely identifies, e.g., obs. 
1 and 4 in my example as belonging to the same group. Starting with 
(imagine var5 contains your actual data of interest):


        var2      var3       var4      var5
 1.        1      1996          a         1
 2.        1      1996          b         2
 3.        1      1996          c         3
 4.        1      1997          a         4
 5.        1      1997          b         5
 6.        1      1997          c         6
 7.        2      1996          a         7
 8.        2      1996          b         8
 9.        2      1996          c         9
10.        2      1997          a        10
11.        2      1997          b        11
12.        2      1997          c        12



. encode var4,gen(class)    required to generate an integer for tsset

. drop var4

. egen tsid=group(var2 class)

. sort tsid var3

. tsset tsid var3
      panel variable:  tsid, 1 to 6
       time variable:  var3, 1996 to 1997


. list

        var2      var3      var5     class       tsid
 1.        1      1996         1         a          1
 2.        1      1997         4         a          1
 3.        1      1996         2         b          2
 4.        1      1997         5         b          2
 5.        1      1996         3         c          3
 6.        1      1997         6         c          3
 7.        2      1996         7         a          4
 8.        2      1997        10         a          4
 9.        2      1996         8         b          5
10.        2      1997        11         b          5
11.        2      1996         9         c          6
12.        2      1997        12         c          6

I believe this is doing what you describe.

Kit


*
*   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/