st: Re: patterns of panel data

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

--On Saturday, August 10, 2002 2:33 -0400 Guillermo wrote:

> I am working with panel data and I found the program xtpattern extremely
> useful. Basically, it creates a string variable for each observation with
> the patterns you would obtain by "xtdes" (describing) the data, sp say:
> 1.1..1
> indicates that a unit of observation (say, a household) is present on time
> periods 1,3 and 6.
> What I need now is to know how many contiguous observations there are, so
> for instance:
> 1.1..1 would be zero
> 111111 would be six
> 111.11  would be (perhaps) three (the max).
>
> The obvious thing would be to parse through the string generated by
> xtpattern... but I couldn't find an elegant way around this problem. Any
> ideas on this?
Dick Campbell responded that if we changed the . to 0 and treated the 
result as a binary number, we could get at the issue. Not quite sure this 
will do that (especially in a language like Stata's that is a bit short on 
bitwise operators--you really need a bitwise 'shift' here), since the issue 
is the number of consecutive 1's. But Nick Cox (Harry Potter's sidekick) 
has also provided a solution to that problem, known as
-spell-.

BUT whilst embedding spell in a hacked version of xtpattern, I noticed one 
leetle flaw in xtpattern. If a year (time period) is missing completely 
from your panel--which is sometimes the case--it is not picked up in the 
pattern. Quite reasonably, since xtpattern can only interpret what is ever 
present. But for Guillermo's purposes I think we better deal with that 
issue.

You may within Stata

. copy http://fmwww.bc.edu/ec-p/baum/xtpattern2.ado .

and you must also

. ssc install spell

If you don't have this installed. Then it will do this job, as

xtpattern2, gen(p2) con(consec)

where the 'con' argument will be a (new) integer variable containing the 
greatest number of consecutive observations, including any missing from all 
panels.

E.g. from a hacked version of the grunfeld data, xtpattern yields:

         firm      year                    p
 1.         1      1935  111111111111.111.11
 2.         1      1936  111111111111.111.11
 3.         1      1937  111111111111.111.11
 4.         1      1938  111111111111.111.11
 5.         1      1939  111111111111.111.11
 6.         1      1940  111111111111.111.11
 7.         1      1941  111111111111.111.11
 8.         1      1942  111111111111.111.11
 9.         1      1943  111111111111.111.11
10.         1      1944  111111111111.111.11
11.         1      1945  111111111111.111.11
12.         1      1946  111111111111.111.11
13.         1      1948  111111111111.111.11
14.         1      1949  111111111111.111.11
15.         1      1951  111111111111.111.11
16.         1      1953  111111111111.111.11
17.         1      1954  111111111111.111.11
18.         2      1935  11111111111.1..11.1
...
81.         5      1949  1111111111111111111
82.         5      1951  1111111111111111111
83.         5      1952  1111111111111111111
84.         5      1953  1111111111111111111
85.         5      1954  1111111111111111111


Note that 1950, which I have dropped from all years, is not flagged.
After running xtpattern2,

         firm      year                    p2     consec
 1.         1      1935  111111111111.11.1.11         12
 2.         1      1936  111111111111.11.1.11         12
 3.         1      1937  111111111111.11.1.11         12
 4.         1      1938  111111111111.11.1.11         12
 5.         1      1939  111111111111.11.1.11         12
 6.         1      1940  111111111111.11.1.11         12
 7.         1      1941  111111111111.11.1.11         12
 8.         1      1942  111111111111.11.1.11         12
 9.         1      1943  111111111111.11.1.11         12
10.         1      1944  111111111111.11.1.11         12
11.         1      1945  111111111111.11.1.11         12
12.         1      1946  111111111111.11.1.11         12
13.         1      1948  111111111111.11.1.11         12
14.         1      1949  111111111111.11.1.11         12
15.         1      1951  111111111111.11.1.11         12
16.         1      1953  111111111111.11.1.11         12
17.         1      1954  111111111111.11.1.11         12
18.         2      1935  11111111111.1...11.1         11
19.         2      1936  11111111111.1...11.1         11
...
81.         5      1949  111111111111111.1111         15
82.         5      1951  111111111111111.1111         15
83.         5      1952  111111111111111.1111         15
84.         5      1953  111111111111111.1111         15
85.         5      1954  111111111111111.1111         15

which properly notes the missing 1950 obs (and thus the length of p2 does 
not agree with that of p from xtpattern).

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/