tsset with two time variables

[Adrienne Wold]

Hi,
Im trying to fill an unbalanced panel where the observations are unique on an id and two time variables (monthly and year) of the form

* Example generated by -dataex-. To install: ssc install dataex
clear
input float(uniqueid year month)
1001 1990  1
1001 1990  2
1001 1990  3
1001 1990  4
1001 1990  5
1001 1990  6
1001 1990  7
1001 1990  8
1001 1990  9
1001 1990 10
1001 1990 11
1001 1990 12
1001 1991  1
1001 1991  2
1001 1991  3
1001 1991  4
1001 1991  7
1001 1991  8
1001 1991  9
1001 1991 10
1001 1991 11
1001 1991 12
1001 1992  1
1001 1992  2
1001 1992  4
1001 1992  5
1001 1992  6
1001 1992  7
1001 1992  8
1001 1992  9
1001 1992 10
1001 1992 11
1001 1992 12
1001 1993  1
1001 1993  2
1001 1993  3
1001 1993  4
1001 1993  5
1001 1993  8
1001 1993  9
1001 1993 10
1001 1993 11
1001 1994  1
1001 1994  2
1001 1994  3
1001 1994  5
1001 1994  7
1001 1994  8
1001 1994  9
1001 1994 10
1001 1994 11
1001 1994 12
1001 1995  1
1001 1995  2
1001 1995  3
1001 1995  4
1001 1995  5
1001 1995  6
1001 1995  7
1001 1995  8
1001 1995  9
1001 1995 10
1001 1995 11
1001 1996  1
1001 1996  2
1001 1996  3
1001 1996  4
1001 1996  7
1001 1996  9
1001 1996 10
1001 1996 11
1001 1996 12
1001 1997  1
1001 1997  2
1001 1997  3
1001 1997  4
1001 1997  5
1001 1997  7
1001 1997  8
1001 1997  9
1001 1997 10
1001 1997 12
1001 1998  1
1001 1998  2
1001 1998  3
1001 1998  4
1001 1998  5
1001 1998  6
1001 1998  7
1001 1998  8
1001 1998  9
1001 1998 10
1001 1998 11
1001 1998 12
1001 1999  1
1001 1999  2
1001 1999  3
1001 1999  5
1001 1999  8
1001 1999  9
end

[/CODE]

but the "tsset" command says too many variables are specified. Is the best way to simply create one year/month variable to address this?
Thank you.

[Nick Cox]

You need a single monthly date variable:

gen mdate = ym(year, month) 
format mdate %tm 
tsset uniqueid  mdate

[Adrienne Wold]

Thank you. I am then trying to use "tsfill" to fill the panel such that all id's have an observation for every month year combination from 2000-2010. But after doing so, I see that not every observation is covered. For example,

uniqueid mdate
1003 2007m12
1003 2008m1
1003 2008m2
1003 2008m3
1003 2008m4
1003 2008m5
1004 2002m10
1004 2002m11
1004 2002m12
1004 2003m1
1004 2003m2
1004 2003m3
1004 2003m4
1004 2003m5

For uniqueid 1004, there are no observations for years 2000-2001, and none for 2009-2010 for uniqueid 1003. Is tsfill not the correct approach for this?

[Nick Cox]

tsfill fills in gaps within panels, but never stretches panels beyond the first or last observation. This is explained in the help for tsfill.

tsfill uniqueid mdate, full

fillin uniqueid mdate

seems to be what you are reaching for, but to what purpose? It can only add missing values on all other variables to the dataset.