Using tsset for time-series - any alternative?

[Dani Vasquez]

Dear Statalist,

I am using three years of household data and I want to declare my dataset to be a time series. As I have more than one cell for (almost) every household ID, I cannot apply tsset. Accordingly, I have "repeated time values within panel" or in other words duplicates. However, I cannot delete these duplicates because my ID is for the whole household. Is there an alternative command I could use?

Thanks

[Clyde Schechter]

Well, what you are saying here is that you do not actually have panel data at the household level. The panel data is, it appears, at the person level. So you should find (or create) the person identifier and use that as the panel identifier in your -xtset- (or -tsset-) command.

[Nick Cox]

That's got to be xtset; tsset allows you to specify just a time variable, but not just an identifier.

[Dani Vasquez]

I see. Thank you, both! Does that mean that I can only specify both (ID and time) when I create a unique ID for each individual of my cross-sectional data? Can you recommend me how I can create unique IDs for 50,000 observations?

Thank you in advance

[Nick Cox]

If you're say that

xtset id time

or

tsset id time 

fails, then it's hard to see that a different identifier would help.

If still puzzled, then please read and act on FAQ Advice #12 and show us example data and code to make your question clearer.

[Dani Vasquez]

here are some variables to understand how my data is constructed:

[CODE]

clear
ID ID_Per Gender Age B1 B2 B3 year
 471 1 1 39 3  9 4 2009
 471 2 2 39 3  6 3 2009
 471 3 2 19 1 13 5 2009
 471 4 1 16 3  9 4 2009
 471 5 1 14 1  8 3 2009
 471 6 2 12 1  5 1 2009
 472 1 1 40 2  9 4 2009
 472 2 2 38 3 12 5 2009
 472 3 1 17 1 11 4 2009
 472 4 1 16 1 10 4 2009
 472 5 2 11 1  5 1 2009
 473 1 1 61 3  6 3 2009
 473 2 2 48 2  7 3 2009
 473 3 2 18 3 12 5 2009
 473 4 1 16 2  8 3 2009
 473 5 2 15 1  9 4 2009
 473 6 1 14 1  8 3 2009
 473 7 2 12 1  6 3 2009
 473 8 1 10 1  4 1 2009
 511 1 2 72 4  0 1 2010
 511 2 2 41 2  7 3 2010
 517 1 1 43 3  9 4 2010
 517 2 2 33 2  8 3 2010
 517 3 2 16 1 10 4 2010
 517 4 2 15 1  9 4 2010
 517 5 1 13 1  7 3 2010
 518 1 1 38 3 12 5 2010
 518 2 2 36 2  8 3 2010
 518 3 1 16 1 10 4 2010
 518 4 1 15 1  9 4 2010
 519 1 1 35 2 11 4 2010
 519 2 2 34 2 11 4 2010
 519 3 2 10 1  4 1 2010
1106 1 1 46 3 12 5 2011
1106 3 1 19 1 13 5 2011
1108 1 1 45 3 14 6 2011
1108 2 2 42 2  6 3 2011
1108 3 1 19 1 12 5 2011
1108 4 2 16 1 10 4 2011
1108 5 2 15 1  8 3 2011
1110 1 1 74 3  9 4 2011
1110 2 2 64 4  0 1 2011
1110 3 2 43 3  9 4 2011
1110 4 2 40 3 16 7 2011
1110 5 2 36 2 11 4 2011
1110 6 2 27 3 16 7 2011
1110 7 1 24 1 14 5 2011
1112 1 2 69 4  0 1 2011
1114 1 1 28 2 11 4 2011
1114 2 2 24 2  8 3 2011
1116 2 2 41 2  7 3 2011
end
format %ty year

(ID_Per = number of person of household ID, B1=school attendence, B2= years of schooling, B3= educ.attainment)


Though I do not have a real panel data set, I want to declare the data to be a timeseries.

[Nick Cox]

Thanks for the data example. You lost the input code in copying. That's easy brain surgery.

Whatever this is, it's not a pure timeseries. You can force it to be panel data after e.g.

egen ID_Joint = group(ID ID_Per), label

but that loses the main structure and seems artificial. I would

xtset ID

and use year as a covariate.