Panel Data: max( l(0/t).x )

[Christian Haefke]

Hi,
thanks for the super helpful forum. It seems, today for the first time, I cannot find a solution here. It would be great if you could please point me to one, or help.

I have a panel data set and would like to create a new variable that contains the maximum value over t previous values.
Consider t = 3:

Works:
qui gen x_3 = max(L0.x,L1.x,L2.x)

Does not work:
qui gen x_3b= max( L(0/3).x)
returns: unknown function L()

Background:
Data is per-second measurements with one file per day, each day is about 2.5G
Unfortunately I will need to do this for many values of t for every day over five years.
Eventually the particular values of t will be provided by the user in a numlist,
so I need to generate the input for max() automatically.

Thanks for your help!

Christian

[Leonardo Guizzetti]

It's not clear if you panel data (-xtset-) or survival time data (-stset-). I assume panel data, and I assume that you have no gaps in the timing variable for your panels.

In that case, if you really only care about a small, fixed window size (like 3 seconds in your example) you can hand code the varlist yourself. The other option doesn't require data to be -xtset- (though it's a good idea for some error checking), and uses -rangestat- (available from SSC).

input id time x
1 1 1
1 2 3
1 3 2
1 4 5
1 5 4
end

xtset id time

* option 1 - smal and fixed rolling interval (easily adaptable to a loop to build the varlist)
gen want1 = max(x, L1.x, L2.x)

* option 2 - using rangestat
sort id time
rangestat (max) want2=x, interval(time -2 0) by(id)

Result

     +-------------------------------+
     | id   time   x   want1   want2 |
     |-------------------------------|
  1. |  1      1   1       1       1 |
  2. |  1      2   3       3       3 |
  3. |  1      3   2       3       3 |
  4. |  1      4   5       5       5 |
  5. |  1      5   4       5       5 |
     +-------------------------------+

[Christian Haefke]

the rangestat command solves my problem. I learned a lot on the way.
Thanks Leonardo,