Construct random panel spell dataset based on structure of real data

[Henry Strawforrd]

I want to create a toy dataset for students with a good number of observations (ie too many to input completely by hand), whose data is completely fake (ie contains no personal information) but which captures the basic logical structure of my real administrative dataset. I may use

descsave

to get labels and summary stats from the real dataset.

Here is an example to illustrate the structure of the dataset

input id birthdate education spelltype spellbeg spellend establishment_id firmsize
1 01jan1970 10 unempl 01jan1990 30dec1990 . .
1 01jan1970 10 unempl 01jan1991 30dec1992 . .
1 01jan1970 11 empl 01jan1992 30dec1995 1 10
1 01jan1970 11 empl 01jan1996 30dec1998 2 100
1 01jan1970 11 empl 01jan1999 30dec1999 2 105
end

• id is an identifier for each person, fixed over time
• birthdate is an example of variable that would be constant over time within id
• education is an example of a variable that is time-varying (and has logical implication that it should not decrease over time)
• Then there is information about spells which can be either employment, inactivity or unemployment. This can be completely random.
• Each spell has beginning and end date (and here again there is logical implications that spell end > spell begin AND spell begin[_n+1] > spell end [_n])
• If the spell is an employment spell then there is another panel dimension, namely the establishment, which has an id that is fixed over time.
• Establishments also have characteristics that may be fixed or vary over time. Firmsize is an example of time-varying variable that may decrease or increase over time.

Is this way too difficult or possible?