Hello friends,
I have panel data with repeated time values. The reason is that there are more than one observations for some years for some individuals. Variable spell counts for observations. The data is something like below
person_id spell year age wage cba
1364 1 2000 22 30 1
1364 2 2000 22 35 1
1364 3 2000 22 45 1
1364 1 2001 23 47 2
1364 1 2002 24 44 3
1365 1 2005 34 45 3
1365 2 2006 35 50 1
1366 3 2002 38 52 2
entry year is when spell==1, lets call it t0. Also, spell is uniquely identified when there are observations with repeated time values.
I want to regress wage 3 years after entry on age and cba at the entry (spell == 1) but time dummy is at the current year (same as wage, i.e. 3 years after entry). regression equation would be: wage_(t0+3) = b1age_t0 + b2 cba_t0 + year_dummy_(t0+3) if there is repeated time values for wage observation at t0+3 I would like to consider the first one appear in the data (i.e., the one with the smallest spell) for wage_t0+3.
Could you plz help me how I can do that? thanks.
Can I recommend that you proceed in two steps:
1) Data curation
You remove repeated data
Then you may still have missing data, "holes"
2) Data interpolation, extrapolation/prediction
if you have for person x wage (t) until t=t1 but you are missing wage(t) from t1+1 until t2-1. Then you have wage(t) for t=t2 to t3.
Interpolation is estimating likely wage(t) for t between t1+1 and t2-1.
It depends on the statistics if wage before and after, any curve fitting by maximum likelihood...
Extrapolation is where data stops after t1, then you have to predict wage(t)...
Less constraints, more risk.
Imagine we look at linear evolution, or exponential, or quadratic, etc.
You have to analyse the data of John over time, or the data of people supposed to be comparable to John, or the data of all people...
It's what data analysts do, with many more dimensions though :)
Hope it helps
Let me know
Hi!
If I understood correctly, you can do it using Stata in this way:
sort person_id year
keep if spell == 1
by person_id: gen age_3 = age[_n_3]
by person_id: gen cba_3 = cba[_n_3]
reg wage age_3 cba_3 i.year
Can I recommend that you proceed in two steps:
1) Data curation
You remove repeated data
Then you may still have missing data, "holes"
2) Data interpolation, extrapolation/prediction
if you have for person x wage (t) until t=t1 but you are missing wage(t) from t1+1 until t2-1. Then you have wage(t) for t=t2 to t3.
Interpolation is estimating likely wage(t) for t between t1+1 and t2-1.
It depends on the statistics if wage before and after, any curve fitting by maximum likelihood...
Extrapolation is where data stops after t1, then you have to predict wage(t)...
Less constraints, more risk.
Imagine we look at linear evolution, or exponential, or quadratic, etc.
You have to analyse the data of John over time, or the data of people supposed to be comparable to John, or the data of all people...
It's what data analysts do, with many more dimensions though :)
Hope it helps
Let me know
Hi. I think you should pair your observation with the year by using STATA.
Gegam Shagbazian Thank you for your practical solution, I solved the problem with a very similar approach. Actually, I found the way to solve problem by using the correct application of "sort" and "by sort" functions in Stata.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Regression