If my panel data contain studies appended from 10 different data sets, there are certain variables which shows duplicate values if I calculate incidence or prevalence of an event. I have explored the Data utilities option and tried reporting duplicate variables and drop variables. Any other better method to do this? If someone can share good literature on panel data representation in medical science, that will be of great help.
Thanks in advance.
A typical solution is to sort the data and then to remove duplicate observations. Assume that the your units are denoted by "unit" and the time is denoted by "year. Then you can use the following code:
sort unit year
drop if unit[_n]==unit[unit_n-1] & year[_n]==year[_n-1]
You would be removing those observations that appear more than once.
I hope it helps.
Thanks for the reply José-Ignacio Antón . This will surely help. But I would like to keep the observations rather than dropping it, as I need to calculate the number of times this observation repeated in a particular duration of time. For example. if a child is having multiple episodes of diarrhea in 6 months, Stata will count the child multiple times if we are estimating incidence. So that will be over estimation. And if I remove the multiple observation, other factors like fever, pneumonia episodes will also be lost.
It is not easy to guess how the data are from here.
If you need to count, you can create and auxiliary variable. For instance, let's say that you want to compute how many times a condition is registered.
gen aux_condition = 0
replace aux_condition = 1 if condition == 1
bysort unit: egen ncondition = total(aux_condition)
ncondition will be measuring how many times each unit has recorded "condition".
If you have many conditions you can use a loop:
foreach var in condition1 condition2 condition3 {
gen aux_`var' = 0
replace aux_`var' = 1 if `var' == 1
bysort unit: egen n`var' = total(aux_`var')
}
What are your panel and time variables? Over which dimension do you have repeated values?
Is the problem that you have several observations of the same person/ID/etc in one time interval, like for example multiple monthly observations when your time variable is years? If so, a solution like what José-Ignacio Antón , suggested, namely summing or counting data up from months to years, creating a single observation for the year with those sums, and then dropping the excess monthly observations is a good one.
If your data issue is not that, more details would help. :)
- Badges
- Science topic
- Similar topics
- Social Psychology
- Attitude