Dear all,
it´s my first time working with mi impute and I´m a bit confused how to use the data generated by mi impute.
I have got a large panel dataset with 12,000 individuals in which I want to impute missing data for the ears between 1986 and 2010.
My data structure looks like this:
Now I want to to implement a dynamic microsimulation model which uses the information I have in t-1 to impute whether a person changes their employment status in t.
The variable "transition" indicates whether a person changes their employment status in the next period.
So here in the example I want to use information from 1989 to impute the employment status in 1990.
Now I want to write a loop implementing this. My code so far looks like this:
But some things unfortunately do not work / I do not understand.
1. Since I set the imputation format to wide, it adds the variables _1_transition_work _2_transition_work ..... and only in _6_transition_work the actual value for transformation_work is implemented for t-1.
Why is that? Should I rather mi set my data to flong or mlong?
2. Then if I want to replace the missing information of transformation work in t-1 with the imputed value of _6_transition_work it does not work, because it sometimes said that this variable could not be find. And if I run the code with the replace command, _6_transition_work is not generated and just 5 imputation up to _5_transition_work were added to the dataset.
So basically my question is, if there is an easy way to straight replace the imputed value into transition_work, so that then I can update the employment status in t?
Thank you very much in advance!
it´s my first time working with mi impute and I´m a bit confused how to use the data generated by mi impute.
I have got a large panel dataset with 12,000 individuals in which I want to impute missing data for the ears between 1986 and 2010.
My data structure looks like this:
Code:
* Example generated by -dataex-. To install: ssc install dataex clear input long id int syear float(work_status flag_impu_work) byte(transition_work _mi_miss) float(_1_transition_work _2_transition_work _3_transition_work _4_transition_work) byte(_5_transition_work _6_transition_work) 01 1986 1 0 0 0 0 0 0 0 0 0 01 1987 1 0 0 0 0 0 0 0 0 0 01 1988 1 0 0 1 0 0 0 0 0 0 01 1989 1 1 . 1 . . . . . 0 01 1990 . 0 . 1 . . . . . . 01 1991 . 0 . 1 . . . . . . 02 1986 1 0 0 0 1 1 1 1 1 1 02 1987 1 0 1 0 0 0 0 0 0 0 02 1988 0 0 0 1 0 0 0 0 0 0 02 1989 0 1 . 1 . . . . . 0 02 1990 . 0 . 1 . . . . . . 02 1991 . 0 . 1 . . . . . . end
Now I want to to implement a dynamic microsimulation model which uses the information I have in t-1 to impute whether a person changes their employment status in t.
The variable "transition" indicates whether a person changes their employment status in the next period.
So here in the example I want to use information from 1989 to impute the employment status in 1990.
Now I want to write a loop implementing this. My code so far looks like this:
Code:
mi set wide
mi register imp transition_work
mi register regular age age2 age3 edu_highest migration region
*** START LOOP
forval y=1986/2012 {
capture drop flag_impu_work
gen flag_impu_work =0
#d;
sort pid syear;
by pid: replace flag_impu_work=1 if syear==(`y'-1) & work_status!=. & work_status[_n+1]==. & id==id[_n+1] & year_last_observation<=(`y') & syear>=entry+1;
mi impute logit transition_work = c.age c.age2 c.age3 i.edu_highest i.migration i.region if syear==`y'-1 & work_status!=. , noisily rseed (1234) add(1) augment iterate(100);
#delimit cr
* updating information
* replace missing value in t with imputed value
replace transtion_work = _6_transition_work if syear== `y' & transition_work ==. & _6_transition_work !=. & flag_impu_work==1
* update employment status
replace work_status=work_status[_n-1] if flag_impu_work==1 & work_status==. & transition_work[_n-1]!=. & transition_work==0
replace work_status=1 if flag_impu_work==1 & work_status==. & transition_work[_n-1]!=1 & transition_work[_n-1]==1
replace work_status=0 if flag_impu_work==1 & work_status==. & transition_work[_n-1]!=1 & transition_work[_n-1]==0
}
But some things unfortunately do not work / I do not understand.
1. Since I set the imputation format to wide, it adds the variables _1_transition_work _2_transition_work ..... and only in _6_transition_work the actual value for transformation_work is implemented for t-1.
Why is that? Should I rather mi set my data to flong or mlong?
2. Then if I want to replace the missing information of transformation work in t-1 with the imputed value of _6_transition_work it does not work, because it sometimes said that this variable could not be find. And if I run the code with the replace command, _6_transition_work is not generated and just 5 imputation up to _5_transition_work were added to the dataset.
So basically my question is, if there is an easy way to straight replace the imputed value into transition_work, so that then I can update the employment status in t?
Thank you very much in advance!
