I am trying to run the following code for Grouped Fixed Effects that can be found on the link.
While running I get the error "file assignment.txt not found". Is it a bug or am I doing something wrong? Thank you for your time!
This is how my data look like:
And this is the code:
************************************************* ************************************************** **
* Generic code for the Grouped Fixed-Effects Estimator (GFE) *
* Bonhomme and Manresa, "Grouped Patterns of Heterogeneity in Panel Data", to appear in Econometrica *
************************************************** ************************************************** **
* Stata version 11.2
* The code computes the optimal grouping for individuals in a generic unbalanced panel dataset into G different groups.
* The calculation of the Grouped Fixed Effects (GFE) estimator is done by means of an executable FORTRAN file.
* Throughout the program XXX denotes inputs to be filled by the researcher.
* The user needs to do the following tasks before running this program:
* - Save the input dataset in the folder "GFE_generic" in stata format.
* - Make sure that the panel contains the same number of periods for all units, possibly with missing information (if unbalanced).
* - Fill in appropiately the XXX in the "USER INPUTS" part of this program. Specifically:
* -- Enter the path to the folder "GFE_generic" (line 56).
* -- Fill in the name of the dataset in "UPLOAD DATA" (line 61).
* -- Specify the individual/unit identifier as ID (line 64).
* -- Specify the time identifier as TIME (line 65).
* -- Label the outcome (dependent variable) as "OUTCOME" (line 74).
* -- Label the list of covariates (independent variables) as "COVARIATES" (line 78).
* Once this is done, the program is ready to be run in Stata.
* When the program runs, an executable file appears in order to calculate the GFE estimator.
* At this point, the researcher is asked to specify:
* - The number of groups G
* - The number of covariates (0 or greater). The number of covariates must match the dimension of "COVARIATES".
* - The type of algorithm: "0" for algorithm 1 in the paper (iterative), and "1" for algorithm 2 (variable neighborhood search)
* - The parameters of the algorithm:
* -- For algorithm 1: only the number of starting values (ex: 1000).
* -- For algorithm 2: the number of starting values, the number of neighbors, and the maximum number of iterations (ex: 10,10,10).
* - The type of standard errors (1 if bootstrapped), together with the computation algorithm (ex: 5,10,5 for Algorithm 2, with 100 replications).
* The executable file generates the "outputobj.txt" file, which contains:
* - The value of the objective function attained at each iteration in the algorithm (Nsim in the paper).
* - The estimate of the slope parameters (theta in the paper).
* Inspection of this file provides an idea of the numerical stability of the solution.
* Finally, the optimal grouping per unit is added to the input dataset as an additional variable named "assignment".
* The resulting dataset, containing all existing variables as well as the new "assignment" variable is saved as a new dataset named "DATA_GFE.dta".
* "DATA_GFE.dta" is contained in "GFE_generic".
* Bootstrapped standard errors (optional) can be found in "standard_errors_bootstrap"
* The code contains two optional parts:
* - The algorithm may be estimated in deviations to the mean, so as to remove time-invariant fixed-effects (line 80).
* - A selection criterion may be applied to remove units that have less than a prespecified number of time periods (line 112).
* Final word: PLEASE USE! ANY COMMENTS OR SUGGESTIONS ARE WELCOME! - [email protected], [email protected]
clear
set more off
*****************
* USER INPUTS *
*****************
* ENTER PATH HERE
cd "THE RESPECTIVE PATH"
* UPLOAD DATA
use "GFE_generic.dta", clear
* IDENTIFIERS (UNIT & PERIOD)
gen ID = individual
gen TIME = year
* PANEL DATA SETUP
tsset ID TIME
sort ID TIME
save DATA_inter, replace
* OUTCOME (TO BE FILLED IN)
* Example: "unab OUTCOME: Y"
unab OUTCOME: vote
* LIST OF COVARIATES (TO BE FILLED IN)
* Example: "unab COVARIATES: X1 X2"
unab COVARIATES: age race faminc education gender
* We construct deviation to the mean (optional)
/*
foreach vv of varlist `OUTCOME'{
bys ID (TIME): egen inter=mean(`vv')
replace `vv'=`vv'-inter
drop inter
}
foreach vv of varlist `COVARIATES'{
bys ID (TIME): egen inter=mean(`vv')
replace `vv'=`vv'-inter
drop inter
}
*/
********************
* AUTOMATIC CODE *
********************
* Define non-missing indicators, as well as a global missing indicator
gen nomis_tot=1
foreach vv of varlist `OUTCOME'{
gen nomis_`vv' = (`vv'~=.)
replace nomis_tot=nomis_tot*nomis_`vv'
}
foreach vv of varlist `COVARIATES'{
gen nomis_`vv' = (`vv'~=.)
replace nomis_tot=nomis_tot*nomis_`vv'
}
* Compute the number of non-missing observations, by ID
bys ID (TIME): egen sumnomis=sum(nomis_tot)
* Keep only if number of non-missing observations>=THRESH (optional)
/*
scalar THRESH = XXX
keep if sumnomis>THRESH
*/
* Recompute the total non-missing indicator
drop sumnomis
bys ID (TIME): egen sumnomis=sum(nomis_tot)
* Create a file with the unit identifiers
preserve
gen uno = 1
collapse (count) uno, by(ID)
drop uno
save ID_list.dta, replace
restore
* We use the convention that, when OUTCOME or any of the COVARIATES is missing,
* then all of them are set to zero
keep ID TIME nomis_tot sumnomis `OUTCOME' `COVARIATES'
foreach vv of varlist `OUTCOME'{
replace `vv'=0 if nomis_tot==0
}
foreach vv of varlist `COVARIATES'{
replace `vv'=0 if nomis_tot==0
}
* Store N and T_max (ie, the maximum number of periods per individual)
preserve
qui su ID, det
qui count if ID==r(min)
scalar Tmax=r(N)
count
scalar NTmax=r(N)
scalar N=NTmax/Tmax
gen var1 = N
gen var2 = Tmax
collapse var1 var2
outsheet using InputNT.txt, nonames replace
restore
* We print in txt file the pattern of missing data 5
preserve
keep ID TIME nomis_tot
sort ID TIME
reshape wide nomis_tot, i(ID) j(TIME)
drop ID
outsheet using Ti_unbalanced.txt, nonames replace
restore
sort ID TIME
keep `OUTCOME' `COVARIATES'
order `OUTCOME' `COVARIATES'
outsheet using data.txt, nonames replace
* We call the fortran executable in order to compute the GFE assignment
shell Bootstrap_version.exe
* We incorporate the group assignement into the original data
clear
insheet using assignment.txt
rename v1 assignment
merge 1:1 _n using ID_list.dta
drop _merge
merge 1:m ID using DATA_inter.dta
drop _merge ID TIME
************
* OUTPUT *
************
* Save data in DATA_GFE
save DATA_GFE.dta, replace
************
* CLEAN UP *
************
* We delete all auxiliary files created as input for the GFE.exe program
shell del data.txt InputNT.txt Ti_unbalanced.txt assignment.txt ID_list.dta
* We rename bootstrap output files: "replications.txt" contains the estimation of the replications and "standard_errors_bootstrap.txt" contains the standard errors.
shell ren assignment_bootstrap.txt replications.txt
shell ren outputobj_bootstrap.txt standard_errors_bootstrap.txt
While running I get the error "file assignment.txt not found". Is it a bug or am I doing something wrong? Thank you for your time!
This is how my data look like:
And this is the code:
************************************************* ************************************************** **
* Generic code for the Grouped Fixed-Effects Estimator (GFE) *
* Bonhomme and Manresa, "Grouped Patterns of Heterogeneity in Panel Data", to appear in Econometrica *
************************************************** ************************************************** **
* Stata version 11.2
* The code computes the optimal grouping for individuals in a generic unbalanced panel dataset into G different groups.
* The calculation of the Grouped Fixed Effects (GFE) estimator is done by means of an executable FORTRAN file.
* Throughout the program XXX denotes inputs to be filled by the researcher.
* The user needs to do the following tasks before running this program:
* - Save the input dataset in the folder "GFE_generic" in stata format.
* - Make sure that the panel contains the same number of periods for all units, possibly with missing information (if unbalanced).
* - Fill in appropiately the XXX in the "USER INPUTS" part of this program. Specifically:
* -- Enter the path to the folder "GFE_generic" (line 56).
* -- Fill in the name of the dataset in "UPLOAD DATA" (line 61).
* -- Specify the individual/unit identifier as ID (line 64).
* -- Specify the time identifier as TIME (line 65).
* -- Label the outcome (dependent variable) as "OUTCOME" (line 74).
* -- Label the list of covariates (independent variables) as "COVARIATES" (line 78).
* Once this is done, the program is ready to be run in Stata.
* When the program runs, an executable file appears in order to calculate the GFE estimator.
* At this point, the researcher is asked to specify:
* - The number of groups G
* - The number of covariates (0 or greater). The number of covariates must match the dimension of "COVARIATES".
* - The type of algorithm: "0" for algorithm 1 in the paper (iterative), and "1" for algorithm 2 (variable neighborhood search)
* - The parameters of the algorithm:
* -- For algorithm 1: only the number of starting values (ex: 1000).
* -- For algorithm 2: the number of starting values, the number of neighbors, and the maximum number of iterations (ex: 10,10,10).
* - The type of standard errors (1 if bootstrapped), together with the computation algorithm (ex: 5,10,5 for Algorithm 2, with 100 replications).
* The executable file generates the "outputobj.txt" file, which contains:
* - The value of the objective function attained at each iteration in the algorithm (Nsim in the paper).
* - The estimate of the slope parameters (theta in the paper).
* Inspection of this file provides an idea of the numerical stability of the solution.
* Finally, the optimal grouping per unit is added to the input dataset as an additional variable named "assignment".
* The resulting dataset, containing all existing variables as well as the new "assignment" variable is saved as a new dataset named "DATA_GFE.dta".
* "DATA_GFE.dta" is contained in "GFE_generic".
* Bootstrapped standard errors (optional) can be found in "standard_errors_bootstrap"
* The code contains two optional parts:
* - The algorithm may be estimated in deviations to the mean, so as to remove time-invariant fixed-effects (line 80).
* - A selection criterion may be applied to remove units that have less than a prespecified number of time periods (line 112).
* Final word: PLEASE USE! ANY COMMENTS OR SUGGESTIONS ARE WELCOME! - [email protected], [email protected]
clear
set more off
*****************
* USER INPUTS *
*****************
* ENTER PATH HERE
cd "THE RESPECTIVE PATH"
* UPLOAD DATA
use "GFE_generic.dta", clear
* IDENTIFIERS (UNIT & PERIOD)
gen ID = individual
gen TIME = year
* PANEL DATA SETUP
tsset ID TIME
sort ID TIME
save DATA_inter, replace
* OUTCOME (TO BE FILLED IN)
* Example: "unab OUTCOME: Y"
unab OUTCOME: vote
* LIST OF COVARIATES (TO BE FILLED IN)
* Example: "unab COVARIATES: X1 X2"
unab COVARIATES: age race faminc education gender
* We construct deviation to the mean (optional)
/*
foreach vv of varlist `OUTCOME'{
bys ID (TIME): egen inter=mean(`vv')
replace `vv'=`vv'-inter
drop inter
}
foreach vv of varlist `COVARIATES'{
bys ID (TIME): egen inter=mean(`vv')
replace `vv'=`vv'-inter
drop inter
}
*/
********************
* AUTOMATIC CODE *
********************
* Define non-missing indicators, as well as a global missing indicator
gen nomis_tot=1
foreach vv of varlist `OUTCOME'{
gen nomis_`vv' = (`vv'~=.)
replace nomis_tot=nomis_tot*nomis_`vv'
}
foreach vv of varlist `COVARIATES'{
gen nomis_`vv' = (`vv'~=.)
replace nomis_tot=nomis_tot*nomis_`vv'
}
* Compute the number of non-missing observations, by ID
bys ID (TIME): egen sumnomis=sum(nomis_tot)
* Keep only if number of non-missing observations>=THRESH (optional)
/*
scalar THRESH = XXX
keep if sumnomis>THRESH
*/
* Recompute the total non-missing indicator
drop sumnomis
bys ID (TIME): egen sumnomis=sum(nomis_tot)
* Create a file with the unit identifiers
preserve
gen uno = 1
collapse (count) uno, by(ID)
drop uno
save ID_list.dta, replace
restore
* We use the convention that, when OUTCOME or any of the COVARIATES is missing,
* then all of them are set to zero
keep ID TIME nomis_tot sumnomis `OUTCOME' `COVARIATES'
foreach vv of varlist `OUTCOME'{
replace `vv'=0 if nomis_tot==0
}
foreach vv of varlist `COVARIATES'{
replace `vv'=0 if nomis_tot==0
}
* Store N and T_max (ie, the maximum number of periods per individual)
preserve
qui su ID, det
qui count if ID==r(min)
scalar Tmax=r(N)
count
scalar NTmax=r(N)
scalar N=NTmax/Tmax
gen var1 = N
gen var2 = Tmax
collapse var1 var2
outsheet using InputNT.txt, nonames replace
restore
* We print in txt file the pattern of missing data 5
preserve
keep ID TIME nomis_tot
sort ID TIME
reshape wide nomis_tot, i(ID) j(TIME)
drop ID
outsheet using Ti_unbalanced.txt, nonames replace
restore
sort ID TIME
keep `OUTCOME' `COVARIATES'
order `OUTCOME' `COVARIATES'
outsheet using data.txt, nonames replace
* We call the fortran executable in order to compute the GFE assignment
shell Bootstrap_version.exe
* We incorporate the group assignement into the original data
clear
insheet using assignment.txt
rename v1 assignment
merge 1:1 _n using ID_list.dta
drop _merge
merge 1:m ID using DATA_inter.dta
drop _merge ID TIME
************
* OUTPUT *
************
* Save data in DATA_GFE
save DATA_GFE.dta, replace
************
* CLEAN UP *
************
* We delete all auxiliary files created as input for the GFE.exe program
shell del data.txt InputNT.txt Ti_unbalanced.txt assignment.txt ID_list.dta
* We rename bootstrap output files: "replications.txt" contains the estimation of the replications and "standard_errors_bootstrap.txt" contains the standard errors.
shell ren assignment_bootstrap.txt replications.txt
shell ren outputobj_bootstrap.txt standard_errors_bootstrap.txt
Comment