Supplementary Materials for Replication:
"Educational Policy and Intergenerational Mobility"
by Tomoaki Kotera and Ananth Seshadri
February 2017

This readme.txt file contains detailed information to replicate results in the paper. Section 1 details the data used in the paper; while Section 2 is about the computer codes.

1. Data.

a. rankrank_county.dta: This data file contains rank-rank slope and population at the county level for Table 1. Data is available from

rank-rank slope: Table III of "rank_rank_chetty.xls" provides the data on rank-rank slope.
population: The folder named "population_county" contains population at the county level in each state.


b. school_district_information.dta. This data file contains

i. state: state name
ii. schooldistrict: school district name
iii. county: county name
iv. schoolyears: average school years at the county level. 
v. childincome: child income in each county (Data is available only at the county level)
vi. averageexp: average public school spending

The detailed description of the last three variables is as follows:

school years: Census 2000 School District Tabulation (STP2) Data give "Sex by Educational Attainment For the Population 25 Years and Over". In each download, the corresponding information in only one county is available. 
To compute average school years, we need to know how many school years are associated with each education attainment. We use "school_years_for_each_education_attainment.txt". This text file is available in the data file.     

child income: Table III of "rank_rank_chetty.xls" provides the data on median child income. We regard it as child income in each county.

averageexp: The folder named "public_school_spending" includes public school spending from 1990 to 1998. Total public school spending is defined as the sum of current spending on instruction, support services, and other elementary-secondary programs. Then, we take the average. 



2. Code

Code is written for Fortran, specifically, using the Intel Fortran compiler, circa 2011.

For estimation and baseline results, we use

i. nr.f90
ii. nrutil.f90
iii. nrtype.f90
iv. Minimization.f90
v. Baseline.f90

Use the above order in the compilation call.

To run the program, we need to import several text files. (Change file name if necessary)

i.   Wage rate: Computation of  wage rates is described in our manuscript. It is easy to compute from school_district_information.dta
ii.  average school years: This information is available from school_district_information.dta
iii. weight: The text file is contained in the same folder. We use population in school district as the weight. The folder named "population_school_district" in the data contains population at the school district level in each state
iv.  moments: The complete information is available in Appendix. We can also compute from school_district_information.dta
v.   location of school district (county level):  This information is available from school_district_information.dta
vi.  random component (used for a child's learning ability): We employ a standard normal distribution for random numbers 


Runtime hinges on the number of school districts in a state. In Colorado (number of school districts=118), it takes 3 minutes for each iteration. 
The main results in the paper are in the following text files. 

i.   estimated results: Result.txt
ii.  simulated moments: Moment_Simu.txt
iii. actual moments: Moment_Actual.txt
iv.  income slope coefficient: Slope.txt
v.   school district income: income_adult.txt
vi.  child income: income_child.txt
vii. public school spending: public.txt


For counterfactual simulations, we use Counterfactual1.f90, Counterfactual2.f90,and Counterfactual3.f90, respectively.