Alex Maynard and Jiaping Qiu, "Public Insurance and Private Savings: Who is Affected and by How Much?," Journal of Applied Econometrics, Vol. 24, No. 2, 2009, pp. 282-308. The file mq-files.zip contains the following files. Except for table1_rows1to10.dta, these are all ASCII files in DOS format. file size filename purpose ------------------------------------------------ 6685137 table1_rows1to10.csv ASCI SIPP data file 4113452 table1_rows1to10.dta STATA version of SIPP data file 2580 mquant.m matlab replication code 2283 Table1-2.do STATA replication code 882 Table3matrixforCHcode.do STATA replication code 866 Table5matrixforCHcode.do STATA replication code 998 Table6AmatrixforCHcode.do STATA replication code 923 Table6BmatrixforCHcode.do STATA replication code 1251 Table7.do STATA code The household asset composition data, used for Rows 1 to 10 of Table 1, can be found in the either the file "table1_rows1to10.csv" (in ASCII (text) format with commas separating columns) or "table1_rows1to10.dta" (in STATA format). This data comes from the Survey of Income and Program Participation (SIPP), using the years 1984 through 1993. The wealth composition information is available in the fourth and seventh waves of 1984, 1985 and 1986 (the first wealth supplement for 1985 was actually in the third interview), the fourth wave of 1987, 1990, and 1992, and the seventh wave of 1991. There are 52,706 household observations in total. Each household is on a separate row. The data contains 14 variables ordered in the following columns: column 1. suid: Sample unit identifier column 2. entry: Entry address ID column 3. pnum: Personal number column 4. panwave: Data wave and year column 5: mhhtnw: Net wealth column 6: totalassets: Total assets column 7: totaldebt: Total debt column 8: debtpositive: Indicator for households with positive debt column 9: hhscdbt: Secured debt column 10: hhusdbt: Unsecured debt column 11: hhtheq: Net equity in home column 12: hhvehcl: Net equity in vehicles column 13: netexeq: Net worth excluding home equity column 14: netexeqvehcl: Net worth excluding home and vehicle equity All variables have been deflated by 1987 constant dollars. For the remaining rows 11-15 of Table 1 and Tables 2-7, we use the same data used in: Jonathan Gruber & Aaron Yelowitz, 1999. "Public Health Insurance and Private Savings," Journal of Political Economy, vol. 107(6), pages 1249-1274, December. The raw Gruber-Yelowitz data used to form the control variates is also originally from SIPP and spans the years 1984-1993. However, this data was obtained from SIPP at an earlier date, and likely due to data revisions, there are some very minor discrepancies in household net worth between this data and the data in "Table_rows1to10.csv" used for the wealth composition. (The Gruber-Yelowitz data does not include wealth composition data). Also included in the raw Gruber and Yelowitz data set are the simulated values that are used to create the Medicaid Eligible Dollars (MED) variable and its instrument, the Simulated Medicaid Eligible Dollars (SIMMED) variable. These simulated values were constructed in previous work by Janet Currie and Jonathan Gruber in conjunction with the following two papers: Currie, Janet and Jonathan Gruber (1996a). Health insurance eligibility, utilization of medical care, and child health. Quarterly Journal of Economics 111. and Currie, Janet and Jonathan Gruber (1996b). Saving babies: The efficacy and cost of recent changes in the Medicaid eligibility of pregnant women. Journal of Political Economy 104. The raw Gruber-Yelowitz data are posted on Professor Aaron Yelowitz's website at: http://gatton.uky.edu/Faculty/yelowitz/index.html in the zip file called "gruberyelowitz.zip" located under the heading "Data for Gruber and Yelowitz (JPE 1999)" in the drop-down menu entitled "View Economics Papers, Data and Other Links." The zip file includes 23 raw SIPP data sets in STATA (.dta) format. It also contains a STATA do file called run01.do, which can be used to create the final-form data set employed in the regression and quantiles regressions from the raw data set. To create the final-form data set, first unzip "gruberyelowitz.zip." Then edit the program run01.do at line 7585 just below the lines: * Table 6, column 2; reg lhhtnw kp_com8 combo8 $hdvar $hhvar _I* (kp_com8b combo8b $hdvar $hhvar _I*); to add the following two new lines of code: keep hhtnw2 lhhtnw combo8 combo8b kp_com8 kp_com8b $hdvar $hhvar _I* save "c:\finaldata.dta" Then run run01.do. This will create the final-form data set that we employ in a file called finaldata.dta with 52,706 households arranged by row and the variables assorted by column with the following variable definitions: hhtnw2: household net-worth lhhtnw: ln(hhtnw2) combo8: Medicaid eligible dollars combo8b: SIMMED (IV for combo8) kp_com8: asset testing%G%@ combo8 interactions kp_com8b: asset testing%G%@combo8b interactions female: head if female age: age of head age2_100: age2/100 black: head is black white: head is black ed12: head is high school graduate ed15: head has some college ed16: head is college graduate married: head is married ed12sp: spouse is high school graduate ed15sp: spouse has some college ed12sp: spouse is college graduate fearn: household labor income Aa*: controls for age-gender group. Ageed*: controls for age-education group _Ifnp*: dummy variables for family size _Isy*: state, year fixed effects and state year interactions The STATA code used to calculate the summary statistics of Tables 1 and 2 is in the program file "Tables1-2.do". If one is only interested in the asset composition of Rows 1 to 10 in Table 1, he/she need only use the data file "Table_rows1to10.csv" and run Part A (comment out Part B) of the program "Tables1-2.do". Since the information for rows 11-15 of Table 1 and all rows of Table 2 is from the Gruber-Yelowitz data, if one wants to fully replicate Tables 1 and 2, he/she needs to merge "Table_rows1to10.csv" with the final-form data from Jonathan Gruber & Aaron Yelowitz, being sure to match rows on the following four variables: "suid entry pnum panwave." The order of the columns does not matter. Then save the merged data in a file called "mergedata.dta" and run the program "Table1-2.do". For the quantile regression results in Tables 3-7, we use the final-form data from Gruber-Yelowitz in conjunction with the instrumental variable quantile regression method of Chernozhukov, Victor and Christian Hansen (2005). "An IV model of quantile treatment effects," Econometrica 73, 245-262. The MATLAB code for the instrumental quantile regression is freely available and posted on Professor Christian Hansen's web page. It can be downloaded from the following URL: http://faculty.chicagogsb.edu/christian.hansen/research/iqrmat.zip To use Chernozhukov and Hansen's code, one needs to create 4 matrices: y, z, d, and x, where y is the dependent variable, z is instrumental variable, d is endogenous variable, and x contains the other control variables. The codes to create these matrices for Table 3, Table 5, and Panels A and B of Table 6 are in the STATA do files named "Table3MatrixforCHcode.do", "Table5MatrixforCHcode5.do", "Table6AMatrixforCHcode.do", and "Table6BMatrixforCHcode.do", respectively. Following this, the included MATLAB program file, "mquant.m", was used to load these variables and call Chernozhukov and Hansen's instrumental quantile regression procedures in order to produce the final instrumental quantile estimates. This code can be used for the instrumental quantile estimates in Tables 3, 5, and 6. The program itself should not need any change when moving from one of these tables to the next. What needs to be changed are the contents of the data files (y, z, d, and x) that this program loads. This is accomplished by re-composing these data by running the included STATA do file corresponding to the table in question. As explained in the program, there were a few cases when convergence was not obtained for the quantile in question. In this case the quantile was shifted by 0.01. Finally, the results in Table 7 are produced using the standard two-stage instrumental variable regression. The code is in the STATA do file named "Table7.do".