Herman J. Bierens and Jose R. Carvalho, "Semi-Nonparametric Competing Risks Analysis of Recidivism", Journal of Applied Econometrics, Vol. 22, No. 5, 2007, pp. 971-993. The data set used in this paper comes from the Inter-University Consortium for Political and Social Research and was originally collected by the Bureau of Justice Statistics (BJS). The software used for the econometric analysis is EasyReg International, the free software package developed by the first author. This software package can be downloaded from the URL http://econ.la.psu.edu/~hbierens/EASYREG.HTM The SNP competing risks models have been estimated by EasyReg module SNPSURVIVAL2 (via Menu > Miscellaneous modules). The Logit models have been estimated by EasyReg module LIMDEP (via Menu > Single equation models > Discrete dependent variables models). The integrated conditional moment (ICM) tests have been conducted by re-estimating the Logit models as nonlinear regression models (via Menu > User-defined nonlinear models > Nonlinear regression), using the Logit estimation results as start values, and then selecting the option "Integrated Conditional Moment (ICM) test of model correctness" in the "What to do next?" window. The data involved are available in three formats: 1. Microsoft Excel CSV format (for US number setting of Windows) File name: Recidivism_Excel.csv 2. EasyReg text format File name: Recidivism_EasyReg.txt 3. Journal of Applied Econometrics text format File name: Recidivism_JAE.txt All three files, which are in DOS format, are zipped in the file bcdata.zip. The first two files can be imported directly in EasyReg (Only one of the two needs to be imported). The file Recidivism_JAE.txt contains the same data matrix as file Recidivism_EasyReg.txt, except that in file Recidivism_JAE.txt the variable names and corresponding columns are listed at the end of the file, whereas in file Recidivism_EasyReg.txt the variable names precede the data matrix. The first record of the latter file is "32 -99999", the number of variables and the missing value code. Thus, in the two data text files, Recidivism_EasyReg.txt and Recidivism_JAE.txt, missing values are indicated by the value -99999. Excel indicates a missing value as an empty data entry, i.e., a missing values in file Recidivism_Excel.csv is indicated by two adjacent commas (..,,..) or by a comma at the end of a record. These data files contain 16355 observations on the following 32 variables: Column 1: Observation number This is just a counter Column 2: CASE ID This the case ID assigned by the BJS to each of the 16355 ex-inmates. Column 3: TIME (days/1000) Time in days per 1000 between release from prison and the first arrest, or the length of follow-up period in the case of right-censoring. Column 4: C Right-censoring indicator: C = 1 if right-censored, i.e., no arrest has occurred in the follow-up period, C = 0 if arrested during the follow-up period. Column 5: F Arrest type indicator: F = 1 for felony arrest, F = 0 for misdemeanor arrest or right-censoring. Column 6: MALE Gender indicator: 1 = male, 0 = female. Column 7: BLACK Race indicator: 1 = African American, 0 = other race. Column 8: RELEASE Release indicator: 0 = released on parole or probation, 1 = otherwise. Column 9: AGE (days/1000) Age of the ex-inmate in days divided by 1000 at the time of release from prison. Column 10: SENT (days/1000) Length of the last sentence served, in days divided by 1000, before release from prison. Column 11: Dummy California State indicator: 1 = California, 0 = other state. Column 12: Dummy Florida State indicator: 1 = Florida, 0 = other state. Column 13: Dummy Illinois State indicator: 1 = Illinois, 0 = other state. Column 14: Dummy Michigan State indicator: 1 = Michigan, 0 = other state. Column 15: Dummy Minnesota State indicator: 1 = Minnesota, 0 = other state. Column 16: Dummy New Jersey State indicator: 1 = New Jersey, 0 = other state. Column 17: Dummy New York State indicator: 1 = New York, 0 = other state. Column 18: Dummy North Carolina State indicator: 1 = North Carolina, 0 = other state. Column 19: Dummy Ohio State indicator: 1 = Ohio, 0 = other state. Column 20: Dummy Oregon State indicator: 1 = Oregon, 0 = other state. Column 21: Dummy Texas State indicator: 1 = Texas, 0 = other state. Column 22: Select California Selection variable: 1 = California, missing value = other state. Column 23: Select Florida Selection variable: 1 = Florida, missing value = other state. Column 24: Select Illinois Selection variable: 1 = Illinois, missing value = other state. Column 25: Select Michigan Selection variable: 1 = Michigan, missing value = other state. Column 26: Select Minnesota Selection variable: 1 = Minnesota, missing value = other state. Column 27: Select New Jersey Selection variable: 1 = New Jersey, missing value = other state. Column 28: Select New York Selection variable: 1 = New York, missing value = other state. Column 29: Select North Carolina Selection variable: 1 = North Carolina, missing value = other state. Column 30: Select Ohio Selection variable: 1 = Ohio, missing value = other state. Column 31: Select Oregon Selection variable: 1 = Oregon, missing value = other state. Column 32: Select Texas Selection variable: 1 = Texas, missing value = other state. To estimate the state models, the variable TIME has to be multiplied by each selection variable (via Menu > Input > Transform variables > Multiplicative combination of variables), so that the new variable is TIME for the state involved, and a missing value for all other states. EasyReg will automatically skip any observation for which a model variable is a missing value. In addition to the data files, a separate appendix to the paper, recidivism_app.pdf, is also included. This appendix is also downloadable from URL http://econ.la.psu.edu/~hbierens/RECIDIVISM_APP.PDF