Due at classtime, Tuesday 4 April 2000
Set up a Stata program (do-file) to provide the empirical results requested. Hand in a copy of the program, annotated with your comments as warranted. The comments may be handwritten on the printout if they are clearly legible.
Use the Wooldridge PRISON dataset, available from within Stata via the command
use http://fmwww.bc.edu/ec-p/data/wooldridge/PRISON
This dataset contains 714 observations, longitudinal data on the following variables:
1. state alphabetical; DC = 9 2. year 80 to 93 3. govelec =1 if gubernatorial election 4. black proportion black 5. metro proportion in metro. areas 6. unem proportion unemployed 7. criv violent crimes per 100,000 8. crip property crimes per 100,000 9. lcriv log(criv) 10. lcrip log(crip) 11. gcriv lcriv - lcriv_1 12. gcrip lcrip - lcrip_1 13. y81 =1 if year == 81 14. y82 15. y83 16. y84 17. y85 18. y86 19. y87 20. y88 21. y89 22. y90 23. y91 24. y92 25. y93 26. ag0_14 prop. pop. 0 to 14 yrs 27. ag15_17 prop. pop. 15 to 17 yrs 28. ag18_24 prop. pop. 18 to 24 yrs 29. ag25_34 prop. pop. 25 to 34 yrs 30. incpc per capita income, nominal 31. polpc police per 100,000 residents 32. gincpc log(incpc) - log(incpc_1) 33. gpolpc lpolpc - lpolpc_1 34. cag0_14 change in ag0_14 35. cag15_17 change in ag15_17 36. cag18_24 change in ag18_24 37. cag25_34 change in ag25_34 38. cunem change in unem 39. cblack change in black 40. cmetro change in metro 41. pris prison pop. per 100,000 42. lpris log(pris) 43. gpris lpris - lpris[t-1] 44. final1 =1 if fnl dec on litig, curr yr 45. final2 =1 if dec on litig, prev 2 yrs
Fit and evaluate the following models. You will want to use 'iis' and 'tis' to convince Stata that these are panel data.
1. a. Log of violent crime rate = f(black, metro, polpc), with year dummies, via fixed effects (see xtreg). Discuss expected signs and findings.
b. Same model, via GLS random effects.
c. Perform the Hausman test and evaluate the results.
2. a. Log of property crime rate = f(black, metro, polpc), with year dummies, via fixed effects.
b. Same model, via GLS random effects.
c. Perform the Hausman test and evaluate the results.
3. Repeat #1 and #2 adding lpris (log of prison population per 100,000) to the fixed effects and random effects formulation. Do your judgments from the Hausman test results change? Why or why not?
4. Generate lagged values of gcriv and gcrip, the changes in the log crime rate variables. (Careful: this must be done for each state; see 'by'). Test OLS regression models in which gcriv = f(cag15_17, gpris, lgcrip) and gcrip = f(cag15_17, gpris, lgcriv). Comment on expected signs. What is being implied by the lagged variables' placement in these equations? Comment on their estimated coefficients. What does the constant term in these equations imply?
5.Reestimate the two OLS equations above using Zellner SURE (see sureg) with the corr and isure options. Comment on the SURE results versus the OLS results.