150 likes | 279 Views
Instrumenting Police Levels with Federal Law Enforcement Spending: An Assessment of Recent Efforts. John L. Worrall Tomislav V. Kovandzic. Police Levels and Crime. Police levels may be endogenous Reverse causality Methods of dealing with (or skirting) endogeneity
E N D
Instrumenting Police Levels with Federal Law Enforcement Spending: An Assessment of Recent Efforts John L. Worrall Tomislav V. Kovandzic
Police Levels and Crime • Police levels may be endogenous • Reverse causality • Methods of dealing with (or skirting) endogeneity • Time series w/ lagged police • Time series w/ creative interventions (e.g., terror alert levels—Klick and Tabarrok, 2005) • Granger causality test • IV Regression • Need instrument(s) for police levels
Review of Non-spending Instruments • Levitt (1997) • Electoral cycles • Levitt (2002) • Firefighters • Cornwell and Trumbull (2004) • Offense ratios and tax revenues • Other studies • Various combinations of demographic variables
Federal Spending (Grants) Instruments • GAO (2005) • 7 instruments (Hiring, MORE, Innovative COPS grants, Misc. COPS grants, Byrne, LLEBG, Non-cops grants) • Overidentified • Evans and Owens (2007) • “paid officers granted” instrument = .75*UHP + CIS • Just identified
Instrument Requirements • Two requirements for “good” instruments • Relevance (correlated with endogenous variable) • Significant in first stage regressions & joint F-stat. > 10 (Staiger & Stock, 1997) • Validity (independent of error process in main equation) • Hansen’s J (Hansen, 1982)
Our Contribution • Research questions • Do federal law enforcement grants make “good” (i.e., relevant & valid) instruments? • Are police levels associated with crime? • Answer to question 2 • Yes and no • Hiring instruments are better (generally relevant & valid) • Other federal grants make weak instruments • Answer to question 2 • Yes, but not as much as previously estimated
Data • Data • Panel of 5,199 cities, 1990-2001 • Dependent—7 index crimes (cleaned) • Independent • Police levels, UHP, CIS, DNP, Hiring, MORE, Innovative COPS grants, Misc. COPS grants, Byrne, LLEBG, Non-cops grants (grant draw-downs) • Income, Nonwhite, Under 24, Unemployment (county) • Unit and year dummies
Estimation Procedure • Analysis • GMM Models w/ -xtivreg2- in Stata • Analysis in levels & rates per 10,000 or percentages • Instruments lagged one period for delayed effect • Robust std. errors w/ corrections for state-level clustering • Models • Hiring grants (roughly in line with Evans and Owens) • UHP, DNP, CIS • All federal law enforcement grants (GAO’s focus) • Hiring, MORE, Innovative COPS grants, Misc. COPS grants, Byrne, LLEBG, Non-cops grants
Conclusion • Be cautious with grants as instruments • Grants may be correlated w/ unobservables (heterogeneity in grant-getting); lagged police may temper that effect