Quarterly journal of economics
Yüklə 337,5 Kb. Pdf görüntüsü
|
|
Violent crime 77.11
83.18 66.13
Property crime 132.26
116.46 86.89
Murder 51.00
66.57 55.39
Prisoners per 1000 residents 2.83
1.26 0.86
Police per 1000 residents 2.85
0.64 0.27
State personal income per capita ($1997)
23207 3408
1361 AFDC generosity per recipient family (t–15) 7242
2905 1364
State unemployment rate (percent unemployed) 6.15 1.55
1.21 Beer consumption per capita (gallons) 23.03 3.32
1.24 Poverty rate (percent below poverty level) 13.80
3.51 1.64
Violent crime arrests per 1000, under age 25
3.18 1.46
0.49 Property crime arrests per 1000, under age 25 12.36
3.76 1.44
Murder arrests per 1000, under age 25 0.11 0.06 0.03
Violent crime arrests per 1000, age 25 and over 2.04 1.06
0.34 Property crime arrests per 1000, age 25 and over 4.82
1.58 0.65
Murder arrests per 1000, age 25 and over
0.06 0.03
0.01 All values reported are means of annual, state-level observations for the period 1985–1997 with the following exceptions. Arrest data cover the years 1985–1996, and AFDC generosity data are for the years 1985–1998. The police and prisons data are once-lagged, and thus correspond to the years 1984 –1996. The values reported in the table are population weighted averages. The effective abortion rate is a weighted average of the abortion rates for each cohort born in a state, with weights determined by the percentage of arrests by age for a given crime category in the United States in 1985 as shown in equation (1). All summary statistics are based on 663 observations, except where otherwise noted. Because of missing data, arrest statistics are based on 574 observations, compared with a theoretical maximum of 612. AFDC statistics are based on 714 observations. See Data Appendix for further details. 403
by the coef cients on abortion is substantial. An increase in the effective abortion rate of 100 per 1000 live births (the mean effective abortion rate in 1997 for violent crime is 180 with a standard deviation of 96 across states) is associated with a reduc- tion of 12 percent in murder, 13 percent in violent crime, and 9 percent in property crime. In Table II, comparing the states in the top third with respect to abortions to the states in the bottom third, our parameter estimates imply that crime fell an additional 16 –25 percent in the former states by 1997 due to greater usage TABLE IV
P ANEL
- DATA
E STIMATES OF THE R ELATIONSHIP BETWEEN A BORTION
R ATES AND
C RIME
Variable ln(Violent crime per capita)
ln(Property crime per capita) ln(Murder per capita) (1)
(2) (3)
(4) (5)
(6) “Effective” abortion rate ( 3 100)
2 .137
2 .129
2 .095
2 .091
2 .108
2 .121
(.023) (.024)
(.018) (.018)
(.036) (.047)
ln(prisoners per capita) (t 2 1) —
.027 — 2 .159 — 2 .231 (.044)
(.036) (.080)
ln(police per capita) (t 2 1) —
.028 — 2 .049 — 2 .300 (.045)
(.045) (.109)
State unemployment rate (percent unemployed) — .069
— 1.310
— .968
(.505) (.389)
(.794) ln(state income per capita) —
— .084
— 2 .098 (.213) (.162)
(.465) Poverty rate (percent below poverty line) — 2 .000 — 2 .001 — 2 .005 (.002)
(.001) (.004)
AFDC generosity (t 2 15) (
3 1000) — .008 — .002
— 2 .000 (.005) (.004)
(.000) Shall-issue concealed weapons law — 2 .004 — .039 — 2 .015 (.012) (.011)
(.032) Beer consumption per capita (gallons) — .004 — .004
— .006
(.003) (.003)
(.008) R 2 .938 .942 .990
.992 .914
.918 The dependent variable is the log in the per capita crime rate named at the top of each pair of columns. The rst column in each pair presents results from speci cations in which the only additional covariates are state- and year- xed effects. The second column presents results using the full speci cation. The data set is comprised of annual state-level observations (including the District of Columbia) for the period 1985–1997. The number of observations is equal to 663 in all columns. State- and year- xed effects are included in all speci cations. The prison and police variables are once-lagged to minimize endogeneity. Estimation is performed using a two-step procedure. In the rst step, weighted least squares estimates are obtained, with weights determined by state population. In the second step, a panel data generalization of the Prais-Winsten correction for serial correlation developed by Bhargava et al. [1982] is implemented. Standard errors are in parentheses. Data sources for all variables are described in the Data Appendix. 404
QUARTERLY JOURNAL OF ECONOMICS of abortion. One additional abortion is associated with a reduction of 0.23 property crimes, 0.04 violent crimes, and 0.004 murders annually when a cohort is at its peak crime age. Comparing these estimates to average criminal propensities among 18 –24 year olds, those on the margin for being aborted are roughly four times more criminal. These estimates are roughly consistent with, but somewhat larger than, the back-of-the-envelope predictions in Section III. The other coef cients in the model appear plausibly esti- mated. The elasticities of incarceration and police with respect to crime all carry the expected sign, with prison associated with signi cant reductions in property crime and murder, and police associated with signi cant reductions in murder. 28 A higher state unemployment rate is associated with signi cant increases in property crime, but not violent crime, consistent with previous research [Freeman 1995]. The three other measures of state economic conditions—per capita income, the poverty rate, and AFDC generosity (lagged fteen years to roughly correspond with the early years of life of the current teenagers) do not systemat- ically affect crime. Shall-issue concealed carry laws appear to signi cantly increase the amount of property crime, but have no effect on violent crime or murder. Finally, beer consumption is weakly linked with higher crime rates, but never signi cantly so. Table V investigates the sensitivity of the abortion coef - cients to a range of alternative speci cations. We take the spec- i cations with the full set of controls in Table IV as a baseline. The abortion coef cients from those regressions are reported in the top row of Table V. Each row of the table represents a different speci cation. The sensitivity of the results to large states (since the regressions are population weighted) and states with very high or low abortion rates is examined rst. Removing New York reduces the estimates for violent crime and murder, while eliminating California increases the abortion coef cient for those two crime categories. Dropping Washington, DC, which is an extreme outlier (with an abortion rate over four times the national average) increases the estimated impact of abortion. 28. The estimated effects of incarceration are consistent with previous cor- relational panel-data studies (e.g., Marvell and Moody [1994]). The prison coef - cients obtained here are approximately the same magnitude as Levitt [1996] nds when correcting for the endogeneity of the prison population using prison over- crowding litigation as an instrument. Levitt [1997] nds a negative impact of police on crime using electoral cycles in large cities as an instrument for the size of the police force. 405
Dropping all three of those high abortion states leads to higher estimates across the board, suggesting that the crime-reducing impact of abortion may have decreasing returns. Omitted variables may also be a concern in the regressions given the relatively limited set of covariates available. One crude way of addressing this question is to include region-year interac- tion terms in an attempt to absorb geographically correlated TABLE V
S ENSITIVITY OF A BORTION
C OEFFICIENTS TO A LTERNATIVE S PECIFICATIONS Speci cation Coef cient on the “effective” abortion rate variable when the dependent variable is ln (Violent crime per capita)
ln (Property crime per capita) ln (Murder per capita) Baseline
2 .129 (.024) 2 .091 (.018) 2 .121 (.047) Exclude New York 2 .097 (.030) 2 .097 (.021) 2 .063 (.045) Exclude California 2 .145 (.025) 2 .080 (.018) 2 .151 (.054) Exclude District of Columbia 2 .149 (.025) 2 .112 (.019) 2 .159 (.053) Exclude New York, California, and District of Columbia 2 .175 (.035) 2 .125 (.017) 2 .273 (.052) Adjust “effective” abortion rate for cross-state mobility 2 .148 (.027) 2 .099 (.020) 2 .140 (.055) Include control for ow of immigrants 2 .115 (.024) 2 .063 (.018) 2 .103 (.047) Include state-speci c trends 2 .078 (.080) .143 (.033) 2 .379 (.105) Include region-year interactions 2 .142 (.033) 2 .084 (.023) 2 .123 (.053) Unweighted 2 .046 (.029) 2 .022 (.023) .040 (.054) Unweighted, exclude District of Columbia 2 .149 (.029) 2 .107 (.015) 2 .140 (.055) Unweighted, exclude District of Columbia, California, and New York 2 .157 (.037) 2 .110 (.017) 2 .166 (.075) Include control for overall fertility rate (t 2 20) 2
2 .093 (.019) 2 .123 (.047) Long difference estimates using only data from 1985 and 1997 2 .109 (.054) 2 .077 (.034) 2 .089 (.077) Results in this table are variations on the speci cations reported in columns (2), (4), and (6) of Table IV. The top row of the current table is the baseline speci cation that is presented in Table IV. Except where noted, all speci cations are estimated using an annual, state-level panel of data for the years 1985–1997. Standard errors (in parentheses) are corrected for serial correlation using the Bhargava et al. [1982] two-step procedure for panel data. The speci cation that corrects for cross-state mobility does so by using an effective abortion rate that is a weighted average of the abortion rates in the state of birth for fteen year-olds residing in a state in the PUMS 5 percent sample of the 1990 census. Controls for the ow of immigrants are derived from changes in the foreign-born population, based on the decennial censuses and 1997 estimates, linearly interpolated. Region-year interactions are for the nine census regions. 406
QUARTERLY JOURNAL OF ECONOMICS shocks. The abortion coef cients are not substantially affected by this approach. Since we are measuring the effect of abortions in a state on crime in that state up to a quarter century later, the issue of cross-state mobility should be considered. Theoretically, the pres- ence of such cross-state movements will tend to systematically bias the abortion coef cient toward zero since the true effective abortion rate is measured with error by our proxy that ignores mobility. In order to adjust for migration, we determined the state of birth and state of residence for all fteen year-olds in the 1990 PUMS 5 percent sample. Using this information, we recal- culated effective abortion rates as weighted average abortion rates by the actual state of birth of fteen year-olds residing in a state. For all three crime categories the estimated impact of abortion increases with the migration correction, although the changes are not large. We perform a range of other sensitivity checks. Controlling for the ow of immigrants to a state somewhat reduces the estimated effect of abortion on crime (particularly for property crime), but it does not change their signi cance. When we include state-speci c time trends, the estimates change somewhat errat- ically, and the standard errors double for murder and property crime and triple for violent crime. Unweighted panel data regres- sions (as opposed to population weighted) yield sharply smaller coef cients, but this is exclusively due to Washington, DC as an outlier (owing in all likelihood to mismeasurement in the DC abortion rate). Excluding District of Columbia alone, or District of Columbia in combination with California and New York, leads to coef cients from the unweighted regressions that are greater than the baseline estimates. Including controls for lagged changes in overall fertility rates for the same era as our abortion measures has almost no impact on our estimated coef cients. Regressions using only the 1985 and 1997 endpoints of our sample (“long-differences”) yield coef- cients similar to, although somewhat smaller than, the baseline coef cients for the overall panel. V. T HE
MPACT OF A BORTION ON A RRESTS BY A GE OF
O FFENDER
The preceding section highlighted a strong empirical corre- lation between abortion rates after Roe v. Wade and crime changes in recent years. In this section we explore the extent to 407
LEGALIZED ABORTION AND CRIME which arrest patterns substantiate a possible causal interpreta- tion of these results. In particular, if legalized abortion is the reason for the decline in crime, then one would expect that de- creases in crime should be concentrated among those cohorts born after abortion is legalized. 29 Testing that hypothesis is complicated by the fact that the age of criminals is not directly observable. The age of arrestees, however, is reported. 30 Thus, we can analyze whether arrests by cohort are a function of the abortion rate. The basic speci cations used to explain state arrest rates by age category are identical to the crime regressions in the preced- ing section, except that the dependent variable is the (natural log of the) arrest rate per capita for those under age 25 rather than the overall crime rate for all ages, and 1997 is excluded from the sample because the necessary arrest data are not yet available. 31 Results from the estimation are reported in columns 1–3 of Table VI. Two speci cations per crime category are presented: the top row of results just includes the effective abortion variable and year- and state- xed effects, while the bottom row adds to these the remaining covariates that were used in Table IV above. Be- cause the dependent variable is denominated by the population under age 25, the abortion coef cients only re ect changes in arrest rates per person. If the impact of abortion was solely through changes in cohort size, then the per capita speci cations we run would yield zero coef cients on the abortion variable. In all six cases, lagged abortion rates are associated with decreases in arrests per capita by those under the age of 25, with estimates 29. It is possible that crime by older cohorts may be affected indirectly by abortion. For instance, if there are fewer criminals in younger cohorts, this may increase additional criminal opportunities for older individuals (particularly in activities such as drug distribution where there may be easy substitutability). On the other hand, to the extent that lower crime by the young increases the criminal justice resources available per older criminal [Sah 1991], crime among older cohorts may also fall. Moreover, as noted above, if abortion results in smaller family sizes and a concomitant increase in parental resources per child, the effect of legalization could be observed in crime reductions for older siblings. All of these effects are likely to be of second-order magnitude, however. 30. Arrest data may not accurately re ect criminal activity for a number of reasons. Greenwood [1995] argues that juvenile crime is more likely to be com- mitted in groups so that the arrest frequency of juveniles overstates the true fraction of crime they commit. Also, if there are differences across criminals in avoiding detection, arrests will be skewed toward the less pro cient criminals. 31. We use an age cutoff of 25 because it is approximately the age of the oldest cohorts affected by legalized abortion. Arrest data are available by single year of age up to age 24, but only in ve-year groupings thereafter. The results presented are not sensitive to small perturbations of the age groupings. 408
T A B L E V I T H E I M P A C T O F A B O R T IO N R A T E S O N A R R E S T S B Y A G E (A L L V A L U E S IN T H E T A B L E A R E C O E F F IC IE N T S O N T H E E F F E C T IV E A B O R T IO N R A T E (3 10 0) , O T H E R C O E F F IC IE N T S A R E N O T R E P O R T E D ) S pe ci ca ti on ln (a rr es t pe r pe rs on ,u nd er ag e 25 ) ln (a rr es ts pe r pe rs on ,a ge 25 + ) ln (a rr es ts pe r pe rs on ,u nd er ag e 25 ) m in us ln (a rr es ts pe r pe rs on , ag e 25 + ) V io le n t cr im e P ro pe rt y cr im e M ur de r V io le n t cr im e P ro pe rt y cr im e M u rd er V io le n t cr im e P ro pe rt y cr im e M u Yüklə 337,5 Kb. Dostları ilə paylaş:
|