Quarterly journal of economics
Yüklə 337,5 Kb. Pdf görüntüsü
|
|
rd er E ff ec ti ve ab or ti on ra te (3 10 0) on ly ,n o co va ri at es in cl ud ed 2 .0 95 2 .0 85 2 .2 14 .0 22 2 .0 19 2 .0 34 2 .1 16 2 .0 66 2 .1 80 (. 02 9) (.0 23 ) (.0 51 ) (. 05 4) (.0 37 ) (. 03 7) (.0 42 ) (.0 23 ) (. 03 4) E ff ec ti ve ab or ti on ra te (3 10 0) ,i nc lu di ng fu ll se t of co va ri at es 2 .0 44 2 .0 54 2 .1 80 .0 33 .0 08 2 .0 36 2 .0 62 2 .0 63 2 .1 37 (. 03 0) (.0 23 ) (.0 62 ) (. 04 6) (.0 31 ) (. 05 0) (.0 34 ) (.0 19 ) (. 04 6) R eg re ss io ns ar e id en ti ca lt o th os e in T ab le IV ,e xc ep t th at th e de pe n de nt va ri ab le s ar e ar re st ra te s br ok en do w n by ag e ca te go ry in st ea d of ov er al l cr im e ra te s. T he to p ro w of th e ta bl e pr es en ts re su lt s fr om sp ec i ca ti on s in w hi ch th e on ly ad di ti on al co va ri at es ar e st at e- an d ye ar - xe d ef fe ct s. T h e bo tt om ro w of th e ta bl e pr es en ts re su lt s us in g th e fu ll sp ec i ca ti on .C ov ar ia te s in cl ud ed in th e bo tt om ro w ar e on ce -l ag ge d po lic e an d pr is on er s pe r ca pi ta in lo gs ,s ta te un em pl oy m en t ra te ,l og ge d st at e in co m e pe r ca pi ta ,t he po ve rt y ra te , la gg ed A F D C ge ne ro si ty ,s ha ll -i ss u e co nc ea le d w ea po ns la w ,a nd be er co n su m pt io n pe r ca pi ta .T he re gr es si on s us e an n ua ls ta te -l ev el da ta fo r th e pe ri od 19 85 –1 99 6 (1 99 7 ar re st da ta by ag e ar e no t ye t av ai la bl e) .B ec au se of m is si ng da ta ,t h e nu m be r of ob se rv at io ns va ri es ac ro ss co lu m ns be tw ee n 55 5 an d 55 7, co m pa re d w it h a th eo re ti ca lm ax im um of 61 2. St at e- an d ye ar - xe d ef fe ct s ar e in cl ud ed in al ls pe ci ca ti on s. T he pr is on an d po li ce va ri ab le s ar e on ce -l ag ge d to m in im iz e en do ge ne it y. E st im at io n is pe rf or m ed us in g a tw o- st ep pr oc ed ur e. In th e rs t st ep ,w ei gh te d le as t sq ua re s es ti m at es ar e ob ta in ed ,w it h w ei gh ts de te rm in ed by st at e po pu la ti on .I n th e se co nd st ep ,a pa ne ld at a ge ne ra li za ti on of th e P ra is -W in st en co rr ec ti on fo r se ri al co rr el at io n de ve lo pe d by B ha rg av a et al .[ 19 82 ] is im pl em en te d. S ta nd ar d er ro rs ar e in pa re n th es es . 409 LEGALIZED ABORTION AND CRIME ranging between 2 .044 and 2 .214. The abortion coef cient is statistically signi cant in ve out of six speci cations. If the arrest data are measured without error and there are no spillovers between the crime of the young and the old, then we would not expect legalized abortion to affect the crime of those born prior to the law change. Columns 4 – 6, which relate arrest rates of older cohorts to abortion rates, thus provide a natural speci cation test for our hypothesis. In none of the crime catego- ries does the abortion rate variable have a statistically signi cant impact on arrests of older cohorts. In three instances the coef - cient is positive; in the other three cases the coef cient is nega- tive. All of the estimates are much smaller in magnitude than was the case for arrests of those under the age of 25. The last three columns of the table show “difference in differences” estimates of the impact of abortion on cohorts born after legalization relative to those born before. In all cases, the coef cients are similar to those in the rst three columns of the table. This result strength- ens the causal interpretation of the abortion coef cients on the arrest patterns of the young. The implied magnitude of the abortion effects on arrests is smaller than the parallel estimates presented in the preceding section analyzing crime rates, but is of the same order of magni- tude. On average, about half of those arrested are under the age of 25.
32 Thus, to generate the crime reduction in Table IV requires coef cients on young arrests that are twice as large as the coef- cients on overall crime. With the exception of murder, the arrest coef cients are actually smaller than the crime coef cients. Part of this discrepancy may be attributable to the fact that the arrest regressions re ect only reductions in per capita crime by the young, not smaller youthful cohorts, but this can explain only a portion of the gap. It remains an open question as to whether this discrepancy represents a partially spurious relationship in the crime regressions, measurement error in the arrest data, or a relationship between crime and arrests that is not proportional. It is important to stress, however, that while the magnitude of the effects differs between the crime and arrest regressions, the basic story with respect to abortion is present in both cases. 33 32. Over the sample period, those under the age of 25 accounted for an average of 49 percent of violent arrests, 62 percent of property arrests, and 48 percent of murder arrests. 33. We replicated the sensitivity tests that were presented in Table V for the baseline Table IV regressions using Table VI as the baseline estimates. These 410
As a further test of our hypothesis, we analyze arrest rates by state by single year of age. These data are available for the ages 15 and 24 covering the period 1985 through 1996. If abortion legalization reduces crime, then we should see the reduction begin with, say, fteen year-olds about sixteen years after legal- ization, then extend to sixteen year-olds a year later, and so on. Because we observe many cohorts in a given state and year, we are able to include controls for state-year variation. Thus, unlike the preceding table, where state-year variation was our source of identi cation, in the analysis that follows our estimates are based on differences in abortion rates and crime rates across cohorts within a given state and year. The regression we run takes the following form: (3)
ln ( ARRESTS stb ) 5 b 1
sb 1 g s 1 l tb 1 u st 1 e stb , where s, t, and b index state, year, and birth cohort, respectively. The variable ARRESTS is the raw number of arrests for a given crime. Unlike previous tables, we do not divide arrests by popu- lation to create per capita rates because of the absence of reliable measures of state population by single year of age. As our mea- sure of the abortion rate for a particular cohort, we use the abortion rate in the current state of residence in the calendar year most likely to have preceded the arrestees birth. 34 Cross-state migration will not be captured by this measure, but the results in earlier sections suggest that the impact of migration on the esti- mates is small (and that any migration correction would, if any- thing, strengthen our results). Because the unit of observation in the analysis is a state-birth cohort and cohorts are observed repeatedly over time, we will include controls for age, national year-cohort interactions, state-year interactions, and (in some cases) state-age interactions. We cannot, however, include state- regressions again revealed the robustness of the coef cient estimates, exhibiting patterns similar to the sensitivity analysis for the full sample. These results are available from the authors on request. 34. For example, we use the abortion rate in 1980 to re ect the abortion exposure of fteen year-olds arrested in 1996. Because the arrest data cover a calendar year, there is a possible 730-day window into which an arrestee’s date of birth may fall (i.e., an arrest is made on January 1 of someone who is 16 years and 364 days old versus an arrest is made on December 31 of someone who is 16 years and 1 day old). With a six-to-seven-month lag from likely time of abortion to time of birth, this 730-day window is centered on the calendar year that we use to capture abortion exposure. More complicated attempts to measure abortion expo- sure yield estimates similar to the ones we present. 411
birth cohort interactions without absorbing all of the variation in the abortion exposure of a state-birth cohort. Table VII presents the results of this analysis for violent crime and property crime. There are too few murder arrests per single age category per state to enable us to provide similar estimates for murder. We present estimates restricting the im- pact of abortion to be constant over the entire age range (odd columns) and allowing the impact of abortion to vary by age (even columns). Some of the regressions include state-age interactions, others just have state- xed effects. All of the speci cations in- clude year-age interactions to control for national-level uctua- tions in the age-crime pro le. 35 In all cases, standard errors have been corrected to re ect correlation over time in a given birth cohort’s observations. The top row of Table VII presents estimates restricting the abortion coef cient to be constant across the ages 15–24. In all instances, the coef cient is strongly signi cantly negative, imply- ing that higher abortion rates around the time a cohort is born are associated with lower arrest rates in their teens and twenties. When the abortion coef cient is allowed to vary by age, 38 of the 40 parameter estimates are negative; more than two-thirds of these estimates are statistically signi cant at the .05 level. The greatest impact of abortion appears to occur in the age range 18 –22. The effects are generally weakest for the youngest ages in the sample. The coef cients in this table are not directly comparable to those in the preceding tables. Because we are analyzing arrests by single year of age in this table, we are able to use actual abortion rates as opposed to the effective abortion rates that average over many cohorts. Comparing states in the top third and bottom third with respect to abortion frequency, the gap between those sets of states in actual abortion rates was about 350 per 1000 births. Given the estimates in the top row of Table VII, this implies that arrest rates of 15–24 year-olds in the high abortion states are estimated to have fallen between 5 and 14 percent relative to the low abortion states. 35. For instance, the arrival of crack appears to have temporarily raised the violent crime propensities, particularly among youths. 412
T A B L E V II T H E R E L A T IO N S H IP B E T W E E N A B O R T IO N R A T E S A N D A R R E S T R A T E S , B Y S IN G L E Y E A R O F A G E ln (V io le nt ar re st s) ln (P ro pe rt y ar re st s) A bo rt io n ra te (3 10 0) 2 .0 15 — 2 .0 28 — 2 .0 40 — 2 .0 25 — (. 00 3) (. 00 4) (.0
04 ) (.0 03 ) A bo rt io n ra te (3 10 0) in te ra ct ed w it h A ge = 15 — .0 18 — 2 .0 08 — 2 .0 37 — 2 .0 05 (. 00 8) (. 01 0) (. 00 7) (.0 08 ) A ge = 16 — .0 08 — 2 .0 07 — 2 .0 43 — 2 .0 11 (. 00 7) (. 00 8) (. 00 6) (.0
06 ) A ge = 17 — 2 .0 10 — 2 .0 21 — 2 .0 42 — 2 .0 13 (. 00 6) (. 00 7) (. 00 6) (.0
05 ) A ge = 18 — 2 .0 35 — 2 .0 39 — 2 .0 53 — 2 .0 23 (. 00 4) (. 00 7) (. 00 5) (.0
05 ) A ge = 19 — 2 .0 40 — 2 .0 43 — 2 .0 50 — 2 .0 36 (. 00 5) (. 00 7) (. 00 5) (.0
06 ) A ge = 20 — 2 .0 43 — 2 .0 43 — 2 .0 38 — 2 .0 35 (. 00 6) (. 00 7) (. 00 6) (.0
06 ) A ge = 21 — 2 .0 39 — 2 .0 39 — 2 .0 28 — 2 .0 37 (. 00 9) (. 00 8) (. 00 6) (.0
06 ) A ge = 22 — 2 .0 28 — 2 .0 24 — 2 .0 20 Yüklə 337,5 Kb. Dostları ilə paylaş:
|