US law enforcement has increasingly employed asset forfeiture against a variety of criminal enterprises in recent years, but the effectiveness of forfeiture in deterring crime has received little systematic study. Since 1989, Portland, Oregon, has used asset forfeiture to deprive "problem" drunk drivers of the instrumentality of their offenses: their vehicles. Event-history analysis techniques were applied to arrest data covering five years of forfeiture enforcement. Perpetrators whose vehicles were seized could be expected to be rearrested on average half as often as those whose vehicles were not. The most plausible explanations for this result point to a reduced threat to public safety from these problem motorists as a result of Portland's forfeiture program.
PORTLAND'S ASSET FORFEITURE PROGRAM: THE EFFECTIVENESS OF VEHICLE SEIZURE IN REDUCING REARREST AMONG "PROBLEM" DRUNK DRIVERS
After languishing in relative disuse since prohibition, the US "wars" on drugs and organized crime promulgated new statutes and an explosion of interest which revived first criminal and ultimately civil forfeiture as common prosecutorial tools. Across the nation in the late 1980s, many state and local jurisdictions passed measures authorizing novel uses of forfeiture against crime. In 1989 one such measure, Portland, Oregon's, Forfeiture Ordinance, began targeting problem drunk drivers. This study evaluates the effectiveness of that program in reducing rearrest among this highly recidivistic group.
While a large body of literature exists regarding the legal issues surrounding forfeiture, little material exists respecting forfeiture's effectiveness in deterring crime. This dearth of research is bewildering in light of the frequency with which the effectiveness of forfeiture is cited in justification of its employment. In a 1991 review of literature on forfeiture by the Reed College Public Policy Workshop (PPW), Kapsch et. al noted: "Considering the appeals that the courts so often make to the effectiveness of forfeiture as an apology for occasional abuses, it is astounding that so little empirical evidence of that effectiveness has been produced" (p. ii).
Since the 1991 review, forfeiture has continued to be a frequent topic of articles in academic and legal publications, as well as the subject of court decisions and public debate. Unfortunately, this attention has done little to provide any systematic evidence of forfeiture's widely touted effectiveness against any of the many types of crime against which it is now frequently used.
GAUGING FORFEITURE'S IMPACT
The Federal "War on Drugs"
According to the U.S. Justice Department Executive Office for Asset Forfeiture [EOAF] (1994), "[t]he mission of the Department's Asset Forfeiture Program is to maximize the effectiveness of forfeiture as a deterrent to crime" (p. v). While, in the opinion of the EOAF, "revenue is an ancillary benefit" (p. 15), and not the primary goal of the forfeiture program, the amount of revenue derived from seizures and deposited in the Asset Forfeiture Fund "serves as a barometer to measure the success of the program" (p. 16). This amount has grown from $27 million deposited in FY 1985 to more than one half billion dollars in FY 1993, and totals over $3.2 billion since the Fund's inception in 1985 (EOAF). Excluding special deposits related to the Drexel Burnham Lambert case in 1989 and the Michael Milken case in 1991, regular deposits have increased in each year of the Fund's existence (EOAF).
If the fund truly is a barometer of the Asset Forfeiture Program's objective of deterring crime, one might expect to see an impact on the U.S. drug supply which roughly mirrors the growth in annual asset seizures. Yet in the case of cocaine, the flagship target of the national "war on drugs," prices have remained consistently low and purity has remained consistently high in recent years. The number of individuals reporting using cocaine at least once a week has remained relatively constant over the same period (Nat'l Narcotics Intelligence Consumers Comm., U.S. Drug Enforcement Admin. [NNICC], 1994). While the number of people reporting infrequent use of the drug has dropped dramatically since the mid-1980s, it is not clear whether this drop is related in any way to the Asset Forfeiture Program, or if it is the result of increased drug education, cultural trends or a combination of factors (NNICC, 1994). Absent a better measure of the impact of the Asset Forfeiture Program than the mere value of assets seized, it remains an open question whether, as Attorney General Janet Reno asserts, "[a]sset forfeiture has proven to be an effective tool in stripping criminals of the instrumentalities and proceeds of their illicit activities" (EOAF, 1994, Foreward), or whether criminals have merely absorbed the costs imposed by the Program as an inevitable cost of doing business in the multi-billion dollar international drug trade.
State and Local Efforts
At the state and local level, a number of law enforcement jurisdictions have implemented enforcement programs which have included the use of forfeiture and other forms of administrative property seizure against a variety of criminal activities. Studies evaluating these programs, some of them quite sophisticated, nevertheless fail in a variety of ways to conclusively assess the effectiveness of forfeiture in any of the capacities in which it has been employed. Some efforts studied have targeted the "supply side" of criminal activities.
Other efforts have attempted to control or hold accountable individuals who use drugs, or whose possession and use of legal but controlled items, such as weapons, poses a threat to society:
Some programs have used forfeiture in combating both supply and demand of illegal drugs:
While all of these studies provide interesting information on how forfeiture is being employed around the country to address a variety of law enforcement needs, none provides any conclusive evidence of forfeiture's effectiveness as a deterrent of crime.
Forfeiture and Policy Making: Need for Study
If any conclusive studies of forfeiture's effectiveness do indeed exist, certainly none have reached the attention of those who would have the greatest stake in citing their outcomes: the policy makers, public officials and academics who regularly square off in the forfeiture debate. Several papers delivered to a 1994 New York Law School Law Review symposium debating forfeiture assert that forfeiture is an effective crime deterrent (Symposium, 1994). Yet none cites statistics which adequately substantiate this claim. At a 1993 congressional hearing in which civil forfeiture came under intense criticism sparked by well-publicized tales of abuse, a U.S. representative, a state representative, a high ranking Department of Justice official, and a county sheriff all characterized forfeiture as a "powerful weapon" against crime (U.S. Congress, 1993). Yet none cited studies to substantiate this characterization, nor do any documents entered into the record of the hearing contain references to any such studies. A 1992 report by the Bureau of Justice Statistics [BJS] on drug crime characterizes forfeiture in an almost identical manner, again without citation of evidence.
In academic and legal journals, in government reports, and ultimately before the political bodies where policy is shaped, forfeiture continues to be portrayed as a potent weapon against crime without the benefit of any systematic knowledge of its effectiveness. This does not seem to be the result of disingenuousness, but rather of a pervasive conflation of the power of forfeiture to seize assets, which neither proponents nor critics doubt, with the power of forfeiture to deter crime, which is untested. The two are not synonymous. Without proof that the benefits of forfeiture are tangible and significant, it is impossible to state whether the costs of its occasional abuse are justified. Rational public policy making requires well-defined, quantifiable assessments of what forfeiture has and has not achieved. Such assessments are sadly lacking from current policy debate.
PORTLAND'S FORFEITURE PROGRAM
The most well known, debated and publicized aspect of forfeiture in the U.S. in the last
decade has been the cooperative efforts of federal, state and local law enforcement authorities to
wage the war on drugs against the various parts of the organizations which supply narcotics,
from the giant international cartels to the dealers on the street. However, asset forfeiture
programs aimed at "[ensuring] user accountability" (BJS, 1992, p. 186) have been employed in
various jurisdictions at least since 1986 (Purdum, 1986; Johnson, 1986). Typically, these
efforts have targeted the demand-side of the drug equation, seizing the property - usually
vehicles - of users who attempt to purchase drugs. Portland has taken this approach to new
areas by using forfeiture to target other crimes in the commission of which a motor vehicle is
instrumental. Under Portland's Forfeiture Ordinance, in effect since December of 1989,
vehicles may be seized and forfeited from offenders arrested for driving while their licenses are
suspended or revoked (DWS) if the suspension resulted from driving under the influence of
intoxicants (DUII), and from offenders who are arrested as habitual traffic offenders (HO) -
people who have committed three or more serious traffic offenses, at least one of which must be
a DUII to meet the criteria for forfeiture.
Questions and Concerns
Portland's program raises a number of questions and issues. Drinking and driving is a devastatingly serious problem, a problem which is made more troublesome by the fact that many perpetrators are hard-core recidivists whose behavior seems to be all but impervious to modification by means of conventional sanctions. The Forfeiture Ordinance targets these individuals specifically, since one must be a repeat offender to be subject to its provisions. Does seizing these people's vehicles succeed where other measures often fail, or, as some suspect, do they simply replace the seized vehicles with unregistered "junkers" and continue to drive?
In addition to the impact of the Ordinance on offenders, its impact on taxpayers and law- abiding citizens must be considered. Contrary to popular (and often cynical) beliefs about the financial benefits of asset forfeiture to law enforcement, the Portland forfeiture program costs more to administer than it takes in from sales of seized property. Most vehicles seized are never auctioned, but are instead released to third parties, such as spouses and lenders. Of those that are forfeited and auctioned, most tend to be older vehicles of relatively little value. Another concern with the widened use of forfeiture by law enforcement is its perceived potential for abuse. Although the Portland Ordinance contains important safeguards and is administered by men and women of high integrity, the entrusting of such a powerful tool to the hands of law enforcement should be accompanied by clear benefits to public safety. Only if the program is effective in protecting lives on the highways by depriving drunks of their weapon of choice will the real cost in tax dollars and potential cost in liberty seem worth paying.
The 1992 Survey of Offenders
In the Spring of 1992, the PPW conducted a survey to examine the effectiveness of the Portland program in deterring alcohol-related driving activity. The study was designed as a phone survey of a target group consisting of households of offenders, as well as of a control sample of households selected at random from the Portland metropolitan area. To minimize the number of refusals, hang-ups or untruthful responses which might result from asking to speak with the offenders by name, it was decided to request to speak with the individual in each household with the birthday nearest to the survey date. The survey was conducted in cooperation with the Portland Police Bureau (PPB) using the facilities of the PPW and funded through a grant from the Rose E. Tucker Charitable Trust.
Analysis of the data from the survey unfortunately revealed problems with the target group data. Of the 194 households surveyed in the target group, only 78 reported that any member had been stopped for DUII. Of those, only 12 reported having had a vehicle seized or forfeited. This was especially puzzling given the care with which the survey instrument had been adapted from instruments which had already been tested and found to be relatively reliable. It must be concluded either that the perpetrators were no longer or never had been at the phone numbers provided from the PPB computer files, or that the respondents did not answer accurately or truthfully on a wide scale. While there are no doubt important methodological lessons to be learned from the 1992 survey results, they cannot be used to answer the question of whether Portland's forfeiture program has been an effective crime deterrent.
The Current Study
The current study seeks to answer this question using offender data acquired internally from PPB, rather than from a survey. For the purposes of this investigation, the broad notion of deterrence is addressed operationally along the lines of the familiar dichotomy between general deterrence and specific deterrence. General deterrence is the reduction in criminal activity caused by the threat of a sanction in those potentially subject to its imposition. Specific deterrence is the reduction in criminal activity caused by the imposition of a sanction in those to whom it has actually been applied. Despite exploration of a variety of techniques to circumvent the inherent shortcomings of arrest data, the lack of crucial information regarding individual knowledge and perceptions of forfeiture as a sanction prevented a methodologically sound assessment of the general deterrent effect of the forfeiture program. This study therefore focuses on the impact of forfeiture as a specific deterrent in reducing rearrest rates among those whose vehicles have been subjected to it. The body of the report is organized in three sections. "Data" describes the sources from which the data for the study were collected and the organization of the data file used in the analysis. "Methods" gives explains the statistical methodology. "Results" reports and discusses the interpretation of the outcome of multivariate analysis which tests the effect of the forfeiture sanction on rearrest rates among a sample of offenders.
The data for this study were acquired from PPB's Portland Police Data System (PPDS), from the PPB Asset Forfeiture Unit's vehicle seizure records, and from the monthly reports of the PPB Traffic Division. The PPDS data consists of all citations issued from January 1, 1989, to December 31, 1994, for DUII, felony DWS, and HO (N = 22,525). Data prior to 1989 were unavailable due to regular purging of old citation records by the Data Processing Division. Multiple citations may be issued for a single custody, and many perpetrators have multiple citations. Each record of a citation contains variables for unique PPB perpetrator and custody identification numbers, allowing grouping and relational linking of records by perpetrator or custody. There are 21,220 unique custodies and 16,801 unique perpetrators represented in the PPDS data set.
The vehicle seizure data consist of records for all seizures of vehicles for felony DWS or HO subsequent to the institution of the forfeiture ordinance in mid-December, 1989 (N = 746). Traffic Division data consist of a record of hours patrolled by Traffic Division officers by shift (morning or evening) and the total number of DUII citations they issued for each month from January, 1986, to December, 1993. There are gaps of missing values in these data due to transitions in record-keeping staff. The data sets for all analyses were created via manipulation of these three sources.
Unobserved Sources of Heterogeneity
Any individual charged with HO, or with felony DWS during a license suspension for DUII, is potentially subject to vehicle seizure and subsequent forfeiture. In answering the question of whether having a vehicle seized specifically deters, we wish to examine whether rearrest rates differ between individuals arrested for HO or felony DWS based on whether or not their vehicles were seized at the time of initial arrest. Ideally, there should not be any unobserved sources of heterogeneity - unmeasured differences between groups - which make people in one group more or less likely to be arrested than those in another. For example, if seizure were only applied to offenders with particularly egregious driving histories, and data about those driving histories were unavailable for inclusion as controls in analysis, we would be unable to sort out the effects of forfeiture on recidivism from the effects of having such a driving history. Fortunately, this is not the case. However, there is one difference which must be considered between the group of individuals whose vehicles were seized and the group whose vehicles were not.
It is known that all individuals whose vehicles were seized for felony DWS were operating under a suspension for an alcohol related offense, since such a suspension is a criterion for seizure. However, due to the way that offenses are coded in the PPDS data and the purge by PPB Data Processing of all data prior to 1989, it is impossible to know whether the license of an individual charged with felony DWS whose vehicle was not seized was suspended for an alcohol related offense or for some other reason. However, the non-alcohol related license suspensions during which a felony (as opposed to misdemeanor) DWS citation may be issued are generally related to severe and relatively rare offenses, such as suspensions for negligent vehicular homicide or hit-and-run (Oregon Motor Vehicle Code, 1993-94). Consequently, only a very small proportion of felony DWS citations are given to individuals whose licenses were suspended for non-alcohol related reasons. This fact, the fact that controls may be introduced for recent alcohol related driving convictions from the available data, and the large sample size all make it unlikely that the inevitable inclusion of non-alcohol related felony DWS custodies in the group whose vehicles were not seized introduces significant bias.
It should also be noted that even if any bias were introduced by the inclusion of such custodies, such a bias would be conservative with respect to the effect of vehicle seizure on rearrest, if one assumes, plausibly, that offenders charged with felony DWS for driving during non-alcohol related suspensions are less likely to be subsequently commit alcohol-related offenses. All individuals charged with felony DWS whose vehicles were seized are known to have been operating during an alcohol related suspension. Some individuals charged with felony DWS whose vehicles were not seized presumably were operating under non-alcohol related suspensions. If the non-seizure group as a whole were somewhat less likely to offend, then any reduction of the risk of rearrest attributable to having one's vehicle seized would be underestimated, since the group of individuals whose vehicles had been seized would be in general more likely to offend. Since the null hypothesis we wish to reject is that seizure has no effect in reducing recidivism, if seizure exhibits such an effect in analysis, we can be certain that this effect is not due to an unobserved source of heterogeneity related to the inclusion of non- alcohol related felony DWS custodies, and that if the estimation of this effect is at all in error, then such an error is on the side of conservatism.
Structure of the Data Set
With this in mind, the data set was chosen to consist of all custodies between January 1,
1990, and December 31, 1994, for which a citation for felony DWS or HO was issued (N =
5,493). Only custodies for 1990 and later were used to allow the creation of a variable for
number of prior offenses in the previous year. Since no data exist prior to 1989, including cases
prior to 1990 in the analysis would have introduced bias, as the prior arrest variable for such
cases would not reflect a full year of data, as it would for all subsequent cases. For each case, a
variable was created for the date on which the next subsequent felony DWS, HO or DUII arrest
was observed for the individual involved in the custody. Many individuals were not rearrested
within the observation period. A "dummy variable," that is, a dichotomous variable having the
value of either one or zero, was created to indicate whether the rearrest variable contained the
date of a subsequent arrest, or whether there was no rearrest observation in the study period.
Cases for which there was no rearrest are considered to be "censored" by the end of the study
period. Censoring of data is discussed in the methods section, below. Another dummy variable
was flagged to indicate cases where there had been a vehicle seizure at the time of arrest (N =
Enforcement Level Covariate Vector
It is likely that the probability of being arrested at any given time depends in part on the level of police enforcement in effect at that time. Traffic enforcement is carried out both by the officers of the Traffic Division and by regular patrol officers on the street. There are, unfortunately, no available data on Bureau-wide traffic enforcement activity. Missing data can often be extrapolated from available data if a model with reasonable assumptions can be fitted which reliably predicts missing values as a function of other complete data. The Traffic Division in the past has issued monthly reports containing information on its patrol activities. Complete data does exist for the total number of DUII citations issued per month Bureau-wide through December, 1994, as well as for the number of DUII citations per month issued by the Traffic Division through August, 1993. If a model were found which could reliably predict Traffic Division hours patrolled as a function of Traffic Division DUII citations issued, then this model could be used to predict Bureau-wide patrol hours on traffic enforcement from Bureau-wide DUII citations issued, assuming that regular officers, when engaged in traffic enforcement, are approximately as efficient at issuing citations as Traffic Division officers.
Unfortunately, the best model capable of being constructed with the available data was only able to account for approximately 39% of the variance in Traffic Division hours patrolled as a function of Traffic Division citations issued. Introduction of controls to account for seasonal variation in offense rates did not significantly improve the model. In other words, approximately 60% of the variation in DUII citations issued by the Traffic Division is accounted for by factors other than hours patrolled and seasonal variance. As sufficient data is not available to reliably predict missing values for Traffic Division hours patrolled, there is no way to predict Bureau-wide traffic enforcement, even if the assumption of equal enforcement efficiency were justified.
While we cannot extrapolate the total Bureau-wide traffic enforcement, the number of patrol hours by the Traffic Division in the evening (when most citations are issued) does significantly predict over 37% of the variance in Bureau-wide DUII citations issued. Traffic Division evening patrol hours may therefore be a significant predictor of a portion of the variance in the likelihood that an individual will be arrested for DWS, DUII or HO at any given time. We may test this hypothesis by analyzing the subset of cases for which complete Traffic Division evening patrol data are available. The data on Traffic Division enforcement were used to create for each case a vector of 44 variables containing values for hours patrolled in each of the up to 44 months subsequent to the date of arrest for which data exist. Although this is less than ideal, the subset of complete cases from January, 1990, through August, 1993, is sufficiently large to allow testing of whether Traffic Division hours patrolled had a significant effect on rearrest rates.
Regression and Time-to-Event Data
The most common regression methods are often inappropriate for analysis of the effects of independent variables on a dependent variable containing time to an event. In most techniques, values for the dependent variable must be a number or dichotomous categorical. Although these methods can be used with time-to-event data, for example, if the dependent variable is coded to reflect whether or not, or how often, an event has occurred in an arbitrarily specified follow-up period, such an approach is wasteful of information for a number of reasons. First, and most obviously in the present case, all custodies whose follow-up period extends beyond the end of the study period would have to be eliminated from analysis, since we could not specify a value for the dependent variable for them. If the follow-up period were, for example, one year, no custodies after December 31, 1993 could be used as cases in the analysis, since the period for which data exist ends December 31, 1994, and these custodies would not have a full year of subsequent observations for the determination of the dependent variable. Second, even for cases where the initial offense occurred before December 31, 1993, information about reoffenses which may occur subsequent to the follow-up period would be lost to analysis. Lengthening the follow-up period only reduces the number of usable cases by lengthening the period prior to the end of the study in which cases could not be used, while ameliorating the loss of cases by shortening the follow-up period exacerbates the loss of potentially interesting reoffense data beyond the follow-up period.
A third problem with customary regression techniques when applied to time-to-event data is apparent when we consider that in the case of criminal recidivism, the amount of time from initial offense to reoffense is highly interesting. This information is available in our data set, but is wasted when only whether or how often an individual is rearrested within a given period is considered. It might be thought that this deficiency could be corrected in a linear regression model by using time to reoffense as the dependent variable. However, for individuals who are not rearrested by the end of the study period, the value of the dependent variable is unknown, or censored by the arbitrary imposition of the time cut-off at the end of the study period. Assigning the end date of the study period to the dependent variable would introduce bias by underestimating the actual time to reoffense in most cases, while assigning any other date would be completely arbitrary and result in an under or overestimation for an unknowably large part of the sample. The only other alternative would be to treat censored cases as missing, and thus exclude them from analysis, introducing yet a different bias and losing valuable cases. A further problem with common regression methods for time-to-event data is the fact that certain independent variables, such as an individual's age, are not constant, but vary through time. Ordinary regression techniques offer no way to estimate the effects of time-dependent variables. A different approach is plainly needed.
EVENT HISTORY ANALYSIS
The various techniques of event history analysis are superior to other regression techniques for time-to-event data in that they allow censored observations adequately to be taken in to account, and they permit the use of time-dependent variables. A number of concepts are common to all methods of event history analysis. A case for which an event, such as reoffense, could occur at some time is said to be "at risk" at that time. The total number of cases at risk in any given time period is known as the "risk set." The probability that an event will occur in a particular time period for a particular case in the risk set is termed the "hazard rate." Certain event history models incorporate regression techniques to allow the estimation the effects of covariates on hazard rates. Of these, the Cox proportional hazards log-linear regression model (Cox, 1972) is especially powerful and non-restrictive, given that certain assumptions are adequately fulfilled.
The Cox Proportional Hazards Model
Two of the advantages which Cox models have over many other methods of event history are worthy of note. First, as we have noted, certain covariates, such as the age of a research subject, may change in value during the time that the subject is at risk, and Cox models can use time-dependent variables in regression analysis. Second, many other continuous-time methods use "parametric" models. Such models require the researcher to specify prior to analysis the over-all form of the hazard rate as a function of time. Often, there is very little information available on which to base such a specification. As "non-parametric" models, Cox models require no specific assumptions about the form of the underlying hazard function, and are thus much more general and flexible than parametric models. It is primarily because the Cox model combines the use of time dependent variables with a non-parametric model that it has become the method of choice for event history analysis when it is appropriate.
The Proportionality Assumption
Cox models are not, however, always appropriate for all data. For a Cox model to be
appropriate, it must be assumed that the effects of differing values for the independent variables
are proportional over time. For example, if the covariate "sex" is included in the model, the
Cox model is appropriate only if the hazard function for males differs from that for females only
by a constant factor at all times. A simple statistical method of checking proportionality with
respect to a variable is available by means of testing the significance of the effect of the
interaction of that variable with the log of the time on study minus the log of the mean time to
event for the entire sample. If the effect of this interaction variable is not significant at a chosen
level of significance (as it is not for the variables used in this analysis at pŁ0.05), then the data
may be assumed to be roughly proportional and the Cox model may be used (Blossfeld,
Hamerle, and Mayer, 1989).
Stepwise Regression and Model Building
Building the best model for predicting observed values of a dependent variable involves testing candidate independent variables for inclusion and removal from the model such that the final model contains only those independent variables which contribute significantly to the overall goodness of fit of the model, and excludes those which do not. With any more than a few explanatory variables, manually building a model can be very time consuming. A stepwise regression is an automated procedure for performing this potentially tedious task. In our analysis, variables considered likely to contribute to the model based on theoretical considerations and exploratory results were included in the model on the first step, and those considered unlikely to make a significant contribution were excluded. In subsequent steps, variables in the model were tested for removal and variables not in the model were tested for inclusion. Variables were removed if their removal did not significantly degrade the predictive accuracy of the model, and were included if their inclusion significantly improved the model (p to includeŁ0.1, p to removeł0.15). Significance levels were calculated using the maximum partial likelihood ratio method. Stepwise regression proceeds iteratively until no variables meet the significance criteria for inclusion or removal. The variables still remaining at this point constitute the final model.
Constant explanatory variables tested for inclusion and removal were the sex and race of the subject, the number of prior felony DWS, HO or DUII offenses in the preceding year, whether the subject's vehicle had been seized at the time of custody, and whether the vehicle was subsequently auctioned. The time-dependent variable of the age of the perpetrator was tested using the entire sample, as was the monthly number of evening hours patrolled by the Traffic Division in a model using only cases through August of 1993.
Effects of Variables on Rearrest Rate
Table 1 shows the effects of explanatory variables on time to rearrest in terms of regression coefficients with associated significance levels from the Cox proportional hazards regression model. Only variables having a significant effect on time to rearrest are included in Table 1. Evening hours patrolled by the Traffic Division did not have a significant effect on rearrest in the subset of cases through August, 1993. The model therefore was estimated using all available cases from January 1, 1990, through December 31, 1994.
Effects of Explanatory Variables on Time to Rearrest
Predicted # Rearr./Mo. Predicted Time to Rearr. Variable Coeff. % Increase (Decrease) % Increase (Decrease) --------------------------------------------------------------------------------------- Sex (Male) 0.4467* 56.32 (36.03) Age -0.0192* (1.90) 1.94 Race: Black 0.6900* 99.38 (49.84) Asian -1.8141* (83.70) 513.50 Other 0.3934** 48.19 (32.52) Prior Offenses 0.2543* 28.96 (22.46) Vehicle Seized -0.6887* (49.78) 99.12 --------------------------------------------------------------------------------------- * pŁ0.01. ** pŁ0.05. --------------------------------------------------------------------------------------- Model Chi-Square=724.02, DF=7, pŁ0.01.
Regression coefficients indicate the magnitude and the direction of the effect of each explanatory variable on the hazard rate. A positive coefficient indicates a greater number of expected rearrests in a one month period of time based on an increase of one unit in the value of an explanatory variable, and a shorter expected time to rearrest based on the same increase. A negative coefficient indicates the opposite effect. By calculating the exponent of the coefficient, we arrive at the percent increase or decrease in the hazard rate predicted by a positive change of one for an explanatory variable. Thus being male, as opposed to female (the arbitrarily chosen reference category), corresponds to a 56.32% increase in the number of expected rearrests per month. 100% minus the inverse of this percentage gives the percent expected increase or decrease in time to rearrest - for males, a 36.03% decrease in expected time to reoffense as opposed to females.
No entry for "Race: White" is included in Table 1, as Whites are the reference category for the categorical variable "race" (though any other category could have been chosen). All estimates for the effect of race contrast the effect of being in a certain racial category as opposed to being White. We can thus see that expected time to rearrest is slightly less than half as long for Blacks than for Whites, and over five times longer for Asians than for Whites. Time to rearrest did not differ significantly for Hispanics or American Indians from that for Whites, and these categories are therefore not shown in Table 1. Considered together, other races than those considered specifically had a predicted time to rearrest about a third shorter than that for Whites. Each additional year of age increased the expected time to rearrest by about 2%. We can also see that each prior arrest predicts a 32.52% decrease in expected time to rearrest. Most interestingly, having a vehicle seized nearly doubled expected time to rearrest. Having a vehicle actually forfeited did not have a significant effect over and above that associated with simply having it seized. All of these results are highly statistically significant. Vehicle seizure is a strong and significant predictor of reduced rearrest for DWS, HO and DUII with several other important factors taken into account.
Interpretation of statistical results is not a deductive process, but rather involves choosing among explanations which are consistent with an outcome based on their plausibility. Before concluding that seizure has resulted in reduced recidivism, we must consider consistent alternatives. A classic example of a sanction reducing rearrest rates within a certain geographical area without affecting recidivism is the case of prostitution. There is good reason to believe that when stronger anti-prostitution enforcement is applied in a certain area, arrests in that area may fall, but often only because prostitutes and "johns" relocate to a different area where they may conduct their business with less interference. A similar phenomenon is common with respect to drug activity and enforcement. As state-wide data on offenders were not available for analysis, it may be questioned whether individuals whose automobiles were seized merely continued to reoffend in jurisdictions other than Portland, just as prostitutes or drug-dealers may ply their trades in less well-patrolled sections of town when enforcement is strengthened in their customary area of operations. Could individuals whose vehicles have been seized simply have continued to reoffend at the same rate, but in another jurisdiction as subsequent to vehicle seizure?
There is a fundamental difference between driving on the one hand, and prostitution and drug-dealing on the other, which suggests that the answer to this question is negative. Stepped- up enforcement in one area only requires that a prostitute or drug-dealer travel to a different area to conduct his or her business. No relocation of domicile is required. But an individual whose license has been suspended cannot simply continue to drive in another jurisdiction without relocating his or her place of residence. To completely avoid the prospect of seizure while continuing to drive, an offender must physically relocate his or her residence to another jurisdiction. Such an individual might theoretically reduce his or her chances of apprehension by striving to the greatest degree possible to drive in other jurisdictions when conducting business, minimizing time spent driving within Portland. Yet such a strategy would still involve the risk of regular driving within the city limits, and require a great deal of additional time in performing even the most routine errands. It is highly unlikely that such relocation, either or domicile or driving, is responsible for the dramatic increase in expected time to rearrest predicted by vehicle seizure. More plausible than relocation is the possibility that offenders are continuing to drive after seizure or forfeiture, but that they are driving more carefully to avoid detection. While it is highly likely that this occurs, it seems doubtful that it accounts for the magnitude of the effect on rearrest rates. Presumably, the offenders did not try to get caught the first time. It should also be noted that even if the only effect of the forfeiture program were to run offenders out of town, to cause them to drive as much a possible in other jurisdictions or just to drive much more carefully, this result in itself would be highly desirable from the standpoint of Portland motorists.
If seizure does result in reduced recidivism, how does it do so? Could seizure of vehicles be physically preventing people from driving? While actual forfeiture did not predict any reduction in rearrest over and above that predicted by seizure alone, this does not mean that physical prevention of driving through the loss of a vehicle is not an important factor in reducing rearrest rates. Vehicles which are not forfeited are released to lien holders, spouses and other innocent owners on the understanding that their use will be withheld from offenders. Yet any offender who is able and who wishes to may purchase a beat-up used car for very little money, neglect to register and insure it, and continue driving. If offenders are not driving subsequent to seizure, it is likely not because, strictly speaking, they are physically prevented from doing so, but rather that they choose not to take the necessary steps and resume driving, that is, they are deterred.
Why would seizure deter where other sanctions have failed? While offenders may view brief jail terms with indifference and simply fail to pay fines, the loss of use of a vehicle through seizure or forfeiture is a tangible penalty. Many offenders have few financial resources. The investment which is lost in a vehicle which is forfeited may be considerable to them, even if the vehicle was of relatively little value. The cost of replacing a vehicle can serve as an unavoidable fine, even if a vehicle is only seized and released, if an offender also loses access to it. With vehicles which are released, theca consequences incurred at the hands of third parties also may enhance the deterrent effect of seizure. New York prosecutor Sterling Johnson, speaking of young suburbanites who travel to the city to buy crack and whose cars are seized, put it well: "When they come home without momma's car or without daddy's car, the criminal justice system is going to be the least of their worries..." (Purdum, 1986).
Proper consideration of the outcome of this study requires that the sharp distinction between the facts revealed and their theoretical explanation be reiterated. One may perhaps dispute the explanation, but inasmuch as our data are accurate and our methods sound, it is a fact that, other things being equal, having a vehicle seized reliably predicts a doubled expected time to rearrest for individuals arrested for DWS in the city of Portland between Jan 1, 1990 and December 31, 1994. Explanation of the facts is based on inference and is open to interpretation. Reduced driving as a result of physical incapacitation or deterrence, or driving more carefully are plausible explanations and are consistent with the observed reduction in rearrest rates. Most probably, a combination of these factors is responsible for this result. What is important is that following any of these plausible strategies for avoiding rearrest also serves to make an offender less of a danger on Portland's roads. Any positive modification of the behavior of a group of offenders as recalcitrant as the subjects of this study is a significant accomplishment. Unlike the many forfeiture initiatives which have only been anecdotally justified, if Portland's forfeiture program achieves nothing else, it is still a verifiable success story.