|Home | Contents|
EFFECT OF OHIO VA LAW
This section describes a comparison of DUI recidivism and moving violation rates and crashes of DUI before and after the ALS laws implementation in September 1993.
Under the new laws implemented September 1, 1993, drivers arrested for DUI not only lost their driving privileges immediately under the ALS law, but some offenders also had their vehicles towed and impounded or immobilized at the time of arrest. Those subject to the VA law were multiple DUI offenders and, if suspended for DUI, first and multiple DWS offenders (see Figure II-1). The vehicle was held for at least 5 days pending a hearing. Following the hearing, the vehicle might be released pending trial, but in most cases continued to be impounded or was immobilized for the period provided by law. Since all of the drivers eligible to receive this sanction were suspended for at least 1 year, the vehicle action occurred during the time the offenders license was suspended.
A significant impediment to studying the effect of the Ohio VA law was that, while the notice of the offense was recorded, the vehicle impoundment sanctions were not recorded on the State drivers file. Therefore, whether the offender was sentenced to a vehicle action sanction had to be obtained from local court records. Then, the record had to be traced to the local police department to determine if the sanction was carried out and, if so, the dates of immobilization. Because these local data were required, a statewide study was beyond the resources of this research project; therefore, the evaluation was limited to two large urban counties.
Studies were conducted in two Ohio counties, Franklin (Columbus) and Hamilton (Cincinnati), on the effect of that law. This section evaluates the results of those two studies. Each study was independentHamilton County served as a replication of the study in Franklin County. As described in the Appendix A, the two counties implemented the VA law using different methods. Hamilton County used only vehicle impoundment, and Franklin County initially impounded and then transferred the vehicle to immobilization on the offenders property. The effect on the offenders whose vehicles were impounded or both impounded and immobilized was essentially the same: loss of the use of the vehicle for 30 to 180 days.
In court systems, the application of law varies to some extent among prosecutors and judges. In Franklin and Hamilton Counties, some offenders who were eligible for a vehicle penalty under the VA law were not sanctioned (see Appendix A). Further, personnel limitations and backlogs in the data system sometimes resulted in failure to impound eligible vehicles at the time of arrest. Finally, after a U.S. District Court decision in April 1995, many vehicles registered to someone other than the offender did receive the VA sanction.
Because not all eligible offenders received the vehicle sanction, it was possible to compare eligible offenders whose vehicles were impounded or immobilized with similarly eligible offenders who escaped this sanction. Since the offenders were not assigned to these two groups at random, it was necessary to use demographic and driving record data to equate the groups through covariate analysis.
As noted in Section I, through the cooperation of the Ohio BMV, the driving records of State residents who were cited with DUI offenses between January 1, 1990, and September 1, 1995, were extracted from the States driver record file for analysis. Studies of the VA law made use of the pool of records of the 16,494 drivers who committed a total of 18,185 DUIs in the 2 years following the implementation of the ALS and VA laws. From this data set, drivers records in Franklin and Hamilton Counties were matched with court records obtained for Franklin and Hamilton County residents who were charged with a second or third DUI or with a first or second DWS offense between September 1, 1993, and September 1, 1995. Where the court record indicated that an offenders disposition included a vehicle sanction, the police department record of the impoundment or immobilization was obtained.
DUI offenders who were subject to vehicle forfeiture were not included in the study because there were insufficient cases to provide the statistical power needed for a valid analysis. Financial Responsibility Assurance (FRA) and wrongful entrustment offenders were also subject to having their vehicles impounded. Here again, however, there were too few cases to study. Thus, the State drivers record for each subject in Franklin or Hamilton Counties was merged with his or her court record. Where a VA law order was issued, the case was merged with the local police records to verify that the vehicle driven by the offender had actually been impounded and/or immobilized. (In Franklin County, the records of the court-based immobilization coordinator were used to verify immobilization.)
Data for this study were collected over 2 yearsfrom September 1, 1993 (the date the law went into effect) to September 1, 1995. The length of time that postoffense driving records could be followed varied with the date of offense. Those who offended in the fall of 1993 were tracked for 2 years, up to 23 months or more after the sanction period. Those who offended late in the study did not complete their sanction periods by the September 1, 1995, cutoff date. Driver records were not drawn from the BMV until November 1995 so that court actions occurring through August of that year could be posted to the drivers records.
The proportion of eligible offenders in each county who received a vehicle sanction is shown in Tables IV-1 and IV-2. In Franklin County, only one in four of the DWS offenders who were eligible to receive a penalty actually had their vehicles impounded and/or immobilized. The proportion of DUI offenders receiving the penalty was somewhat higherbetween 30% and 40%. Compared to Franklin County, Hamilton County DWS offenders were almost three times more likely to receive a vehicle sanction, while a somewhat larger portion of the DUI offenders had their vehicles impounded.
In each county, a large majority of the vehicles seized under the VA law were held for the full period provided by the law. One exception to this generalization was first DWS offenders in Hamilton County. Nearly a third of the first DWS offenders served less than half the 30-day impoundment period. Despite this, the average length of the exposure period (shown in the next to last line of Table IV-2) for these first offenders was 29.6 days, almost equal to the minimum 30-day sentence, because many offenders delayed retrieving their vehicles beyond the 30 days.
The average length of driving exposure when the vehicle was impounded or immobilized and the average period of exposure after the offender retrieved his or her vehicle is shown in the lower two lines of Tables IV-1 and IV-2. These are the periods during which repeat DUI and DWS offenses were analyzed for each offender group using survival analysis. The power of these analyses depended upon both the number of offenders in each group studied and the average length of time in which a subsequent offense could occur. Thus, for example, the shorter impoundment periods (30 and 60 days) for DWS offenders made it less likely that reductions in recidivism would be detected than for those with longer impoundment periods (90 to 180 days) imposed DUI offenses.
The average period of exposure while the vehicle was impounded or immobilized was about equal to the sentence length (30, 60, 90, or 180 days) in both counties. Thus, the statistical power to detect a change in the offense rate during the sanction period was a function of both the length of the sanction period and the number of offenders receiving the penalty. On the other hand, the average period of exposure following the return of the vehicle was relatively even for all groups in both counties (about 1 year). Therefore, the most important determinant of power for detecting change during the after-sanction period was the number of cases in the group.
Survival analysis is the method of choice for determining differences in recidivism (Lee, 1992). It uses all the subject days available for analysis in the study database, thereby generally providing the greatest statistical power to detect change. It also allows comparison of rates as they change over the entire length of exposure, as opposed to one discrete, fixed period of exposure. In the recidivism example, one can test the survival (or hazard) rates of groups given different sanctions or against the rates of a baseline comparison group.
Two survival analysis procedures were used in the present study: Cox Regression and the Kaplan-Meier (1958). Both analytic techniques allowed for separate baseline survival/hazard functions for each of the four sanction groups (first DWS, second DWS, second DUI, third DUI) or strata, and the effect of the action (impoundment and/or immobilization versus neither) can be testedeither pooled across the four separate groups or separately within groups.
The Kaplan-Meier method has one limitation. Unlike Cox Regression, it does not permit the use of covariates beyond a single stratification to adjust for prior differences between groups or to explain individual variation within groups in survival times or rates. Because random assignment was not possible, Cox Regression was used, in addition to the Kaplan-Meier procedure, to permit the application of covariates for age, gender, and prior record. In the current study, these two analytical methods yielded similar results.
Kaplan-Meier is an analysis of the difference between or among survival distributions across time. Unlike Cox Regression, it does not assume that the distribution for one group (or level of a factor) is a constant proportion of the survival distribution of another group throughout the period analyzed. The Kaplan-Meier test only determines if they are different. The two curves being compared may actually cross, with one having lower survival rates at first but then higher survival rates later, yet they still test out as statistically different. As such, it is analogous to a nonparametric test for categorical differences in that a significant result does not necessarily indicate that one is consistently or even generally higher than the other.
For this reason, it is difficult to produce an estimate of effect size or a summary measure of the difference in survival over time. A Kaplan-Meier result that shows the two survival curves as different does not provide a measure of how much higher (or lower) one group is than another despite a significant result. To provide comparable measures that are frequently reported in traffic safety literature, the cumulative recidivism rate at a fixed point in time into the survival distribution, rather than across-time summaries (such as a ratio of areas under the recidivism curves), are reported in the present study.
In choosing the time point at which to slice into the survival functions to compute the rates, it is important to ensure that a relatively high percentage of the cases of that group have not yet been censored since the standard error (and volatility) of the functions increases toward the end of the functions as fewer cases remain. For the period of vehicle impoundment or immobilization, most cases had available exposure at least as long as the prescribed length of the sentence for that group. Therefore, at the end of these survival functions (30, 60, 90, and 180 days), the cases with exposure are still very high (80-90%). Within a few days of that point, nearly all sanctioned offenders came off the sanction, so all subjects were censored at 30, 60, 90, or 180 days. Therefore, the rates for recidivism during the sanction period are based upon the cumulative survival functions at these termination points of the sanction periods.
For the period after the sanction, the deterioration of group sample size began early, and the attrition rate increased over time. Approximately half of the subjects in each group ran out of exposure by 12 months. By 18 months, only 5% to 20% of the subjects remained, making the curves increasingly susceptible to large changes caused by single events. For this reason, the rates reported in Tables IV-3 and IV-4 are based on the point in time 12 months after the end of the vehicle sanction.
There are three test statistics available for Kaplan-Meier analysis, which differ primarily in how they weight the time points of the survival functions. The Log-Rank statistic weights all time points equally, and the Breslow and Tarone-Ware statistics weight each time point relative to the proportion of cases in the group that are uncensored at that time point. This method is preferable when the number of sample size changes so much as to make later time points less reliable, as happens in the postsanction period in this study. The Tarone-Ware statistic, which we used in our analysis, weights each time point by the square root of the number of uncensored cases remaining at the beginning of that time point.
Cox Regression uses models that assume the groups (or factor levels) being compared are based on a single underlying baseline survival function and vary among themselves in proportions of this baseline function, which proportions are constant over time. In other words, the relationship between two groups (vehicle sanction/no vehicle sanction) recidivism rates are assumed to be the same over time. Because of this assumption, Cox Regression is able to produce a single summary measure that estimates how much higher (or lower) the survival rates are over time than those of another group, in a form similar to an odds ratio. Cox Regression also allows for the inclusion of other factors and covariates in the model, to explain more of the between-subject variation in survival time and to render groups equivalent via covariate adjustment, if necessary. The significance of each parameter coefficient in the Cox Regression model is tested via the Wald statistic. The effect sizes reported in the tables are derived from the antilogs of the coefficient (eB) for the vehicle sanctioned group, which estimate the proportional relationship over time to the comparison groups function.
Both Cox Regression and Kaplan-Meier analyses are important in this study. The former is necessary because some comparison groups appear to be different from the vehicle sanction groups historically in terms of prior DUI and DWS offenses. Because prior offenses are the best predictors of recidivism, it is important to control for this factor so that differences between these groups that are attributable to differences in group composition are partialed out and not wrongly attributed to the sanction effect. The Kaplan-Meier approach is necessary as well because the assumption of proportionality may not be appropriate for some groups, or at least across all time periods.
In both counties, four sentence lengths were analyzed: 30 and 60 days applicable to DWS offenders and 90 and 180 days applicable to DUI offenders. In Franklin County, the sanction period consisted of the time when the vehicle was impounded combined with the period when the vehicle was immobilized. This was analyzed separately from the period following the sanction when no vehicle action was in effect. In Hamilton County, the sanction period consisted of the time the vehicle was impounded (immobilization was not used by this county), followed by the period after the vehicle was returned. Two dependent measures were evaluated: DWS and DUI convictions. The relatively few fourth-time DUI offenders and third-time DWS offenders who were subject to vehicle forfeiture were not entered into the analysis.
In each county, the comparison group consisted of DUI or DWS offenders who were eligible for a vehicle sanction but did not receive the penalty. In comparing their recidivism rates with those who did receive a sanction, the period for each comparison group was set to be equal to the average time of the sanction for the equivalent experimental (impounded/immobilized) group. For the analysis of the period following release of the vehicles, the origin of the survival curves for both groups was set to correspond to the dates when the experimental group offenders retrieved their vehicles to avoid any carryover from the sanction period.