• No results found

Using a quasi-experimental design, an impact evaluation was performed to assess the DISP impact on the 446 offenders assigned to it during the period from November 2000 through June 2003, primarily from Multnomah County (n=320), but also including a number of offenders who were residents of other Oregon counties (mostly neighboring counties in the Portland area). As indicated in Table 3, another 21 offenders were included in the analyses who resided in various Oregon counties outside the Portland area. (18 DISP offenders were excluded from this analysis because they were not Oregon residents at the time the Oregon DMV extracted the data in 2009 but were at the time of their conviction for DWI). According to data from their driver records, almost all (83%) of the DISP offenders were repeat DWI offenders, verifiable from Oregon DMV data, with 41% of them having three or more alcohol offenses.

Table 3. DISP Participants County of Residence Number of County of Residence Offenders Multnomah County n=320 Clackamas n=54 Washington n=34

Marion, Yamhill, Columbia, Hood River, & Clatsop n=17 Various counties outside Portland area n=21

Total Assigned n=446*

*Does not include 18 out-of-State offenders assigned to this DISP.

As with the other two ISPs evaluated, random assignment was impossible in Oregon, so it was important to determine post-hoc an appropriate comparison group. A potential study comparison group was derived from all Oregon drivers who had any DWI alcohol conviction and/or

administrative sanction (e.g., ALR/per se, implied consent, open container in vehicle) since 1998 (i.e., DISP subjects and potential comparisons). These records included alcohol offenses with arrest dates as early as 1971 (which is the case for the DISP subjects as well), so a retrospective period for determining prior offenses (i.e., before offenses occurring within the DISP period) was approximately 30 years up through June 2009 for both groups.

From this pool of 122,630 offenders (excluding those assigned to the DISP program), the pool of potential comparison subjects was further delineated as follows, such that they would be

identical to the DISP group on all these key factors:

 Offenders were current Oregon residents, as of July 2009

 Offenders who committed an alcohol offense within the timeframe corresponding with the DISP sentencing period of November 2000 through June 2003

 Offender’s age at time of comparably timed index offense was at least 18 years and less than 75 years (to match the age range of the DISP group)

 Sex coding for DMV was either male or female

 Offenders had fewer than eight prior alcohol offenses to match the range of the DISP group

After selecting for all of these characteristics, there remained 50,029 drivers for the potential comparison pool. For each potential comparison driver, his/her first DWI (or other alcohol- driving-related offense) occurring within the DISP sentencing period became that person’s baseline or index offense for comparison purposes. The number of prior DWI offenses (pre-DISP period) was calculated, and future recidivism incidents computed relative to the exposure

beginning at that index offense. This calculation of priors, and of future recidivism, relative to the period of index offenses 2000-2003, was therefore identical for both groups, DISP subjects and potential comparison subjects alike.

The majority (77%) of the potential comparison pool of offenders (50,029) were first offenders at the time of their “index” offense committed within the November 2000 to June 2003 period. The DISP program targets repeat offenders; less than 17% of the DISP subjects were ostensibly first offenders as recorded by the DMV data. Although it is likely that the statistical methods used for analysis would properly account for any such differences between groups via covariate adjustments, two matched comparison groups were composed via stratified random “quota” sampling, so that similar proportions of the comparisons would have priors distributed similarly to the DISP group within age and sex breakouts. To accomplish this, we performed random selection within each two-way stratum (age by priors) in proportion to the two-way age by prior distribution of the DISP group. Sex or gender was not included as a stratifying component because the overall distribution of sex in the potential comparison pool was already virtually identical to that of the DISP group. Sex was also equivalent across age and by priors, so

sampling for the other two strata automatically retained the proper proportions for sex. Two such random samples were created independently in order to have a resampling validation on

reliability of the comparison sampling procedure. The first comparison sample had 9,185 drivers, and the second comparison sample had 9,142 drivers. Both comparison samples were equivalent in three-way distribution of age by sex by priors to the DISP group’s composition. It should be noted that both of these comparison samples utilized County of residence as a stratification factor, so that the DISP subjects could be contrasted specifically against comparison subjects from within their same county – e.g., Multnomah DISP subjects with Multnomah comparison subjects; DISP subjects from counties outside the Portland area (“rest of State”) would likewise be contrasted with comparable subjects from those same counties.

The recidivism data for all these offenders were analyzed using survival analyses, namely Cox Regression models, as described for Minnesota and New York, adjusting for age and sex and prior cohorts as covariates. Additionally, each driver’s county of residence was included in the model, as there was evidence that recidivism rates differed by county, which could reflect a difference in enforcement exposure and/or socioeconomic differences of the drivers themselves. For all analyses of Oregon’s DISP, a second replication analysis was performed (i.e., using both comparison samples); results were virtually identical, and the differences were trivial. For each one, the most conservative results of the two are reported (i.e., the least significant) in order to have the most guarded estimates of outcome effectiveness.

All offenders’ exposure periods were censored at the end of June 2009 if they had not already recidivated before that point; additionally, drivers’ exposures were censored earlier at a deceased date if they died. This permitted a usable exposure period of up to 8 years and 8 months for both groups. Offenders in either of the Oregon groups who recidivated more than once had each of their recidivism offenses counted toward the group’s overall rate. All other statistical