P.R Marques, A.S. Tippetts, R.B. Voas

Pacific Institute for Research and Evaluation Calverton, MD USA 20705

Abstract

The alcohol interlock record is the first extensive behavioral record that has ever been available to help researchers understand DWI offenders. The average interlock device in Alberta Canada logs over 2300 breath tests during an average 9 months of interlock use. Analysis of the patterns of violations makes available new predictive information for scaling driver risk. Recent work (Marques et al., 2001, Marques et al., 2002) has shown the overall rate of elevated BAC tests during the first several months of use strongly predicts repeat DWI after the interlock has been removed. This report further analyzes the record of 5.5 million BAC tests in Alberta and finds that even after accounting for strong predictors of post-interlock recidivism such as prior repeat DWI and the overall prediction based on higher rates of failed interlock BAC tests, if the BAC tests failures occur in the morning (e.g., BAC still >=.04 gm/dl presumably from previous night drinking) there is an additional 45% higher likelihood of later having a new repeat DWI. Also, evidence is presented that offenders who decrease their overall interlock vehicle use (fewer tests taken over interlock use time) have a higher likelihood of recidivism post-interlock.

Introduction

Two particularly interesting findings from analysis of the interlock breath test record reported in Marques et al., (1999) warranted further investigation and the results of those analyses are now reported here. Both questions involved the topography of BAC tests logged by interlock users that were warn (.02-.039 gm/dl) and fail-levels (>.04 gm/dl) during the period they were required to use an interlock. These questions relate to 1) examining time of day when elevated BAC tests are logged, and 2) the pattern elevated BAC tests during the entire period of interlock use. The purpose is to relate distribution of elevated interlock BAC tests to post-interlock recidivism.

[Back to top]  

Daily patterns: are failed morning BAC tests evidence of dependence?
The question is whether the time of day when BAC tests are logged at warning or fail violation levels is uniquely predictive of recidivism. In the 1999 paper we reported that the morning workday hours (7-9am), and weekend hours (10-12pm) had the highest frequency of elevated BAC tests. We've conjectured that the occurrence high BAC at the first startup in the morning might be evidence of greater alcohol problem severity since these BAC tests may reflect alcohol from a previous evening of drinking. Higher alcohol problem severity, as indicated by high consumption and elevated morning BAC, suggests more frequent episodes of impaired driving and possibly more convictions during the post-interlock years. This was considered to be of interest for several reasons but partly because during our intervention study (Marques et al., 2000), counselors reported that offenders, upon learning about the hours required for alcohol clearance from circulation at higher BACs, began to believe their own BAC might really still be elevated in the morning. The initial reaction to failed BAC tests in the early morning was to assume interlock equipment failure, but a discussion of the timeframe for detoxification of ethanol sometimes helped convince the offenders that failure to start the interlock car may have been due to residual BAC.

Overall pattern: is there evidence for learning?
The 1999 paper reported that, over the period of time that the interlock was used, data aggregated across all interlock users showed that the proportion of all tests taken resulting in BAC>=.02 gm/dl was greatest at the beginning of interlock use and then the violation rate declined over time, giving the appearance of learning curve which was best modeled by a quadratic function. During the period of interlock use, the proportion of all tests that were warns or fails dropped by about 50% (in the aggregate) from beginning to end. Since our finding that the rate of BAC violations logged by the interlock is a strong predictor of later DWI convictions we wanted to know if those who later received repeat offenses deviated from this pattern - whether they failed to learn. Did those who lowered the rate of BAC test violations during interlock usage later have a a lower likelihood of receiving a new repeat offenses after the interlock was removed. Is failure to reduce BAC violations evidence of less learning as indicated by future DWI?

Methods

The detailed methods germane to this study have been reported previously in the three interlock recorder papers referenced above. Procedural detail is not repeated here.

[Back to top]  

Table 1: BAC tests by day of week. Data from 2310 interlock offenders.

Briefly, however, between the years 1994-1998, 2310 convicted DWI offenders in Alberta who had between 0 and 9 prior DWI offenses used an alcohol interlock device for periods of at least 5 months ranging up over 24 months. Half of the subjects were first time offenders. The average period of use was 9 months. After the interlock was removed, 182 additional repeat offenses occurred as of late Yr 2000 representing a variable length post-interlock observation period, representing 9% of the sample. While the interlock was still on the cars, offenders provided nearly 5.5 million breath tests. Only one half of one percent of these are in the warn and fail range (BAC>=.020 gm/dl), but with the large sampling base this represents 22,828 test, an average of approximately 10 elevated BAC results for each of the 2310 interlock users; 7,286 elevated BACs were in the fail range (BAC>=.040 gm/dl). Table 1 shows that the pattern of elevated BACs is highest during the weekends (Saturday, Sunday) even while the actual number of tests taken during the weekend are fewer than those on weekdays (Monday - Friday).

Results

Does high morning BAC reflect greater problem severity?
The data in Figures 1a and 1b show failed BAC tests stacked into two ranges (.040-.079 and .080+ gm/dl) and broken out by hour of the day that the fail test result occurred. Figure 1a (left) is Monday to Friday, Figure 1b (right) is Saturday and Sunday. The pattern of failed tests differs distinctly between these two periods. On typical working days, Monday-Friday, the morning hours between 6:00-9:00 has the largest number of failed tests. However, this is not because that is the time of the largest number of attempted starts. The largest number of attempted starts is during the hours 16:00-17:00 (not shown). On weekends the largest number of failed BAC tests increases gradually with a peak occurring 10:00-12:00 and then declining through the day.

Failed BAC tests are an aggravation to the driver (since the car won't start), and overall these occur infrequently, so it is probably valid to assume failures were surprising to most drivers. Anyone surprised by an elevated BAC (first thing in the morning after awakening) may have a poor understanding of their level of problem drinking and this lack of awareness may reflect their drunken driving risk. In assess if there was something unique about the morning failures, these were studied in isolation of other intervals. For this purpose, five different periods were defined on the basis of the hour by day distributions. Morning was defined as encompassing the hours of 6-10 am on workdays and 10-12 on weekends. The remaining intervals evaluated are as shown in Table 2. The table also shows the percent of all interlock users who had at least one fail, or two or more BAC fails (.040+ gm/dl) during the intervals noted.

Previous evaluation of post-interlock recidivism with the Cox proportional hazard model (survival analysis) showed that failed interlock BAC tests and prior DWIs are the only two predictor variables that entered the equation to predict post-interlock DWI (Marques et al., 2002). These were the strongest from among a dozen or more variables tested which derived from both driver records (moving violations, driving while suspended charges) and from questionnaire data (self-report consumption, demographics etc.). New analyses were conducted using dummy-coded variables from Table 2 representing when during the day or week, 2 or more failed BAC tests occurred. In this case, as before, both the number of prior DWI and the proportion of all vehicle start attempts that are fails (BAC >=.04 gm/dl) entered the equation as significant predictors. In addition to those variables, receiving at least 2 fails in the morning hours was the only other variable that entered the predictive equation. If a driver had at least 2 fails during the morning hours, it increased the likelihood of a post-interlock repeat DWI by 45% (Exp B=1.45), significant at P=.05. No other evaluated period during the day came close to entering the equation; 2 or more nighttime fails (P=.3) was the closest.

[Back to top]  

Figure 1: Fail level BAC tests stacked into two intervals (.040-.079 and .080+ gm/dl)

Data are shown by hour on weekday Fig 1a, or weekend Fig 1b.

Table 2: Breakout of failed BAC tests into different time interval within a week

Do later recidivists fail to change behavior over time on the interlock?
The curve describing rate of change in BAC over the time that an interlock was in use (not shown here but found in Marques et al., 1999) portrays a 50% decline in the proportion of all start-up BAC test that are elevated above .02 gm/dl. Factors associated with prior DWI or breath tests were evaluated as nineteen different variables in attempt to identify predictors associated with those who ultimately recidivate post-interlock, and by so doing, show evidence of having failed to learn to control their drinking. To limit the sample to those for whom a reasonable demonstration of learning is possible, subjects with no BACs >=.04 gm/dl were excluded, as were those few who had an interlock for less than 5 months.

Among the tested variables were the best predictors found so far including proportional rate of failed BAC tests during the first five months of interlock use and more prior DWI offenses. In addition to these, several variables representing BAC tests over time were added to the model. These variables include the slope of warn, fail, or high fail tests over time within subjects, the slope of total tests taken (a proxy for vehicle use), the T-score of each of these (calculated as the slope divided by the standard error) reflecting 12 months interlock learning data, the transformed or normalized rates of warns, fails and high fails during the first two or the first five months. The predictors, all of which represent either the offenders' past driving record or their BAC performance while the interlock was installed, were analyzed with two different methods: survival analysis and logistic regression. The dependent measure in both analyses is postinterlock DWI offenses. The survival analysis takes into consideration the time frame in which the post-interlock repeat offense occurs and is therefore a continuous measure. The logistic regression only considers if the DWI occurred or did not occur. The data in Table 3 summarizes the results of the proportional hazard survival analysis and the logistic regression.

Table 3: Survival analysis (upper) and logistic regression analysis (lower) to predict postinterlock recidivism

[Back to top]  

Of the nineteen variables evaluated, both analyses' procedures identified four common predictor variables. The relative strength of each predictor can be approximated using the R value, a square root measure of effect size. The order of presentation in the tables roughly reflects the potency of the predictor. The predictors that were dummy-coded as dichotomous (the two DWI variables) show Exp(B) values, understandable directly, and are shown in the table. The likelihood of any repeat offense (relative to those who are first time offenders) increases the likelihood of a future DWI by about 57-58%. Having five or more DWI offenses reflects on only 155 of the offenders in the sample but this increases the likelihood of a repeat DWI postinterlock by an additional 131%.

In both analyses, the overall proportional rate of failed BAC tests with the interlock installed is, here again, a key predictor in both analyses. In Cox regression, the effect size and significance of this variable is similar to the high risk of 5+ repeat offenders. In the logistic regression, the failed BAC tests is substantially more potent a predictor than any of the other entries with R=.15.

Notable for its absence in the equation however, is any measure of slope. This suggests that change in the rate of warn, fail, or high fail BAC tests was not predictive of greater or lesser likelihood of post-interlock repeat DWI convictions. On the other hand, in both analyses change in the rate of vehicle use did enter the equations. It is not a high potency predictor but it is consistently present. Its sign is negative meaning that over time on the interlock, those who reduced the amount of vehicle use (i.e., drivers who showed a decline in the number of times they started the vehicle) had a higher post-interlock repeat DWI offense. This suggests there may be a subset of drivers who were finding it difficult to use their vehicles due to regularly elevated BAC so they decided to not use the interlock vehicle. Less vehicle use could be a form of self-imposed suspension (from the interlock vehicle). It could reflect less total driving, or it may only mean that driving shifted to a non-interlock vehicle. Shifting to a non-interlock vehicle is the weakest link in the interlock strategy since anyone stipulated to an interlock vehicle, especially people with more financial resources, can easily choose an alternate car.

[Back to top]  

Discussion

The alcohol interlock data recorder is the most significant source of objective driver BAC test data ever available for the understanding of DWI offenders. On average the interlock records about 265 tests each month. The occurrence of elevated BAC tests, particularly those above .04 gm/dl strongly predicts the likelihood of repeat DWI. Also, those who have failed BAC tests in the morning are 45% more likely to recidivate (even after accounting for prior offenses and higher total rates of failed tests).

Licensing authorities or courts that sanction DWI offenders, and which have interlock programs, could use this information right now to improve the identification of high risk drivers. If sanctions were flexibly applied the interlock would be left on the vehicle until the driver shows evidence of having successfully reduced driving (knowingly or unknowingly) with elevated BAC. Caution is necessary, however, and other risk indicators may be needed since it now appears that some offenders who had subsequent DWIs reduced vehicle use over time when required to drive with an interlock.

References

Marques, P.R., Voas, R.B., Tippetts, A.S., Beirness, D.J. (Dec 1999) Behavioral monitoring of DUI offenders with the alcohol interlock recorder. Addiction. 94(12), 1861-1870.

Marques, P.R., Tippetts, A.S., Voas, R.B., Danseco, E.R., & Beirness, D.R., (2000). Support services provided during interlock usage and post-interlock repeat DUI: Outcomes and processes. In H. Laurell & F. Schlyter (Eds.). Alcohol, Drugs and Traffic Safety - T 2000: Proceedings of the 15th International Conference on Alcohol, Drugs and Traffic Safety, May 22- 26, 2000. (Vol. 4). Stockholm, Sweden: ICADTS.

Marques, P.R. Tippetts, A.S. Voas, R.B, Beirness, D.J. (2001). Predicting repeat DUI offenses with the alcohol interlock recorder. Accident Analysis & Prevention. 33(5), 35-45.

Marques, P.R., Tippetts, A.S. & Voas, R.B. (2002). Comparative and joint prediction of DUI recidivism from alcohol ignition interlock and driver records. Submitted, March 2002 - Journal of Studies on Alcohol.

[Back to top]