Method

For this study it was necessary to create a single database that contained information relevant to all the GDL system restrictions (night-time curfew, young passengers, alcohol) but was not biased with respect to the factors being investigated. The best source of crash data was the police traffic crash reports, but it was of concern that this database may be biased towards the factors being studied. For example, it is known that there is under-reporting of crashes to the police,^12–14 so if there were changes in the reporting rate of crashes that involved one or more of the driving restrictions, the results may be due to changes in reporting rates, rather than the actual restrictions.

For the present study this potential problem was overcome by matching police reports to the New Zealand Health Information Service (NZHIS) hospital inpatient file.¹⁵ The NZHIS maintains a record of all persons admitted to a public hospital for treatment of an injury including crash victims, and all injury discharges are E coded.¹⁶ The sole criterion for a record to be included on the hospital file is admission to hospital, which can be considered a proxy measure of injury severity. This should not be influenced by police reporting practices, or biased by whether a crash occurred at night, involved young passengers, or involved alcohol.

Following the procedure established by Alsop and Langley,¹⁴ individual records in the New Zealand police traffic crash report file for 1980–95 were matched to the records for the same individuals and crash events in the NZHISs' hospital inpatient file. On the hospital file traffic crash victims were identified by E codes 810–819 and motor vehicle occupants by using the fourth digit of the E code (0 for driver, 1 for passenger).¹⁷ The linked database, therefore, contained only records where a vehicle occupant had been admitted to hospital. Records were excluded where the licence status of the driver was undetermined or missing, or was recorded as unlicensed or disqualified. A summary of the licence status for all drivers with a linked record is given in table 1. This shows that of the grand total of 6693 cases, 4505 (67%) were included in the analyses. Of these, 2252 (50%) were licensed pre-GDL, 1273 (28%) had a full graduated licence, 980 (22%) a restricted licence; 3435 (76%) were male, and 1069 (24%) female.

The main unit of analysis was a motor vehicle involved in a crash where the driver was 15–19 years, inclusive. The traffic crash reports were used to identify the night-time (10 pm to 5 am) crashes and the presence of passengers. The involvement of passengers less than 20 years of age was limited to the crashes involving an injured passenger (that is, young casualty). The involvement of alcohol was assessed by the police officer who recorded “whether the driver was thought to have been drinking”. There is some evidence that this is a reasonably accurate indicator if alcohol is present.¹⁸

For the statistical analyses multivariate logistic regression were used to compare the pre-GDL group with the restricted licence group, and the pre-GDL group with the full GDL group. Separate regression models were fitted where each of the four driving restrictions were the response variables.

1. Night-time curfew crashes.

2. Crashes involving passengers of all ages.

3. Crashes involving young casualties.

4. Crashes where alcohol was suspected.

To control for potential confounding effects, factors known to be strongly associated with young driver crashes were included as covariates in the models. These covariates were gender and age. To control for linear trends that may have arisen as the result of other interventions over the period covered by this study, the year in which the crash occurred was included the model.

Results

A summary of the number and percentage of crashes that involved each of the driving restrictions are given, by driver licence status, in table 2. For the night-time, all passenger, and alcohol suspected crashes, all of the linked crashes were used, but the young casualty comparisons were based only on crashes where there was an injured passenger, hence the reduced totals.

Table 3 provides an example of the full logistic regression model that was used for all of the eight regression models. In this example (table 3) the response (or dependent) variable was a night-time crash (yes/no) and the main explanatory (or independent) variable was driver licence status, and in this example the pre-GDL licence group, which was classified as the reference group, was compared with the restricted licence group. The results show that, compared with the pre-GDL drivers, a significantly smaller proportion of the crashes involving a restricted licence driver occurred at night (odds ratio 0.66, 95% confidence interval 0.50 to 0.86, p=0.003).

Four similar, separate logistic regression models were fitted using the same explanatory variables as were used in the example given in table 3. For these four models the main explanatory variable was the type of licence held by the driver (pre-GDL drivers were compared with the restricted licence drivers) and the four (yes/no) response variables were:

1. Involvement in a crash during the night-time curfew.

2. Involvement in a crash with passengers (all ages).

3. Involvement in a crash with a young casualty.

4. Involvement in a crash where alcohol was suspected.

The results for the main explanatory variable, for each of the four response variables, are given in the upper section of table 4. This shows that, compared with drivers licensed pre-GDL, restricted licence drivers had a significantly smaller proportion of crashes that occurred during the night-time curfew (p=0.003), a significantly smaller proportion of crashes that involved passengers of all ages (p=0.018), and also a significantly smaller proportion of crashes where alcohol was suspected (p=0.034), but for crashes involving an injured passenger, there was no significant difference (p=0.980).

The next stage of the analysis compared the crashes that involved a pre-GDL driver with those that involved a full graduated licensed driver. To do this four further models were fitted using the same four driving restriction response variables as were used in the previous models. The explanatory variables were also the same as used previously, except that for licence type those with a full graduated licence were compared with the pre-GDL group. The results for the main explanatory variable, pre-GDL versus full graduated licence, are given in the lower section of table 4 and show that those with a full graduated licence had a smaller proportion of crashes at night (p=0.042) but there were no significant differences for any of the other restrictions.

Discussion

Earlier research has shown that, following the introduction of GDL in 1987, there had been a significant reduction in the number of young people seriously injured in traffic crashes,⁵ in the ratio of crashes involving a young driver compared to older drivers,⁴ and also in the rate of crashes per number of licensed drivers.^{6, 9} The results from the present study show that, of the crashes that occurred between 1980 and 1995, those that involved a driver with a restricted licence were less likely to have occurred at night, less likely to have involved passengers, and less likely to have been suspected of involving alcohol, compared with crashes involving a driver licensed under the old system. This suggests that the GDL restrictions, and in particular the night-time driving curfew, have made a positive contribution to the reduction in young driver crashes that has occurred since the GDL system was introduced in 1987.

A full evaluation of the restriction on the carrying of passengers was not possible because data on passenger age was only available for those who had been injured. The results showed that among this group the restricted licence driver was just as likely to have young passengers as the pre-GDL driver. Although this study was not directly concerned with addressing issues of compliance with the driving restrictions, results such as this draw attention to the findings of previous research that have highlighted some of the problems of compliance,^{19, 20} and especially breaches of the restriction on the carrying of young passengers.

Supporting the findings from other researchers,^21–23 the night-time curfew appears to have contributed to a reduction in night-time crashes. Because this was associated with a reduction in the actual number of crashes, we are confident that this result was not due to an increase in daytime crashes, but was either due to less or safer driving at night by young, restricted licensed drivers. It was also interesting to find that the drivers with a full graduated licence also had a lower proportion of crashes at night than the pre-GDL drivers, even though the night-time driving restriction no longer applied to them.

There are several possible explanations. It may have been that, compared to the pre-GDL drivers, those drivers with a full graduated license were safer drivers at night, they may have driven less at night, they may have been less likely to drive for no real purpose, or they may have been less likely to drink and drive. Irrespective of the reason, it is encouraging that the night-time curfew has helped reduce young driver crashes.

The results of this study also showed that the young restricted licence drivers compared with pre-GDL drivers had a reduced proportion of crashes where alcohol was suspected. There were, however, a number of factors other than the GDL system, that may have contributed to this result. In 1992 the 30 mg per 100 ml of blood alcohol limit was extended to include all young drivers under 20 years, irrespective of the type of licence they held. This would have had direct impact on the last few years of the present study, and hence the results presented here. In addition, a range of other strategies such as compulsory breath testing, booze buses, drink-drive blitzes, and media campaigns, have targeted drinking and driving in the whole driving population. These strategies should have had some effect on the behaviour of young drivers, making it is impossible to disentangle the effect of the GDL system alcohol restriction.

In May 1999, some changes were made to the New Zealand GDL system. Among these was a penalty regime comprising monetary fines and demerit points for violations. Breaches of a condition of either the learner or restricted licence incurs 25 demerit points and a $400 (NZD) fine. If 100 demerit points are accumulated within a two year period, the licence is suspended for three months. Other changes to the GDL system in 1999 made it apply to all new drivers (was 15–24 year olds only), the time discount (that is, time reduction) associated with a learner licence was discontinued, the age limit for the supervisor removed, and the learner licence driver must display “L” plates. At the restricted licence stage, the time can be reduced from 18 to 12 months (six to three months for drivers over 25 years) by taking an approved driver education course. Also, to obtain a full licence the new driver must pass a practical test, which is designed to confirm basic driving skills and the ability to detect and respond to hazards in built-up areas and in higher speed zones.²⁴ The effect that these changes have on the effectiveness of the GDL system remains for the future.

Implications for prevention

The New Zealand experience suggests that GDL has been an effective way of helping to reduce traffic crashes among young drivers. The results from this study indicate that driving restrictions, such as a night-time curfew, have directly contributed to this reduction, and should be an essential component of any GDL system. There was also some evidence that the passenger and alcohol restrictions may have contributed to the crash reduction, suggesting that these restrictions should also be given serious consideration by any jurisdiction developing a GDL system.