At what stages of licensing do graduated driver licensing systems reduce crashes? Example from Queensland, Australia

•Changes to the Graduated Licensing System in Queensland, Australia, were evaluated.•Crash trends of pre-post cohorts were compared to identify when declines occurred.•Substantial crash declines were evident in the first month of provisional licensure.•Learner night crashes and provisional driver passenger crashes also declined.•This approach offers additional policy insights to other-age cohort comparisons. Evaluations of new graduated licensing systems (GLS) commonly examine pre-post young driver crash rates relative to another driver group. This comparison approach is important to account for other influences on crashes over time, but has limited ability to determine which GLS components are most effective and at what stage during the licensing process. We previously identified declines in young driver crashes in Queensland, Australia, following introduction of a new GLS in 2007. The objective of the current research was to conduct complementary modelling to identify at what points through the licensing process had particular GLS policies contributed to reductions. Crash trends were explored for learner and provisional drivers under the new GLS versus previous system for three time periods relative to the month of acquiring a provisional licence: the preceding learner period, the first month of provisional licensure (when crashes typically peak), and the overall provisional period. Interrupted time series analyses were conducted for the log ratio of crashes per 10,000 licensed (learner and provisional) drivers with the total number of licensed drivers as an offset. The greatest declines were found in the first month of licensure, with indications that a longer learner period, higher supervised driving hours, and a new provisional night-passenger restriction were key contributors to provisional crash reductions. There was also some indication that a restriction on all phone use reduced crashes during the learner period. We conclude that time series analysis focusing on licensing stage, rather than calendar time only, offers a complementary approach to analysing GLS effectiveness by better identifying where and how changes impact crashes.