Should you buy your teenager their own car?

Between June 2003 and December 2004, researchers in New South Wales, Australia, enrolled newly licensed drivers aged 17 to 24, all holding a first-year provisional licence, into what became known as the DRIVE study; 20,806 consenting drivers were included in this analysis. At recruitment, each driver answered a detailed questionnaire: how much they drove, how they rated their own ability, how often they took risks behind the wheel, how much they drank, whether they'd had professional lessons, and dozens of other factors. One question matters most for this guide: "Do you have regular access to a car, van, truck or similar vehicle to drive?" 45.7% answered "yes, my own car". Most of the rest shared a parent's or family car.

Then comes the part that separates this from a survey. The researchers did not ring participants years later and ask whether they'd crashed. Instead, they linked each driver, by name and date of birth, to police-recorded crash data, hospital admission records and death registrations, right up to 2016. That means the outcomes are what actually happened, recorded by systems that don't care what anyone remembers or admits. Memory can flatter. A hospital admission record cannot.

One more reason to trust the comparison: the two groups were not wildly different people. Both were mostly 17 to 19 years old, and the demographic spread was broadly similar. Where the groups did differ, in driving hours, risk-taking, drinking and other habits, the researchers measured those differences and adjusted for them. That adjustment is where this study gets interesting, and it is the entire subject of the cross-examination below.

Start with the raw counts, before any statistical adjustment. Over the follow-up, 21.9% of drivers with their own car had at least one crash, against 18.7% of those sharing a family car. For crashes resulting in hospitalisation or death, the proportions were small but the gap was proportionally much larger: 1.3% of own-car drivers against 0.7% of family-car drivers. Nearly twice the proportion, before adjusting for anything.

Raw counts can mislead, because the groups differed in how much they drove and how they behaved. So the researchers adjusted for age, sex, where people lived, how many hours a week they drove, risk-taking, sensation seeking, drinking, drug use, professional lessons, test attempts and more, and expressed the result as a hazard ratio. A hazard ratio of 1.29 means that, at that point in time, own-car drivers were crashing at a rate about 29% higher than comparable family-car drivers. Here is what the adjusted analysis found:

The time course matters as much as the size. The gap was widest at the start, when the drivers were newest, and narrowed as they gained experience. For any crash, the rate was about 33% higher at six months, 29% higher at one year, and it faded slowly, remaining statistically detectable until the end of year seven. For serious crashes, the rate was roughly double at the end of year one and still about 48% higher at year three. After that the serious-crash estimates were no longer statistically distinguishable, partly because, mercifully, serious crashes are rare enough that the data runs thin.

Adjusted mean hazard ratios from Table 3 of Chen et al. (2024). HR above 1 means a higher crash rate for the own-car group. Where the confidence interval includes 1, the difference is no longer statistically distinguishable from zero.

Think about what happens, mechanically, when a young driver has to borrow the family car. They have to ask. The moment they ask, several things occur that nobody planned as a safety system but that function as one. The trip has a stated purpose: you know, at least roughly, where they're going and why. The timing gets negotiated: "have it back by eleven" quietly removes the highest-risk late-night hours from the table. You see them before they drive: tiredness, mood, and whether they've been drinking are all briefly on display in the kitchen. And because the car is shared, the default answer to "will you drive us all?" is harder to give: research the authors cite suggests sharing a family vehicle may reduce the chance of carrying carloads of peers, one of the best-documented multipliers of young-driver risk.

Hand over a car of their own and every one of those checkpoints disappears at once. Not because anyone decided to remove supervision, but because the structure that produced the supervision is gone. The studies the authors draw on (Cammisa et al., 1999; Williams et al., 2006) found exactly this: parents of teens with exclusive access to a vehicle find it harder to set driving limits, and monitoring drops during precisely the months when driving habits are being formed. There is a second, blunter mechanism in the same literature: cars owned by teenagers tend to be smaller and older than family cars, which means less crash protection and fewer of the safety systems that prevent or soften collisions.

This reframing matters because it changes what the finding asks of you. Read naively, the study says "don't buy the car". Read properly, it says the shared car was doing a job you didn't know it was doing, and if you buy the car, that job doesn't do itself any more. Someone has to do it deliberately. The whole of the decision section below is about how. It also connects to a pattern that runs through everything we know about crash causation: the danger in driving is rarely the machinery and nearly always the human context around it, a theme we examine at length in why drivers really crash.

The first thing to translate is the licence system itself. The drivers in this study were on NSW provisional licences, which at the time came with real teeth: an effectively zero alcohol limit (cut to zero outright mid-study, in 2004) and a 90 km/h speed cap, among other rules. An Irish novice driver on N plates keeps the learner's 20mg drink-driving limit for the first two years of their full licence, rather than the 50mg limit that applies to established drivers, but faces no novice speed cap, no passenger restriction and no night-time curfew. That comparison should give an Irish parent pause, and it points in an uncomfortable direction: the doubled serious-crash rate in year one was measured in a system with more legal protection wrapped around new drivers than ours has. There is no strong reason to expect the own-car effect to be smaller here, and at least one structural reason to worry it could be larger, though we want to be clear that is inference, not measurement.

The second Irish translation concerns the learner phase. In Ireland, every first-time learner completes the 12 EDT lessons with an approved instructor and must drive accompanied by a qualified accompanying driver, someone who has held a full licence for at least two years, until they pass the test. That accompanying-driver structure is, in effect, the negotiation mechanism written into law: someone qualified is in the seat, trips are by arrangement, and driving is supervised by default. We've written about how to use both well, in our guides to EDT lessons and the accompanying-driver rules. The point this study adds is about what happens the week the test is passed: in Ireland, all of that structure vanishes on a single day, and if a car of their own is waiting in the driveway, the drop in oversight could hardly be more abrupt: from a qualified adult in the passenger seat to nothing, overnight. The transition, not the test, is the dangerous moment.

The third translation is about who carries the decision. In NSW as in Ireland, no law tells a family whether to buy a seventeen-year-old a car. This is one of the few major safety levers that sits entirely with parents, which is exactly why we built a dedicated hub for parents of learner drivers, including the family-practice ground rules that this study, in hindsight, vindicates. The gap between what the law requires of a fresh full-licence holder and what is actually safe for them is wide, and family rules are the only thing that fills it. Our guide to teen and novice drivers covers why that first unsupervised year behaves the way it does.

First, the reason the authors refuse the simple "just say no" reading, because it does them credit. They cite evidence (Ralph, 2018; Porykali et al., 2021) that reliable car access is linked to better education and employment outcomes, particularly for disadvantaged young people, and that outside cities a car can be the safer transport option, not the riskier one. In their own cohort, own-car rates were higher outside the metropolitan area and in less well-off districts: for many families the car is not a luxury, it is how a young person gets to a job or a course. So the question this study poses is not "car or no car". It is "if a car, then how", and on that the authors make four recommendations. The table below sets out their four recommendations, with our Irish reading of each. If you're already budgeting for lessons and tests, a subject we've costed in what failing the driving test costs, the trade-offs will look familiar.

The option nobody advertises: buy later, or share first

There is a fifth option implicit in the study's time curves that deserves to be said plainly. The excess risk is front-loaded: highest in the first months, with serious-crash rates twice the family-car level at year one, fading by year three. That means delay is a genuine safety lever. A family that shares a car for the first six or twelve months after the test, and buys the teenager their own car once the steepest part of the risk curve is behind them, has used the study's finding about as well as it can be used. Nothing about that requires conflict; it requires saying, out loud, that the first year is different, because it measurably is.

What survives all of that is the part we've built this guide on: a large, prospective, record-linked cohort in which own-car drivers crashed more, most sharply in the first year, with the pattern intact after adjusting for the obvious alternative explanations, and consistent with independent findings from other countries. That is not proof in the mathematical sense. It is about as good as evidence on this question is ever likely to get.