Week 1: Welcome to the Road. Please Remain a Student, Not a Statistic.

The unit's opening lecture — what risk, exposure, outcome measures, and uncertainty mean, worked through a single fictional collision, from a professor whose own qualifications now include "an unusually comprehensive incident report."

22 February 2027

Open the slides

“Welcome to Vehicular Macropod Avoidance. I had originally planned to teach this course theoretically.” — Dr Kargaroo, opening remarks

Dr Kargaroo’s qualifications include a doctorate in ecology and, in his own words, “an unusually comprehensive incident report.” He opens the semester by putting that report to its first test: does one professor’s collision prove that a particular stretch of highway is unusually dangerous? The answer is no, on its own — and working out exactly what “on its own” is missing is what this lecture teaches.

Why kangaroos, why now#

One professor’s collision doesn’t prove a road is dangerous. It does raise the question the whole semester exists to answer:

How can we reduce kangaroo–vehicle collisions, and how would we know whether an intervention actually works?

Every later week — the ecology, the dataset, the physics, the engineering, the policy — builds toward answering that question properly. This lecture sets the standard of evidence the rest of the semester is held to.

One collision, several perspectives#

Take the Yarralong Road collision, invented for teaching and not a real incident: at approximately 6:40pm on a weekday in late autumn, a sedan travelling a regional two-lane road with an 80 km/h limit struck an adult kangaroo that entered from the eastern verge. The driver braked and swerved but did not avoid the animal, stopped, and called it in; a wildlife carer attended and later filed her own notes. No other vehicles were involved. The nearest wildlife-crossing sign sits about 400 metres further along the road, past the collision site.

Five people can look at that same collision and reasonably disagree about what it shows:

• the kangaroo — its movement, habitat, and survival aren’t the point of anyone else’s account, but they’re the reason it was on the verge at all;
• the driver — what she could detect, when, and how she reacted;
• the road authority — whether the current signage and road design are defensible;
• the wildlife carer — the animal’s welfare and an accurate clinical record;
• the researcher — whether any of the above supports a general conclusion, or only a description of this one evening.

You’ll get the full case file — the incident briefing, the driver’s statement, the carer’s notes, and five role cards — in the session immediately after this lecture. Here, it’s the worked example for four ideas you’ll use all semester.

What would we need to know?#

• Risk is not “did a collision happen” — it’s how likely one is, combined with how bad it is if it does. Two separate mechanisms can raise it: dusk lighting can shorten how far ahead a driver can actually see, which cuts the distance available to detect an animal at all, and a higher travel speed can shorten how much time is left to respond once something is detected. The case gives us the road’s 80 km/h posted limit, not the sedan’s actual speed that evening, and the driver’s own unverified estimate of what she was doing. Neither establishes that risk was actually elevated at 6:40pm on this stretch — they tell us which direction each factor would push it, if we knew the true speed.
• Exposure is how much opportunity there is for the event at all — traffic volume multiplied by how much animal activity overlaps with it. Knowing one kangaroo crossed the eastern verge once tells you nothing about how often that verge gets crossed, or by how many animals, without separately measuring exposure.
• An outcome measure is what actually got recorded, and by whom, for what purpose. This one incident already produced two records that don’t agree in scope: the driver’s account, given because a vehicle was involved, and the wildlife carer’s notes, filed because an animal needed attention. Neither is “the number of collisions on this road.” Knowing which outcome measure produced a number is a precondition for using that number at all.
• Uncertainty is what the case doesn’t establish, and inventing an answer for it would be interpretation dressed up as fact. The driver’s speed estimate has no independent record to check it against — no fixed camera, no data logger, just her recollection. The carer’s notes establish the animal’s species and sex, and record injuries consistent with vehicle impact — the animal was deceased on arrival — but not how long it had been near the road, or whether this stretch has a prior collision history at all. The session asks you to keep those categories separate rather than resolve them by picking whichever answer makes the best story.

This case is a single, deliberately detailed example — not a substitute for a larger dataset, and not itself part of one. A rich case and a dataset answer different kinds of questions, which is rather the point of putting them back to back across this unit.

A convincing story is not yet convincing evidence#

A single case, however well documented, can raise a good question; it can’t on its own establish how often something happens or why. A road with more recorded incidents might be more dangerous, or it might just be better monitored. “This road is dangerous” needs to say: to whom, under what conditions, and measured against what. Being the person in the case does not automatically qualify you to interpret it — a lesson Dr Kargaroo, by his own account, learned the hard way.

From blame to a research question#

A research question is something you investigate — it doesn’t turn out to be “true” or “false,” only more or less well-supported by what you find. “Which conditions predict roadside presence?” is a research question: you could spend a career on it. What can be shown wrong is a hypothesis: a specific, falsifiable claim you derive from the question and test against evidence. For example, “roadside sightings are more frequent within 200 metres of a permanent water source than farther away” is hypothesis-shaped — not a finding, just the shape a testable claim takes — because a count of sightings against distance could disprove it.

This distinction fixes three familiar verdicts. None of the example hypotheses below report a real finding — each is only a template for what a falsifiable, measurable claim looks like: a predictor, an outcome, a comparison, and where it helps, a threshold or a time period.

• “kangaroos just jump around” → question: which conditions predict roadside presence? → example hypothesis: recorded sightings within 200 metres of a permanent water source are higher, over the same three-month period, than sightings at an otherwise similar site more than 1 kilometre away.
• “more warning signs would help” → question: does signage change driver or animal behaviour? → example hypothesis: strike counts per kilometre on a signed stretch, over a 12-month period, are no lower than on a matched unsigned stretch, once traffic volume is held constant.
• “the new device obviously works” → question: what does the device actually detect, and under what conditions was it tested? → example hypothesis: the device’s detection rate for animals crossing at dusk is no higher than its detection rate for the same crossing pattern tested at midday, across an equal number of trials.

The skill this unit is after is producing that kind of testable hypothesis, not arriving with the verdict already chosen.

Further reading#

Rowden, Steinhardt & Sheehan (2008), “Road crashes involving animals in Australia”, Accident Analysis & Prevention 40(6) — read this before the session. It’s the real, peer-reviewed source behind the outcome-measure problem above: matching their own in-depth study of serious-injury crashes against official government crash records, the authors found the official figures substantially undercount animal-related crashes, because many incidents that show up in the in-depth data never reach the record type that gets officially counted.

The semester ahead#

The course moves from understanding the problem, to reading and weighing evidence, to analysing candidate interventions, to proposing something testable. The two assessed pieces sit on that same arc: the week 6 literature review and hotspot analysis (40%) establishes what the existing evidence can actually support, and the week 12 research proposal (60%) puts forward an intervention along with a way of judging whether it worked. You are not required to solve road safety in twelve weeks — you are required to explain how you would test your idea.