Week 2: Macropod behaviour and circadian activity

Why kangaroo–vehicle collisions cluster at particular times of day — and why the graph that seems to show this is not yet the explanation for it.

1 March 2027

Open the slides

Dawn, Dusk, and Other Inconvenient Appointment Times.

Week 1 showed that one incident cannot establish a general pattern. This week goes a step further and slightly less comfortable: even a whole dataset can mislead you, if you trust its first visible pattern before asking what produced it.

Your first graph#

In this week’s session, you built a time-of-day distribution from the collision log, a synthetic dataset built for this unit. It looked roughly like this:

More records at dawn and dusk than in the middle of the day or the middle of the night. It’s tempting to read that off in one line: kangaroos are more active at dawn and dusk, so more collisions happen then.

Hold that sentence at arm’s length for the rest of this lecture. It is not necessarily wrong — but it bundles together a behavioural claim and a recording claim, and the graph alone can’t tell you which one you’re looking at.

What is the graph actually a graph of?#

A time-of-day distribution of collision records answers one question directly: which times are recorded in these rows? That is not the same question as when are kangaroos on the road?, and the gap between them doesn’t close just by recording things properly. Even if every collision that happens were recorded, and recorded with an accurate time, a collision record would still only exist at the moment an animal, a vehicle and a stretch of road all coincide — it cannot directly measure the roadside animal activity that never produces a collision at all.

A pattern in the spreadsheet may describe kangaroos — or it may describe the person holding the clipboard. Both explanations are consistent with the same 34 dusk rows. Telling them apart is this week’s actual skill, and the dataset audit you ran in the session is where you practised it. The rest of this lecture explains why that audit worked, and what it means for the conclusion you’re allowed to draw.

Crepuscular activity, and its limits#

Crepuscular means most active around twilight — dawn and dusk — rather than in full daylight or the depths of night. It’s a real, well-documented behavioural tendency in many Australian mammals, kangaroos included: a GPS-tracking study of Eastern Grey Kangaroos in a peri-urban community on the New South Wales north coast found their movement rate was highest at dawn (Henderson, Vernes, Körtner & Rajaratnam, 2018).

Notice exactly what that study measured: how far and how fast a tracked kangaroo moved, over a period of days, for a small number of animals in one location. It is genuine evidence that dawn movement activity rises. It is not a measurement of collision risk, a national activity timetable, or a guarantee that this dataset’s dusk peak is animal behaviour rather than something else. Crepuscular is a behavioural pattern, not a booking confirmation for 6:00pm — it describes a tendency across a population, not a fixed schedule any individual animal keeps.

Two things can both be true: kangaroos really are more active around twilight, and that fact alone cannot explain a specific spike in a specific dataset. The first graph is not necessarily wrong. The first explanation usually is — not because it’s false, but because it’s incomplete, and incomplete in a way that feels finished.

Why kangaroos approach roads at all#

Twilight activity explains when kangaroos move. It doesn’t explain why that movement brings them onto a road corridor specifically. Several ordinary pressures do:

• food and water access — roadside verges are often the most recently disturbed, most open grazing available, and cuttings can hold moisture longer than the surrounding paddock;
• habitat fragmentation — a road frequently cuts straight through a home range or a route between two patches of habitat, so crossing it isn’t a choice so much as a consequence of where the road was put;
• movement routes — established tracks between shelter, water and grazing don’t reroute themselves around a new road;
• seasonal conditions — drought pushes animals to range further for food and water, and joey season changes group movement and risk-taking around a mob;
• group movement — mobs move together, so one animal crossing well often means several more are about to.

None of this needs to be dramatic. Most of it is a kangaroo doing what a kangaroo usually does, on a landscape that happens to have a road through it.

The driver’s side of the same system#

Dawn and dusk aren’t just when animals move — they’re also when driving conditions get worse in ways that have nothing to do with animal behaviour:

• visibility drops as ambient light falls, cutting the distance at which a driver can pick an animal out of roadside shadow;
• glare from a low sun sitting near the horizon at both dawn and dusk can be worse than full darkness, when headlights do most of the work;
• traffic volume peaks at dawn and dusk anyway, because that’s when most people commute — more vehicles on the road at exactly the hours animal activity is rising;
• roadside vegetation can block the sightline to the verge earlier than a driver would otherwise notice motion there.

Every one of these factors would, on its own, produce more collisions clustering at dawn and dusk — with no reference to kangaroo behaviour at all.

A model for what “observed collisions” actually contains#

A collision record isn’t a direct readout of animal behaviour. It’s the product of several things all happening together:

Observed collision records = animal presence × vehicle exposure × probability of collision (given presence and exposure) × probability of reporting

Change any term and the observed count changes, without anything about kangaroo behaviour changing at all. More traffic at dusk raises vehicle exposure. Worse dusk visibility raises the probability of collision given presence. A wildlife carer’s reporting habits can raise or lower the probability of reporting, independent of anything an animal did. A graph of observed records is a graph of all four terms multiplied together — never of animal presence alone.

Back to the dataset: an inconveniently exact appointment#

In the session, you looked closely at the timestamps behind those 34 dusk records, rather than just the day-period label. A disproportionate number sit at exactly 06:00 or exactly 18:00 — not “around dawn,” not “roughly early evening,” but the same two clock times, repeated.

Real kangaroo behaviour is not that punctual, and neither, generally, is traffic. An unusually exact spike at one specific timestamp is a strong signal that something about how the row was made — not what the animal did — produced it.

Checking report_source narrows it further: the exact-06:00 and exact-18:00 rows come overwhelmingly from one source — wildlife carer call-outs — not from police reports or road-authority logs. And data_quality_flag names the problem directly on those rows: the timestamp records the carer’s shift-start time, not the time the collision happened. Eleven records in the dataset carry exactly this flag.

Either the kangaroos became remarkably punctual, or a human system has entered the graph. It’s the second one. A rostering habit — carers logging a case at the start of their shift rather than backdating it to when the animal was found — has been quietly counted as if it were evidence about when kangaroos cross roads.

Revising the conclusion#

Here’s the comparison worth carrying into every other week that uses this dataset:

The first explanation in each row isn’t ruled out by this lecture — it’s just no longer the only thing the pattern is consistent with, and the dataset alone can’t adjudicate between them.

Once the eleven shift-time records are set aside, the honest version of this week’s finding is:

The synthetic dataset contains more dusk records, but some timestamps reflect wildlife-carer shift times rather than incident times. The remaining pattern is consistent with several explanations, including animal activity, traffic exposure and visibility, which cannot be separated using these records alone.

That’s a less satisfying sentence than “kangaroos are crepuscular, so dusk is dangerous.” It’s also the one the data can actually support.

Where this goes next#

This week found a temporal pattern that partly dissolved on inspection. Next week does the same exercise in space instead of time: a map of recorded strikes looks authoritative in a way a spreadsheet column never quite does, which is exactly why it deserves the same scepticism you just practised here. Next week we put the problem on a map, where it will immediately look more authoritative.

Reading#

Henderson, T., Vernes, K., Körtner, G., & Rajaratnam, R. (2018). “Using GPS Technology to Understand Spatial and Temporal Activity of Kangaroos in a Peri-Urban Environment”, Animals, 8(6), 97. GPS-tracking evidence for the dawn activity peak this lecture leans on — read it as evidence about movement, not as a study of collisions.