Week 11: Does the evidence travel?

Two real overseas case studies, and the five questions to ask before importing anything they found into a different road, country, or species.

3 May 2027

Open the slides

Most of the strongest evidence in this unit’s field comes from moose in Scandinavia and deer in North America — not kangaroos, and not Australia. That’s not a flaw in the literature; it’s just where the large, well-funded, long-running studies happen to have been done. The question this lecture teaches is how to use that evidence honestly: what transfers, what doesn’t, and how to tell the difference before writing it into a proposal.

Five questions before you import a finding

  1. Same animal? Moose, white-tailed deer, and kangaroos are not interchangeable — different size, gait, herd behaviour, and road-crossing patterns.
  2. Same road type? A fenced interstate freeway and an unfenced two-lane rural road are different engineering problems even before the animal is considered.
  3. Same mechanism? Is the intervention doing the same thing in both places (physically excluding animals from the carriageway), or does it only look similar on the surface?
  4. What was actually measured? A usage count, a collision count, an exposure-adjusted rate, an injury outcome, and a cost-effectiveness figure are five different things — see the outcome ladder below.
  5. What local-context differences matter? Traffic volume, speed limits, fence-end geometry, reporting practices, and species-specific behaviour can all change an intervention’s real-world result without changing its underlying mechanism.

None of “moose,” “deer,” and “kangaroo” is an internally uniform category either — a moose in dense forest and a moose on open tundra don’t cross roads the same way, and neither do rural-fringe kangaroo mobs and urban-fringe ones (Week 2’s synthetic dataset spans multiple site types for exactly this reason).

The outcome-type ladder

These are not interchangeable, and confusing them is one of the easiest ways to overstate an intervention’s real-world effect:

  1. Usage rate — how often animals use a crossing structure once it exists.
  2. Collision reduction — whether vehicle-animal collisions actually went down.
  3. Exposure-adjusted rate — collision reduction accounting for changes in traffic volume or animal population (Week 3’s material).
  4. Injury outcome — whether the collisions that still happen are less severe.
  5. Cost-effectiveness — outcome achieved per dollar spent.

A high usage rate (animals like using the underpass) says nothing directly about whether collisions fell — Week 7 already established that a crossing structure without fencing showed no detectable collision effect in Rytwinski et al. (2016) despite presumably still being usable. Climbing this ladder from “1” to “5” requires new evidence at every rung, not an inference from the rung below.

Basic principle vs specific parameter

Physical and ecological principles transfer reasonably well: fencing that physically excludes an animal from a carriageway will do that regardless of country, and a crossing structure only works if animals are willing to use it. Specific parameters — the exact percentage reduction, the exact predictor weights, the exact cost per kilometre — are local facts about the study site, its traffic, its species, and its fence geometry, and don’t automatically hold anywhere else.

Related: a pilot study’s effect size is not the effect size you’d get from a national rollout. Pilots are usually sited where the problem is worst and the local team is most engaged — a national programme built from many sites, contractors, and maintenance regimes typically regresses toward a smaller average effect.

Case one: moose, Sweden

Seiler, A. (2005). “Predicting locations of moose–vehicle collisions in Sweden,” Journal of Applied Ecology, 42(2), 371–382.

Case two: deer, Utah

Bissonette, J. A., & Rosa, S. (2012). “An evaluation of a mitigation strategy for deer–vehicle collisions,” Wildlife Biology, 18(4), 414–423.

A local scenario to test the questions on

Grellan Creek Road (Northern Ranges, 22km, roughly 900 vehicles/day, Medium reporting coverage) is patrolled on a roughly weekly cycle, and most reports come from commuting motorists rather than dedicated surveys — which means its own collision counts already carry the Week 3 exposure-bias problem, on top of whatever transfer question is being asked. Suppose someone proposed importing the Utah design wholesale for Grellan Creek Road. Running the five questions: same animal (no — kangaroo, not deer); same road type (partially — rural two-lane, not an interstate freeway, so traffic volume and speed differ substantially); same mechanism (the fencing-plus-crossing-plus-escape-ramp principle is plausible); what was measured (mortality decline, which is further up the outcome ladder than a usage count but still a different quantity from a kangaroo-specific exposure-adjusted rate); what local differences matter (Grellan’s weekly, commuter-biased reporting means even “before” and “after” counts here would need the same exposure-adjustment scrutiny as the sourced case studies, not just a raw before/after comparison). Conclusion: the mechanism is a reasonable starting hypothesis; the specific effect size is not something this evidence entitles anyone to promise.

Where this goes

The final project asks for an intervention argued from at least three of this unit’s strands. Overseas evidence like this is one strand worth using — but only the mechanism-level claim travels for free. Any specific percentage, predictor weight, or cost figure from Sweden or Utah needs to be framed as a hypothesis to test locally, not a result to promise.

Reading

Teaching team