Week 11#
Does the evidence travel?#
Moose, Deer, and the Limits of Borrowing#
Where the strongest evidence comes from#
Moose in Scandinavia. Deer in North America.
Not kangaroos. Not Australia.
That’s not a flaw in the literature — it’s where the big, long-running studies happened to be done.
Five questions before importing a finding#
- Same animal?
- Same road type?
- Same mechanism?
- What was actually measured?
- What local-context differences matter?
The outcome-type ladder#
- Usage rate
- Collision reduction
- Exposure-adjusted rate
- Injury outcome
- Cost-effectiveness
High usage ≠ fewer collisions. Week 7’s unfenced crossing structures proved that already.
Principle vs parameter#
Principles transfer. Fencing excludes animals from the carriageway, anywhere.
Parameters don’t. The exact percentage, the exact predictor weights, the exact cost per km — those are local facts about one study site.
Pilot effect size ≠ national rollout effect size, for the same reason.
Case: moose, Sweden#
Seiler (2005) — 2,000 accident sites vs 2,000 controls, logistic regression, validated on 2,600 road sections in Örebro county.
| Outcome | Share |
|---|---|
| Correctly classified | 72.7% |
| Misclassified | 27.3% |
Traffic volume, speed, and fencing were the dominant predictors. The 72.7% is Sweden’s number, not a promise for anywhere else.
Case: deer, Utah#
Bissonette & Rosa (2012) — fencing + 2 underpasses + escape ramps on I-15, 1998–2003 pre vs 2005–2006 post.
| Area | Mortality decline |
|---|---|
| Treatment (fenced + underpasses) | 98.5% |
| Control (unfenced) | 2.9% |
No end-of-fence increase detected here — in tension with Week 7’s general end-of-fence point. Both can be true; this fence’s specifics aren’t stated in the source.
Testing it on Grellan Creek Road#
22km, ~900 traffic/day, weekly patrol, mostly commuter-reported.
Same animal? No — kangaroo, not deer. Same road type? Partly — rural two-lane, not an interstate. Same mechanism? Plausible. What local differences matter? Weekly, commuter-biased reporting stacks the Week 3 exposure problem on top of the transfer question.
Verdict: mechanism is a reasonable hypothesis. The effect size is not a promise.
Next: the final project#
At least three strands, and overseas evidence can be one of them —
but only the mechanism travels for free. Any specific number needs to be framed as a hypothesis to test locally, not a result to cite as settled.