Week 3 Session: Mapping the hotspots

Turning raw collision counts into a defensible hotspot ranking — and watching that ranking change once traffic and road length are accounted for.

8 March 2027

The road network table#

This session adds a second dataset to the one you already know. Download the road network table (CSV, 7 rows) — like the collision log, it’s synthetic teaching data built by the teaching team, covering exactly the seven roads that already appear in the collision log. It is not a survey of any real road, and it does not contain coordinates or a road network you could plot on a real map.

Data dictionary

A schematic, not a map#

Neither dataset gives you coordinates, so nothing in this session asks you to plot an exact collision location. Instead, a road schematic shows counts or rates at the road level — one bar or one mark per road, arranged for comparison, not for geography. Here’s the format, shown with four hypothetical roads and made-up counts — not this week’s actual data, which you’ll tally and rank for yourself in step 2 below:

Bar length is proportional to count only. Road order, spacing and bar position are arbitrary — this schematic represents aggregated records by road, not any real geography, road shape, or exact collision location. You will build the actual raw-count schematic, and the equivalent schematic for a rate instead of a raw count, for this week’s seven real roads during the session — that’s where you’ll see how much the bars reorder.

The exposure formula#

You’ll need this in step 3 below, before the lecture that debriefs this session has a chance to give it to you:

Recorded collisions per million vehicle-kilometres = recorded collisions / (average daily traffic × days observed × road length in kilometres) × 1,000,000

Use an observation window of 1,409 days for every road — 4 January 2023 to 12 November 2026, counting both dates. Every road must be measured over the same span, or the comparison between roads stops being fair.

Before the session#

Bring your revised week 2 time distribution and skim the data dictionary above. You don’t need to open either CSV before the session starts.

Note: a record with an uncertain incident time (last week’s flagged records) is still fine to use here. Week 2’s timestamp problem was about when a collision happened — it doesn’t affect that record’s road_name, which is all this session needs.

In the session#

This runs as one 50-minute sequence, before the lecture debriefs it. Work in the same pairs as week 2.

1. 5 minutes — predict. Without opening either dataset, each pair predicts which road will look most dangerous, and writes one sentence saying why.
2. 10 minutes — first ranking. Open the collision log and count records by road_name. Rank all seven roads from most to fewest records. This is your first ranking.
3. 10 minutes — join and calculate. Join the road network table to your counts by road_name. For each road, calculate collisions per kilometre, then collisions per million vehicle-kilometres using the exposure formula above and its stated 1,409-day observation window. Show your working for at least two roads.
4. 10 minutes — choose a definition and re-rank. Using your calculated rates, decide on an explicit hotspot definition (a fixed threshold — for example, “at least double the median rate” — not a vague impression), apply it, and produce your revised ranking.
5. 10 minutes — peer challenge. Swap your first ranking, revised ranking and stated definition with another pair. Try to break their definition using the traffic and reporting-coverage figures: does a road get excluded that shouldn’t be, or included on the strength of good reporting rather than real risk?
6. 5 minutes — exit ticket. Submit one sentence naming which conclusion changed between your first and revised ranking, and why.

Afterwards#

This session’s output is the Hotspot Appeal Form — a one-page pair submission. It’s formative and ungraded: it’s practice at defending a threshold against a challenge, not a mark on which road you picked.

The Hotspot Appeal Form

• Your first ranking (from step 2).
• Your revised ranking (from step 4).
• Your annotated road schematic — bars or marks for all seven roads, labelled with the measure you used, in the format shown above.
• Your stated hotspot definition, in one sentence.
• Your exposure calculation, shown in full for at least two roads.
• One reporting-coverage limitation that could change your conclusion if it were different.
• A short statement of what your evidence actually supports — not what you suspect, what the numbers in front of you can defend.

Keep a copy of everything above. The week 6 assignment’s hotspot analysis asks you to state an exposure adjustment and defend a hotspot threshold for the same regional dataset — this session is where you practise doing that once, before you do it again for the assignment.