Client€600M Global Industrial Technology Company
SprintAn 8 week sprint
Fieldwork123,282 rows analysed · 6 months · 12 sites · 7 transport modes
Digital and AI
The answer was that rain did not matter
A sensor manufacturer wanted to know whether its weather data, joined to third party traffic data, showed that rain changes how people travel. Six months of readings from collocated sensors were analysed against counts and speeds for seven modes. The honest answer, on this data, was no. That is a result, and it is one a client is entitled to be told clearly.
106,566
Rows of traffic count data analysed
16,716
Rows of vehicle speed data analysed
0.25 m/s
Difference in mean car speed between wet and dry days
0
Modes where people switched to another mode in the rain
A note on sourcing. Every figure here comes from a SprintlyWorks client engagement. Clients are described, never named. Where a figure is identified, modelled or indicative rather than banked, the line says so.
01
Three hypotheses, stated before the analysis
The work set out to test three things, and wrote them down first. That more rain reduces the number of pedestrians, cyclists and motorbikes. That more rain reduces vehicle speeds. And that fewer pedestrians and cyclists means more cars, because people who would have walked or cycled drive instead.
Six months of 2022 were analysed from sites where a weather sensor and a traffic sensor sit together, across twelve roads, split into three hour time slots and separated into weekdays and weekends. Days were classified as dry, damp or wet by how many hours of rain they carried.
Two approaches were used on each question: rain intensity at the moment, and total hours of rain across the day. The second is the better measure of whether someone decided not to go out.
02
Speed did not move
Average car speed behaved almost independently of cumulative rain intensity. Grouping the whole period into wet, damp and dry days for the middle of the day gave mean speeds of 10.84, 10.64 and 10.59 metres per second, across 730 wet rows, 1,799 damp and 1,499 dry.
That is a quarter of a metre per second between the wettest and driest conditions, and it points the wrong way. Cars were marginally faster in the wet.
One of the five roads did show a 25 per cent drop in average car speed on wet days. The other four showed a slight increase of two to three per cent. One site out of five moving against four is not a finding, it is a site with something else going on.
03
Counts moved on some days and not because of rain
Holding day, hour, location and month constant, pedestrian and cyclist counts showed no pattern against rain intensity. Days with heavy rain carried similar counts to days with none.
Measured by hours of rain rather than intensity, some rainy days did show large falls. On one road, a Tuesday with three hours of rain carried 48 pedestrians against 162 and 175 on two dry Tuesdays, a drop of more than 70 per cent.
The same slide records why that cannot be used. A third Tuesday that month was dry and carried 69, sixty per cent below the other two dry days. If a dry day can fall by sixty per cent, a wet day falling by seventy tells you nothing about rain.
Mode switching
Everything fell, including the thing that should have risen
Change in counts after two hours of rain, one month of weekday mornings across four roads. If people were switching from walking to driving, the car column would be positive everywhere.
| Road | Pedestrians | Cyclists | Cars |
|---|---|---|---|
| Road one | 20% down | 39.5% down | 1.1% down |
| Road two | 8% down | 71.1% down | 3.8% down |
| Road three | 29% down | 32.0% down | 14% up |
| Road four | 26% down | 76.6% down | 4.2% up |
How to read this → Two roads show cars rising. On one of them the increase of 40 cars is 70 per cent smaller than the fall of 71.6 pedestrians, so the people did not become drivers. They stayed at home.
SprintlyWorks analysis
04
Why the data could not be pushed further
Two limits were identified and stated rather than worked around. Rain intensity differs between roads at the same hour, sometimes dramatically, so weather and traffic cannot be aggregated across sites by time. On one August day two roads recorded no rain, two recorded moderate intensity and a fifth recorded much heavier.
And low rain intensity readings vastly outnumber high ones, so comparing traffic against rain intensity directly compares a very large sample against a very small one.
The journey time analysis that had been scoped was not attempted, because the mode of transport was unreliable in that data and hourly figures were not available. Dropping a workstream and saying so is better than running it badly.
05
What the client was left holding
A clear negative on all three hypotheses in this dataset, with the reasoning and the confounders written down. The unaccounted variables are named rather than hidden: journey purpose, distance travelled, drainage, and others.
And four conditions under which the question would be worth asking again. A city with much higher counts, so the base is large enough for a signal to show. A country where heavy rain falls in a defined season, so wet and dry are genuinely different. One or two years rather than six months. And the journey time analysis, on data that supports it.
There is no business result in this work. No revenue, no cost, no efficiency. What it produced was an answer, and the answer was that the relationship the client hoped to sell on is not visible in this data. Knowing that in eight weeks is worth more than a hedged maybe.
Download the full case study
Have a similar requirement?
Contact us today to learn more about on-demand workforce and accelerate development on your most pivotal projects!
Quick Reads for Big Impact
Featured Case Studies
Quick Reads for Big Impact
Accelerating Success for Enterprises in 20+ Geographies
Launch Your Sprint with
Define your project, connect with top-tier consultants, and start making progress fast.