The plain bump won: speed-bump design and driver reaction time

Overview

Does making a speed bump look more dangerous make drivers react faster?

Colour theory has an intuitive answer. Red is the universal alert colour — danger, urgency, stop. So a red-and-white speed bump should be perceived sooner and braked for faster than a plain black one. We set out to test that assumption instead of trusting it.

The study measured how a speed bump’s visual design — its colour, and whether it carried a horizontal “KEEP SLOW” instruction painted on the road — changed drivers’ reaction time. To do it without a driving simulator budget, we built an entire test town inside Minecraft, disguised the real question so participants wouldn’t game their answers, and measured split-second decisions with cognitive-psychology software.

The result overturned the intuition. The red-and-white bump — the design chosen precisely because red signals danger — increased average reaction times. The winner was the humble one: a plain, asphalt-coloured bump with “KEEP SLOW” painted on the road, where 86.67% of the 30 participants reacted faster than to the reference bump, with a mean improvement of 33.08% among them. The lever that mattered wasn’t colour. It was a legible instruction in the driver’s line of sight.

Testing a hunch about colour

The work started from two hypotheses grounded in the literature:

1. Painting the bump red (with reflectors) would improve perception. Under colour theory, red reads as alert, danger and urgency — so it should shorten the time a driver takes to register the bump and slow down.
2. Adding horizontal “KEEP SLOW” signage would reinforce attention. On-pavement warnings and contrasting markings are documented to sharpen driver focus and encourage more cautious behaviour.

The interesting tension is between them. The red hypothesis is the seductive one — visually louder, emotionally coded, the kind of idea that feels obviously right in a review. The whole point of the study was to put that instinct on the same footing as a boring text label and let reaction time decide between them.

To pull them apart, we tested four bump conditions:

• Asphalt-coloured, no signage — the reference against which everything else was measured.
• Asphalt-coloured + “KEEP SLOW”.
• Red-and-white, no signage.
• Red-and-white + “KEEP SLOW”.

Each condition appeared across day and night runs, so low-visibility conditions were part of the test rather than an afterthought.

Building a driving test inside Minecraft

A real driving simulator wasn’t available, so we built one. The town was sketched first as a map in Procreate, then constructed block by block in Minecraft Java Edition (creative mode), using five mods to bring it closer to a real road: proper road blocks and horizontal markings, vertical signage and street furniture, and drivable vehicles. Routes through the town were screen-recorded and cut into decision segments in Adobe Premiere Pro.

Hiding the real question was a design decision, not a detail. If participants realised the study was about speed bumps, they’d anticipate every bump and their reactions would stop being honest. So the routes were seeded with a wide range of unrelated road events — pedestrian crossings (with and without pedestrians), a STOP sign, a roundabout, potholes, loose animals, crossing and overtaking vehicles, a give-way junction. Bumps were just one event type among many. The true focus was only revealed in the very last question of the post-test questionnaire.

Measuring the split-second decision

The reaction times themselves were captured in SuperLab 6, a cognitive-experiment tool that turns the recorded routes into a game-like sequence of stimuli. At each decision moment a screen appeared asking the participant to choose an action — press “A” to slow down, “P” to stop — and the software logged exactly how long they took to press. That interval is the reaction time.

Thirty participants took part — mostly 19–24, digitally comfortable, recruited from the Instituto Politécnico de Setúbal. Each signed a consent form, ran a tutorial and a practice (“simulation”) block to get used to the controls, then completed the day test and the night test, keeping their fingers resting on “A” and “P” so response reads were as clean as possible. An observation notebook recorded behaviour the timings couldn’t — hesitation, facial reactions, the occasional accidental key swap. Data was exported to Excel, tutorial responses were discarded, and every condition was compared against the asphalt-coloured reference bump using both absolute (ms) and relative (%) differences.

Results

Colour theory lost. Legible instruction won.

Measured against the plain asphalt reference bump, the four conditions separated cleanly:

• Asphalt + “KEEP SLOW” — the clear winner. The average reaction time across all 30 participants fell by 24.78%, and 86.67% (26 of 30) reacted faster than to the reference. Among those who improved, the mean improvement was 33.08%.
• Red-and-white — it backfired. The average reaction time rose by 62.08%, and only 23.33% (7 of 30) got faster. The design meant to shout “danger” actually slowed people down.
• Red-and-white + “KEEP SLOW” — signage pulled it back, but not to positive. The average still rose, but only by 10.10%, and 43.33% (13 of 30) improved. The instruction rescued some of the loss the colour created — without ever beating the plain bump.

The report’s own reading of why red failed is the most useful part for a designer: the red-and-white combination likely imposed higher cognitive load. An unfamiliar colour, plus letters, plus the surrounding vertical signage, is simply more information to parse in the moment a decision has to be made — and parsing costs time. The plain bump with a short, familiar instruction gave the eye one clear thing to read.

Two honest caveats on the numbers. First, that +62.08% mean is inflated by a single extreme outlier — one participant recorded roughly 3,086 ms on the red-and-white bump, several times the group norm — so the mean overstates how badly the condition did for a typical driver; the direction holds, the magnitude is fragile. Second, the study reports descriptive statistics only: percentages and averages, but no significance test, confidence interval or p-value. With n=30 and that much variance, the findings are a strong, honest signal — not a statistically confirmed effect.

Learnings

Evidence beat intuition — which was the entire point. The “obviously safer” red design lost to a plain bump with a text label. If we’d shipped on instinct, we’d have shipped the slower option. This is the case study I point to when I say I design for evidence rather than aesthetics: the flashier idea was on the table, and the data took it off.

Cognitive load is a design variable, not a footnote. More visual salience is not more communication. The red-and-white bump added information — colour, letters, competing vertical signage — and the driver paid for it in milliseconds. The lesson generalises straight to interface work: every element you add to a moment of decision is something the user has to process before they can act.

Disguising the objective protected the data. The pedestrian crossings, STOP signs and stray animals weren’t set dressing — they were the mechanism that kept reactions genuine by hiding the real variable. Good research design is as much about what you conceal from participants as what you measure.

The limitations are the roadmap. Descriptive-only statistics, a small and homogeneous sample, a mean skewed by one outlier, and a Minecraft world that is emphatically not a road. The honest next steps write themselves: a larger, more varied sample; inferential testing to confirm the effect; and a higher-fidelity simulation — because a blocky game can validate the comparison between designs, but only a realistic environment can validate the reaction times themselves.