Texting and Driving: The Famous Number Is About Truck Drivers

The risk is real, it has been measured well, and the number everybody quotes is not the number for you.

“Texting makes you 23 times more likely to crash” comes from a study of 203 commercial motor vehicle drivers in 55 trucks. It is genuine federal research. It does not describe ordinary motorists, and its dataset contains 21 actual crashes.

The figure for ordinary drivers, from the largest naturalistic study ever run, is 6.1. For comparison, in the same dataset, alcohol impairment is 35.9.

Both facts should change how you talk about this. Texting while driving is dangerous. It is not the most dangerous thing measured, and the campaign statistic overstates it by roughly four times while attaching it to the wrong population.

What was actually measured

The best evidence here does not come from simulators. It comes from instrumenting real cars and waiting.

Dingus and colleagues (2016), in PNAS, analysed the SHRP 2 Naturalistic Driving Study: 3,542 drivers aged 16 to 98, over 35 million miles, and 905 injurious or property-damage crash events.

Behaviour	Odds ratio (95% CI)	Prevalence in normal driving
Handheld - talk	2.2 (1.6-3.1)	3.24%
Handheld - text	6.1 (4.5-8.2)	1.91%
Handheld - dial	12.2 (5.6-26.4)	0.14%
Handheld - browse	2.7 (1.5-5.1)	0.73%
Handheld - reach for phone	4.8 (2.7-8.4)	0.58%
All handheld phone use	3.6 (2.9-4.5)	6.40%
Reaching for a non-phone object	9.1 (6.5-12.6)	-
Emotional state (anger, sadness)	9.8 (5.0-19.0)	-
Drowsiness or fatigue	3.4 (2.3-5.1)	-
Drug or alcohol impairment	35.9 (17.0-75.8)	-

Two things in that table are worth sitting with.

Dialling is worse than texting, at 12.2 against 6.1 - but it is also far rarer, at 0.14% of driving time against 1.91%. Risk per moment and total exposure are different quantities.

Reaching for a coffee cup (9.1) rates higher than sending a text (6.1). That is not an argument for texting. It is an argument that the mechanism is eyes and hands off the road, not phones as such.

The paper's own headline estimate: “Potentially 36%, or 4 million of the nearly 11 million crashes occurring in the United States annually, could be avoided if no distraction was present.”

Where the 23 comes from

Olson, Hanowski, Hickman and Bocanegra (2009), in a Federal Motor Carrier Safety Administration report, studied 203 commercial motor vehicle drivers, 55 trucks, 7 fleets, 16 locations, roughly 3 million miles.

Verbatim: “The most risky behavior identified was ‘text message on cell phone,’ with a significant odds ratio of 23.2.” The table prints OR = 23.24, 95% CI 9.69-55.73.

Three qualifications belong with that number every time it is used.

It describes commercial truck drivers. Different vehicle, different task, different exposure, different professional population.

The outcome is overwhelmingly not crashes. The study logged 4,452 safety-critical events, of which 21 were actual crashes - the rest were 197 near-crashes, 3,019 crash-relevant conflicts and 1,215 unintentional lane deviations.

The confidence interval runs from 9.69 to 55.73. That is not a precise estimate of anything.

The figure is repeated without any of these qualifications, including by the research institute that produced it. The defensible number for ordinary drivers is 6.1.

The hands-free question is genuinely unresolved

This is the most interesting disagreement in the field, and honest coverage of it is rare.

The laboratory says hands-free does not help.

Strayer and Johnston (2001), in Psychological Science, ran two experiments (N = 48 and N = 24). Conversation on a phone more than doubled the probability of missing a target, and slowed responses (F = 28.9, p < .01) - and handheld and hands-free produced equivalent deficits. Listening to the radio or a book on tape did not, which located the cost in the generative demands of conversation.

One caveat that is almost never printed: despite “simulated driving” in the title, this was a pursuit-tracking task on a computer display with a joystick, not a driving simulator.

Caird, Willness, Steel and Scialfa (2008) meta-analysed 33 studies, 94 effect size estimates, total N around 2,000. Reaction time correlations: hands-free rc = .460, handheld rc = .546 - a mean increase of about 0.25 seconds. Their conclusion, verbatim: “Common sense and urban legend suggest that hands-free phones may be safer to use while driving, but handheld and hands-free phones produced similar performance decrements.”

The real world says something different.

Klauer and colleagues (2006), in the 100-Car Naturalistic Driving Study (101 primary drivers, 69 crashes, 761 near-crashes), found dialling a handheld device at OR = 2.8 (1.6-4.9) but talking or listening on it at OR = 1.3 (0.9-1.8) - not significant.

Klauer and colleagues (2014), in the New England Journal of Medicine, compared 42 novice drivers and 109 experienced drivers. Among novices, dialling, texting, looking at the roadside, eating and reaching all carried elevated risk. Among experienced drivers, only dialling did - OR 2.49 (1.38-4.54). Conversation did not emerge as a risk in either group.

Dingus is the exception, finding handheld talk at 2.2 - and that estimate has been challenged. Young (2017), in Safety, argues the case and control groups were mismatched, with talk-exposed drivers also performing other secondary tasks while the comparison group performed none, inflating the estimate by roughly 45%. His adjusted figures are near or below 1.0.

We are flagging that Young is affiliated with a private consultancy rather than a university, that we located no funding statement for the paper, and that his conclusion - that conversation may be protective - is an extraordinary claim. It is a contested reanalysis, not a settled correction.

The resolution we think the evidence supports: the manual and visual acts are implicated consistently and strongly in every real-world dataset. Conversation is where the laboratory and the road disagree. Going hands-free removes the part that the naturalistic data actually implicates, and leaves the part only the laboratory reliably detects. That is why hands-free laws underperform without the underlying cognitive science being wrong.

The drunk-driving comparison, precisely

Strayer, Drews and Crouch (2006), in Human Factors, is the source of “phone use is as bad as drink-driving.” Here is what it says and what it found.

The published conclusion, verbatim: “When driving conditions and time on task were controlled for, the impairments associated with using a cell phone while driving can be as profound as those associated with driving while drunk.”

The study used a high-fidelity simulator with 41 adults, alcohol verified by breathalyser at 0.08.

Measure	Baseline	Phone	Alcohol
Brake reaction time	943 ms	1,022 ms	888 ms
Following distance	27.3 m	28.5 m	26.5 m
Collisions	0	3	0

Three things must be said. The claim is “can be as profound as” under controlled conditions, not “worse than.” The two impairments are qualitatively different - phone users braked about 79 ms slower and hung back; drunk drivers braked faster and harder but followed closer and drove aggressively. And the entire comparison rests on three simulator collisions among 41 people, with none in the alcohol condition.

Set that against Dingus’s real-world figures - texting 6.1, alcohol 35.9 - and the popular equivalence does not survive.

Do the laws work?

Abouk and Adams (2013), in the American Economic Journal: Applied Economics, verbatim: “bans appear moderately successful at reducing single-vehicle, single-occupant accidents if bans are universally applied and enforced as a primary offense. Bans enforced as secondary offenses, however, have at best no effect on accidents. Any reduction in accidents following texting bans is short-lived, however, with accidents returning to near former levels within a few months.”

Rocco and Sampaio (2016), in Empirical Economics, using US county data, found handheld bans reduced fatalities significantly while texting bans had a significantly smaller effect - the opposite of the priority legislators chose, and consistent with the naturalistic finding that the dangerous acts are manual.

What this leaves you with

Put the phone down before you drive. That advice survives every dispute in this article. The acts that take your eyes and hands off the road - dialling, texting, reaching - are implicated in every real-world dataset at every level of experience.

Do not quote 23 times. It is truck drivers, mostly near-crashes, with an interval spanning 9.7 to 55.7. Use 6.1, and say where it comes from.

Do not say texting is as dangerous as drink-driving. In the same dataset, alcohol is roughly six times the odds ratio of texting.

A headset is not the fix it is sold as - but the reason is subtler than “hands-free is just as bad.” It removes the part the crash data implicates and leaves the part only laboratories measure.

If you are a new driver, the numbers are different and worse. Distraction risk in this literature concentrates in inexperience.

Nothing here is a recommendation about law or enforcement, and no road-safety policy position appears in this article.

For the pedestrian counterpart and the same gap between simulator and crash data, see distracted walking. For what phones do to attention when the stakes are lower, see what notifications actually do.

Limitations of this article

We read the Klauer (2014) odds ratios in secondary summaries, not the New England Journal of Medicine paper, which we could not access. One confidence interval reported in those summaries is internally impossible, which tells you at least one transcription error exists in them. We have therefore printed only the two figures corroborated across both summaries and have described the paper’s pattern rather than its full table.

We read the Dingus odds-ratio table in a repository deposit, not the publisher's version.

The naturalistic evidence base is concentrated. Klauer 2006, Dingus 2016 and Olson 2009 were all produced by the same research institute, largely under US federal sponsorship. That concentration is worth knowing.

We did not cite the Caird (2014) texting meta-analysis, because we could not read it.

We did not print a widely circulated annual death toll attributed to texting while driving; we could locate no traceable research source for it.

We did not use insurance-industry analyses of ban effectiveness, having found only news coverage rather than primary documents.

We could not verify funding statements for the Strayer papers, and note that the Strayer 2006 sample is reported as 41 in the report version against a commonly cited 40.