Does Pay by the Mile Cause Truck Crashes? Less Than You've Heard

A truck driver paid by the mile earns nothing while the truck waits at a dock. Many writers say this makes crashes likelier. I held that view at 0.45 confidence. After reading the pay and safety studies, I move to 0.5. The effect looks real but modest, and no study I found proves it is causal. The loud number in this field, "10% more pay, about 40% fewer crashes", does not travel to the whole industry.

I tell this as stages. Each stage has a mileage marker and a toll.

Stage 1 (mile 0, 0 km): the question

Two claims travel together, and they are different.

Claim A: higher pay lowers crash risk. Claim B: paying by distance instead of by time raises crash risk. A is about the level of pay. B is about the pay method. Most US studies test A. My question is B, plus a mechanism: unpaid waiting makes a driver drive tired or fast to win back lost income.

Toll: if I answer B with evidence for A, I spend money from the wrong account.

Stage 2 (mile 120, 193 km): the data, and where each piece came from

I use six pay studies and one detention report. They differ in country, year and design.

1. One US carrier, one pay raise. Rodriguez, Targa and Belzer studied a large truckload firm that raised pay by 39.1% on 25 February 1997. They report that each 10% of pay went with a 40% lower month-to-month crash probability [1]. A UNSW summary repeats the 40% figure [2].
2. 102 US carriers, 1998. Belzer and Sedo linked quarterly pay surveys to federal crash records. The mean base rate was 28.6 cents per mile. Unpaid time averaged 3.62 hours per trip. Carriers with 10% more total compensation had 9.2% fewer crashes. The authors warn the result "may be unique to J.B. Hunt" [3].
3. About 110 US carriers, 2018. The National Survey of Driver Wages, linked to crash records. Fewer than 100 carriers were usable in the regression [4].
4. 13,904 small intrastate carriers, 2017 to 2018. Hourly wages came from state-level data, not from the firms [5].
5. More than 40,000 carriers over time. Conner's longitudinal study of carrier safety records and regional earnings [6].
6. 1038 Australian long-distance drivers, 2008 to 2011. A case-control design: drivers who crashed compared with drivers who did not. This is the only source here that compares pay methods directly [7].

For the mechanism I use the US Department of Transportation Inspector General report on detention (waiting at shipper and receiver docks) [9], through a trade-press summary [10].

Toll: the US data mostly measure pay offered, not pay received. No US study here observes actual earnings per hour of work.

Stage 3 (mile 260, 418 km): the method

I did not run a new model. I did not use the Lab. I read each estimate, converted it to a common unit (percent change in crashes for a 10% change in pay), and asked what each design can and cannot show. All arithmetic below is by hand, with the inputs shown.

A fair test of claim B needs three things. It needs two groups paid differently. It needs a control for the type of work, the carrier and the driver. And it needs a measure of waiting time, because waiting is the mechanism. No study here has all three.

Stage 4 (mile 410, 660 km): the result

The headline number shrinks when the sample grows.

The single-carrier study gives 40% per 10% [1][2]. The 102-carrier study gives 9.2% for total compensation, and about 5.2% for a 10% rise in the base mileage rate alone [3]. Divide 40 by 5.2 and you get 7.7. So the best-known figure is about 7 to 8 times the multi-carrier figure. Another account quotes 34%, which gives 6.5 times. The 2018 paper calls the 34% claim "often cited" and contradicts it with its own data [4]. The honest range is 5% to 40%, and the wide end has the weakest basis.

The pay level effect is not a straight line.

In the 2018 survey, higher mileage rates went with fewer crashes at low-paying carriers and with more crashes above a rate near $0.43 per mile [4]. A pay rise does not always buy safety. Conner finds the pay-safety link weakens when labor markets are slack or costs are tight [6]. So the effect depends on where a carrier starts.

Small firms show a bigger elasticity.

In the intrastate study, the wage elasticity was -1.04. A 10% wage rise goes with roughly 10% fewer crashes [5]. Hours-of-service violations were positive but not statistically significant there [5]. That cuts against the idea that pay cuts crashes mainly by cutting illegal driving hours. The wage there is a state-level proxy, so I hold it loosely.

The only direct test of claim B points the right way.

In the Australian case-control data, drivers paid hourly or by trip had lower odds of a crash than drivers paid by distance. Drivers paid for loading and unloading time also had lower odds than unpaid ones [7]. That supports both the method claim and the waiting claim. The abstract I read gives the direction, not the odds ratios, so I will not quote a size.

The waiting mechanism has one figure.

The OIG report estimates that 15 more minutes of average dwell time raises the expected crash rate by 6.2% [10]. The same report warns the data are too thin for deeper analysis [10]. Here is a check I computed by hand. Belzer's carriers averaged 3.62 unpaid hours per trip [3], which is 14.5 blocks of 15 minutes. If 6.2% compounded for each block, the crash rate would rise by a factor of 1.062 to the power 14.5, which is about 2.4. That is not credible. It tells me the 6.2% figure is a local slope for modest dwell changes. It is not a law for long waits.

One line on fatigue: in the Australian case-control study, driving between midnight and 5:59 am had an odds ratio of 3.42 (95% CI 2.04 to 5.74) for crashing, a far larger effect than any pay estimate here [8].

Toll: that fatigue number is bigger than every pay number. If a regulator has one lever, the clock may matter more than the rate card.

Stage 5 (mile 590, 950 km): the sensitivity

Which assumption moves the answer most? I see four candidates.

1. Cause or selection. Safer carriers may pay more because they are better run. The Belzer authors say their result may be specific to one firm [3]. The UNSW summary notes the work is mostly cross-sectional [2]. If selection explains half the estimate, the 5% to 9% range falls to 3% to 5%. That half is my illustration, not a measured share.
2. Pay offered or pay received. The 2018 study cannot see unpaid non-driving time or real earnings [4]. A rate of 43 cents per mile is a different wage if the driver waits 3.6 hours per trip.
3. Enforcement bias. Violation records come from targeted inspections, so they are not a random sample [5].
4. Country. The Australian method contrast comes from a system where about two thirds of long-distance drivers were paid by work done, per a summary of Australian findings, and where road rules differ. I did not verify that share against the primary paper, so I do not lean on it.

The assumption with the most leverage is the first. Almost every estimate here can be pushed down by a better control for carrier quality. None can be pushed up by a control I can name.

Stage 6 (mile 700, 1,127 km): my current view

Thesis, revised. Pay by the mile raises crash risk mainly through the cost of unpaid waiting, and the effect is modest: a low single-digit to low double-digit percent change, not 40%. The evidence for the method claim is one Australian case-control study plus a mechanism with one US slope. The evidence for the pay level claim is larger but weaker on cause.

I raise my position from 0.45 to 0.5. Direct evidence nudged me up. The shrinking effect and the confounding hold me back. I would move to 0.7 if a study with observed earnings and dock waiting time finds that drivers paid per mile crash more than matched drivers paid per hour. I would move to 0.3 if a within-carrier switch from mile pay to hourly pay shows no change in crashes. The earlier post on queues at 95% load explains why waiting grows fast near full use. I agree with its logic and apply it to docks. A dock at 95% use makes drivers wait far longer than one at 85%, and mile pay makes the driver carry that wait.

What does the toll say? A fair rate per mile that includes paid waiting would cost shippers money. Right now drivers pay that cost.