Crash Barriers and the Market for Lemons

In 1970 an economist explained why it is hard to buy a good used car. The same argument, almost unchanged, explains why the cheapest metal beam crash barrier keeps winning the order — and why that is a road safety problem rather than a commercial one.

A Barrier is only worth what it does in the two seconds that matter

A crash barrier spends its entire life doing nothing. It stands at the edge of a carriageway for fifteen or twenty years, and for all of that time a compliant one and a deficient one behave identically: both look like a guardrail, both are galvanised, both are bolted to posts at two-metre centres.

The difference appears once, without warning, when a vehicle leaves the road at speed. A W-beam barrier is not a wall. It is an energy-absorbing system, designed to deform in a controlled way, hold the vehicle, and redirect it back along the carriageway rather than let it through, over, or under. Whether it does that depends on the steel — its thickness, its yield strength, the consistency of its properties along the length of the rail, and whether the posts and spacers deform as the tested system did.

By the time anyone finds out, the acceptance certificate was signed years ago.

Barriers on a completed carriageway. From the road, a compliant rail and an under-specified one are indistinguishable.

The paper that explained the used-car lot

George Akerlof's paper The Market for "Lemons": Quality Uncertainty and the Market Mechanism was published in the Quarterly Journal of Economics in 1970. It was rejected by three journals before it ran. Thirty-one years later it won him a share of the Nobel Memorial Prize in Economic Sciences, alongside Michael Spence and Joseph Stiglitz, for work on markets with asymmetric information.

The argument is short. Imagine a market for used cars containing good ones and bad ones — "peaches" and "lemons". The seller knows which is which. The buyer does not. Unable to tell them apart, the buyer will rationally offer something like the average value of the two.

That average is an insult to the owner of a good car, who withdraws it from the market. Now the pool is worse than it was. Buyers notice, and lower their offers again. That pushes out the next tier of decent cars. The loop repeats.

Akerlof's point was not that buyers are foolish or that sellers are crooks. Every participant behaves rationally at every step. The market still degrades, because the information required to reward quality is not available at the moment the decision is made. Economists call the result adverse selection. In the extreme, the market for good cars stops existing altogether.

The adverse-selection cycle applied to crash barriers. Nobody has to act in bad faith for the average to fall. Each step is individually rational.

Why a Crash Barrier is a textbook Lemon

Used cars are the famous example, but they are a mild case: a buyer can at least take a car to a mechanic. Metal beam crash barriers sit closer to the pure form of the problem, for four reasons.

The quality difference is invisible at handover. A rail rolled from 2.5 mm coil looks exactly like one rolled from 3 mm. Galvanising at 200 g/m² looks exactly as bright as galvanising at 550 g/m². Both differences are smaller than the width of a human hair, or close to it.

The feedback arrives far too late. Under-galvanised steel does not rust on the day it is installed. It rusts in year four instead of year twenty. Under-thickness rail does not fail in the yard. It fails during an impact, years later, by which time the contract is closed, the retention is released, and nobody is systematically attributing that failure back to a coil specification.

The person paying is not the person at risk. A motorist who hits a barrier had no part in choosing it, cannot inspect it, and will never know what it was made from. This is the sharpest difference from Akerlof's used cars: in that market the buyer bears their own bad decision. Here the consequences are borne by someone with no seat at the table.

Award by lowest price completes the circuit. Where the contract goes to the lowest bidder and the acceptance test is essentially visual, price is the only number anyone can truly verify. Everything else on the specification sheet is a promise. Akerlof's model does not need anyone to cheat for quality to fall — it only needs quality to be unverifiable and price to be decisive. Both conditions are routinely met.

Three (3) ways quality leaves the building -

These are not exotic frauds. They are ordinary, quiet economies, each of which produces something that still passes as a crash barrier.

1. Thickness

IRC:119-2015 specifies a 3 mm W-beam rail element. Steel is sold by weight, so thickness is the most direct lever on cost there is. Rolling at 2.5 mm instead of 3 mm removes about 16.7% of the steel from the rail — roughly 1.9 tonnes per kilometre of single-sided barrier.

The bending capacity of a thin-walled section is approximately proportional to its wall thickness, so that rail retains something like 83% of the moment capacity it was designed around. The barrier has not become useless. It has become a barrier designed for a different, gentler impact than the one the road will eventually deliver.

Note that a tolerance is not a target. AASHTO M180 permits a specific under-run — 0.23 mm — below nominal thickness. That is an allowance for the realities of rolling, not licence to aim for the bottom of the band and treat the shortfall as margin.

2. Galvanising

Zinc is expensive, and coating mass is the easiest thing on a barrier to reduce without anyone noticing, because a thin coating and a thick one are the same colour.

A 550 g/m² coating — the figure Indian contracts commonly carry, and the same value as AASHTO M180's Type I, measured as a total across both faces — works out to about 38.5 microns of zinc on each side. Zinc protects steel by corroding in its place, at a rate set by the environment. Under ISO 9223 those rates run from roughly 0.7–2.1 µm per year on an inland highway to 4.2–8.4 µm per year in a coastal or heavily industrial location.

Divide one by the other and the arithmetic is unforgiving. Coating life is close to linear in coating mass: halve the zinc and you halve the years before rust reaches the steel. On a coastal highway, a barrier coated at 200 g/m² can begin failing inside three years.

Years before first rust, by zinc coating mass and environment. Coating thickness divided by Zinc corrosion rates. Coating life is close to linear in coating mass.

3. Inconsistent steel from secondary re-rolled coil

This is the least visible of the three and the hardest to reason about, because the problem is not a number being low — it is a number being unreliable.

Coil produced by re-rolling scrap-melted secondary material can carry chemistry and mechanical properties that vary along the length of the coil and between coils. A test coupon cut from one end may pass comfortably while material two hundred metres along the same coil sits well below it.

For most products, variability is a quality nuisance. For a crash barrier it is a design problem. The system is engineered to deform predictably — the rail must yield, stretch and absorb energy in a known way for the tested geometry to behave as tested. Steel with inconsistent properties can crack where it was meant to stretch, or shear at a bolt hole instead of transferring load along the rail. A barrier that fails unpredictably is arguably worse than one that is uniformly weak, because nothing about its behaviour can be designed around.

The part Akerlof was hopeful about -

The paper is usually remembered for the death spiral. Its final section is more useful, and it is where the practical answer lives.

Akerlof observed that real markets do not always collapse, because they evolve what he called counteracting institutions — mechanisms that move information across the gap between seller and buyer. He named guarantees, brand names, chains, and licensing and certification.

Every one of those has a direct equivalent in this industry:

• Guarantees become the mill test certificate, tied to the batch that actually became your rail.

• Brand names become manufacturers with a long enough record that a failure would cost them more than the saving.

• Licensing and certification become full-scale crash testing at an accredited laboratory, and third-party testing of the delivered lot.

The economics here are worth stating plainly. None of these mechanisms make the honest product cheaper. What they do is make the dishonest one detectable — which changes what the cheap price actually buys, and therefore changes who wins. Verification is not administrative overhead. In a lemons market it is the only thing standing between a specification and a piece of paper.

A galvanised W-beam rail. Thickness and coating mass are both measurable in minutes — but only if someone measures them.

Five checks that take the guesswork out

None of these require specialist knowledge. Most can be done on site, on the day of delivery, by the person receiving the material.

Check 1 — Weigh a beam

This is the fastest and most under-used check in the industry, and it needs nothing but a weighbridge or a platform scale. A 4.318 m W-beam rail at 3 mm contains about 49 kg of steel, plus roughly 1.1 kg of zinc at 550 g/m² — call it 50 kg galvanised. The same rail at 2.5 mm weighs about 42 kg. That is an 8 kg gap on a 50 kg part, and no amount of finish quality hides it. Weigh a random beam from the truck. If it is materially light, everything else on this list becomes urgent.

Check 2 — Measure the thickness, in the right place

Use a micrometer or an ultrasonic thickness gauge. Measure on a flat portion of the web, away from the corrugation radii, where forming has not thinned or work-hardened the material. Take several readings per beam and several beams per lot. Assess two things separately: the average against nominal, and the lowest single reading against nominal minus the permitted tolerance. A lot whose average is acceptable can still contain individual rails that are not.

Check 3 — Measure the zinc

A magnetic or eddy-current coating thickness gauge gives a non-destructive reading in microns in seconds. To convert to the units your specification uses: microns per side × 7.14 ≈ grams per square metre per side, and doubling that gives the both-sides figure. So 550 g/m² total corresponds to about 38.5 µm on each face.

Take readings at multiple points along the rail — coating that is adequate in the middle and thin at the ends is a real and common pattern. Where a reading is disputed, the referee method is gravimetric: strip the zinc from a measured area and weigh the difference. That result is not arguable.

Check 4 — Test the traceability, not the paperwork

Ask for the mill test certificate, then ask the harder question: can the supplier tie it to the steel in front of you? This is where most quality systems quietly fail, and it is worth understanding why. A coil's heat number lives on a tag hung on the coil — it is not marked continuously along the strip. Once that coil is decoiled, roll-formed and sheared, the individual pieces carry no identity of their own. Galvanising then removes whatever surface marking remains: the pre-treatment pickles the steel in acid precisely to strip everything off it, and the zinc covers the rest. Only deformation survives. AASHTO M180-23 has moved in exactly this direction, adding marking requirements for traceability.

Which means the honest test is not reading a number off the steel. It is a traceability drill: take a beam at random from the delivery, hand it back, and ask for its coil number, heat number, mill certificate, production date and galvanising batch within 24 hours. A manufacturer running a real system produces it. One that cannot has told you something more useful than any certificate would have.

Check 5 — Verify the metal, not the paper

Every check above can be defeated if the supplier chooses the sample, and every document can be defeated if you cannot trust who wrote it. The way past both is to test the steel itself.

Pull material at random from the delivered lot and send it to an independent NABL-accredited laboratory for tensile testing, chemical composition, coating mass and coating adhesion. Write this into the contract as a routine lot-acceptance step rather than something triggered by suspicion. A supplier who knows a random third-party test is coming behaves differently from one who does not — which is the entire mechanism, and the reason it works even when the result comes back clean.

How to write a specification that cannot be quietly met

Most specifications in this market are not wrong. They are simply unenforceable, because they state a requirement without stating how it will be verified or what happens when it is not met. Four additions close most of the gap:

1. State the acceptance method next to every number. Not "550 g/m² zinc coating" but "550 g/m² minimum, both faces, verified by coating thickness gauge on every lot."

2. Specify the sampling rule and who picks the sample. Frequency, sample size, and random selection by the purchaser or their representative.

3. Require a traceability drill, not a filing cabinet. The supplier must be able to trace a randomly selected delivered beam back to its coil, heat and galvanising batch within a stated time.

4. Attach a consequence. A requirement with no rejection clause is a preference. Define what happens to a failing lot before the first delivery arrives, not after.

This was never really about cost

It is tempting to frame all of this as value engineering — a debate about lifecycle cost, early replacement, and whether the cheap barrier is a false economy. It usually is a false economy, and that argument is easy to win. But it is not the important one.

A crash barrier is one of the few products where the buyer, the user, and the person who bears the consequences of a bad decision are three different people. The motorist who leaves the carriageway at 90 km/h on a wet night did not choose the barrier, cannot inspect it, and has no way to know whether the steel about to absorb their impact was rolled to 3 mm or 2.5 mm.

They are relying, completely and unknowingly, on someone upstream having checked. Akerlof's insight was that markets do not supply that checking on their own. Somebody has to build it in.

The tools to do it are not expensive. A platform scale, a coating gauge, a micrometer, a lab account, and the willingness to reject a lot. That is the whole list.

Frequently Asked Questions (FAQ)

How can I tell if a Metal Beam Crash Barrier meets specification?

Weigh a beam first — a 4.318 m W-beam rail at 3 mm should weigh about 50 kg galvanised, and an under-thickness rail is roughly 8 kg lighter. Then measure wall thickness with a micrometer or ultrasonic gauge on a flat part of the web, measure zinc with a coating thickness gauge, run a traceability drill on the mill test certificate, and send a randomly chosen sample to an independent accredited laboratory.

What thickness should a W-beam crash barrier be?

IRC:119-2015 specifies a 3 mm W-beam rail element for Indian highway use. AASHTO M180 defines two classes — Class A at 2.67 mm and Class B at 3.43 mm — and permits an under-run of 0.23 mm below nominal. That tolerance is an allowance for rolling variation, not a target to aim at.

How much zinc coating does a crash barrier need?

Indian contracts commonly specify 550 g/m² measured across eah face, which matches AASHTO M180 Type II at 1100 g/m². Because zinc corrodes at a rate set by the environment, coating life is close to linear in coating mass — halving the zinc roughly halves the years before rust reaches the steel.

How do you measure Galvanising thickness on a Crash Barrier?

Use a magnetic or eddy-current coating thickness gauge, which reads in microns non-destructively in a few seconds. To convert, multiply microns per side by 7.14 to get grams per square metre per side, then double it for the both-faces figure. Take readings at several points along the rail, since coating is often thinner at the ends. Where a reading is disputed, the referee method is gravimetric — strip the zinc from a measured area and weigh the difference.

How do you trace a galvanised crash barrier back to its steel heat number?

Usually you cannot do it piece by piece, and a supplier who claims otherwise deserves scrutiny. A coil's heat number lives on a tag, not along the strip, and galvanising pickles and coats away surface marking. Only deformation survives — a die stamp, dot-peen mark or embossed code at 300 to 500 microns reads through a 38-micron zinc coating. Realistic traceability is therefore batch-level and documentary: heat certificate, production run record, galvanising batch log, dispatch note. Test it with a traceability drill — hand back a random beam and ask for its full history within 24 hours.

Why do substandard Crash Barriers keep winning?

Because quality is unverifiable at the moment of purchase and price is not. This is George Akerlof's 1970 market for lemons: when a buyer cannot distinguish good from bad, they rationally pay an average price, compliant suppliers cannot survive at that price, and average quality falls. No one has to act in bad faith. Award by lowest price with visual-only acceptance testing completes the circuit.

Can a Mill Test Certificate (MTC) prove steel quality?

Only if you can tie it to the material actually delivered, which is harder than it sounds. Even then a certificate is a trust mechanism, not a verification mechanism.