Eight risks came back as high impact on federal-aid highway projects no matter which delivery method had been used. The federal study behind the current evidence on this question asked participating agencies to rate 31 risk factors for their impact on cost and schedule, project by project, and the eight that survived every method were delays in completing railroad agreements, project complexity, uncertainty in geotechnical investigation, delays in the right-of-way process, unexpected utility encounters, work zone traffic control, challenges obtaining environmental documentation, and delays in the delivery schedule.
Not one of those is created by a procurement choice, and no procurement choice removes any of them. Several were settled years earlier, during the environmental and right-of-way sequence traced in how highway construction projects are planned. What the choice of method decides is who is holding each risk when it lands, and what that person is being paid to hold it.
Three contracts, and the payment method is where risk actually moves
The federal definitions are narrower than the popular ones. Design-bid-build is the traditional method: the agency contracts separately for design and construction services, the bid is based on complete plans and specifications, and the two phases run in sequence. Design-build means one entity contracted for both under a single contract. Construction manager/general contractor means the agency buys professional services from a construction manager during design, on a qualifications or best-value basis, negotiates a construction price once the design or an individual design package is finished, and the same firm then builds it.
Those descriptions say nothing about money, which is where the risk sits. Across the study’s 291 completed projects, design-bid-build was a unit-price contract 93 percent of the time, so quantity risk stayed with the owner. Design-build was lump sum on 85 percent of low-bid awards and 91 percent of best-value awards, which moves quantity risk to the builder. CM/GC split between unit price at 38 percent and a guaranteed maximum price at 56 percent, a ceiling that was negotiated rather than competed.
Procurement follows the same logic. Design-bid-build was awarded by low bid on 80 percent of projects, with most of the remainder using A plus B bidding that scores time alongside price. CM/GC used best value on 47 percent and qualifications-based selection on 41 percent, and low bid on none. Design-build ran both ways, and the study sorted it accordingly: 39 projects used price as the only factor, and 77 used at least one non-price factor. For the design-build projects that reported it, design was less than 30 percent complete at the request for proposals on more than three quarters of them. That figure is what transferring design risk looks like as a number.
The design that stops being a bidding document
The result that surprised the researchers concerns CM/GC, where the agency still takes design to 100 percent completion exactly as it does under design-bid-build. Mean agency design duration was 361 days for CM/GC and 932 days for design-bid-build.
The explanation offered is mechanical rather than motivational. Having the construction manager on the team lets the agency fast-track design, and there is no need to develop full designs for competitive bidding. The drawings still get finished. They stop having to be finished in a form that lets three strangers price the same thing on the same day.
Whole-project durations move with it, and here the study is careful. Against a mean design-bid-build duration of 1,774 days, CM/GC averaged 929 and low-bid design-build 889, while best-value design-build came in at 1,516. That is 48 percent shorter for CM/GC and about 50 percent for low-bid design-build, but the mean costs are not comparable: CM/GC and best-value design-build projects ran roughly twice the cost of the design-bid-build group, which the report reads as projects twice as large built in half the time, while the low-bid design-build projects were about half the cost.
Matched cost bands do the honest comparison. Between $2 million and $10 million, where mean costs were within a few hundred thousand dollars of each other, design-bid-build averaged 1,506 days against 773 for low-bid design-build, with agency design time down roughly 77 percent. Between $10 million and $50 million, design-bid-build averaged 2,130 days, best-value design-build 1,420 and CM/GC 662, with the CM/GC mean cost about 13 percent higher than design-bid-build. More than half of the duration difference in that band came out of design rather than construction.
Cost certainty is a program metric, and it separates the methods more cleanly than cost does
Cost certainty is the point at which the agency has a reliable project cost. In the $10 million to $50 million band it arrived on day 1,184 under design-bid-build, day 765 under best-value design-build, and day 329 under CM/GC. For an agency letting a program rather than a project, that is the difference between knowing what next year costs and hoping.
Project intensity, final cost divided by actual duration, tells the same story in the language of disruption: $12,802 per day for design-bid-build, $46,450 for CM/GC, $28,527 for best-value design-build and $12,816 for low-bid design-build. Read as productivity it flatters the contractor. Read as exposure it is a public measure, because the denominator is how long the traveling public has to drive past the work. That same quantity is what pushes jobs onto the night shift, as night highway construction sets out.
Where the evidence stops cooperating with the sales pitch
Cost growth from award to final showed no statistically significant difference between any of the four methods at 95 percent confidence. The means were 4.1 percent for design-bid-build, 0.9 percent for CM/GC, 2.8 percent for low-bid design-build and 4.0 percent for best-value design-build. Faster delivery did not buy worse cost control. It did not buy better cost control either.
Award growth, the gap between the engineer’s estimate and the contract award, does separate. Design-bid-build came in 9 percent below estimate on average and CM/GC 3 percent above, and CM/GC’s higher average was statistically significant against all three other methods. The report offers competition and negotiated pricing as plausible causes without asserting them, and notes something budget officers should weigh: CM/GC also produced the narrowest dispersion, a standard deviation of 6 percent. The method that comes in high comes in high predictably.
Alternative technical concepts, the mechanism that lets a proposer offer a modified approach for competitive advantage, moved the wrong way. On best-value design-build projects, cost growth was 6 percent with them and 2 percent without, a statistically significant difference that the report says needs further study, while noting those projects were also the more complex ones.
Construction schedule growth needs its median read aloud. CM/GC had the highest mean at 31 percent, driven by a handful of extreme projects, and a median of zero, the same median as design-bid-build. Low-bid design-build was the only method whose construction schedules came in shorter than planned on average. Change orders behaved as the mechanism predicts: unforeseen conditions were the largest single category overall, which is the ground uncertainty described in earthwork and grading arriving as paperwork, while plan errors and omissions were highest under design-bid-build, where the agency owns the drawings. Total change order impact ran 5.8 percent of award value for design-bid-build against 3.4 percent for CM/GC.
Contracting methods provide the environment for success, but they by no means guarantee it.
A concession is a financing decision wearing a delivery label
Public-private partnerships get listed as a fourth method and do not belong in the comparison. A design-build-finance-operate-maintain agreement bundles construction with decades of operations and maintenance, and sometimes with revenue risk. Colorado’s Central 70 keeps that last piece on the public side: federal records describe availability payments over a 30-year agreement concluding in 2052, against TIFIA-eligible project costs of $1,271 million, with the state paying the concessionaire rather than handing over toll revenue.
The Chicago Skyway shows what happens when revenue risk does change hands. The city leased the 7.8-mile elevated toll road under a 99-year concession commencing January 26, 2005, for an upfront $1.83 billion that funded a $500 million long-term reserve, a $375 million medium-term reserve and a $100 million neighborhood, human and business infrastructure fund drawn down over five years. In November 2015 a consortium of three Canadian pension funds agreed to buy the lease for $2.8 billion. Nothing in that sequence describes how well the road was built. It describes what a decade of traffic and revenue risk turned out to be worth, which is a different question from the one a delivery method answers, and closer to the ground covered in what a mile of interstate costs.
The ordinary methods have a quieter federal history. Since 1990, agencies have been able to test contracting techniques that do not follow standard Title 23 procurement rules. Design-build, CM/GC and indefinite delivery contracting were the techniques that proved out, and those were the ones codified in regulation. Progressive design-build, best value for design-bid-build, and fixed price with variable scope sit in the experimental tier today. The list is a record of what survived testing rather than a ranking, and it is the closest thing more on road construction has to a controlled experiment on how work gets bought.
The comparison that still has not been made
For all its size, the dataset thins out exactly where the argument is loudest. Duration data came from fewer projects than cost data. Only 49 of the 291 projects could supply the numbers needed to compute overall project schedule growth, and the report says outright that there are not enough data to draw substantial conclusions from them.
That leaves agencies choosing a method before they know which of the eight unavoidable risks will be the one that costs eight months, using national averages built from projects that had already finished under conditions nobody controlled. The gap is fillable, and not by another synthesis. A single state that published its own award estimates, cost certainty dates and change order causes by method, for every project in a program, would be more useful to the next agency than the entire national mean.