The Estimate Gets Honest When the Schedule Gets Real

A busway in Hartford was supposed to cost $88 million. It opened fifteen years later at $546 million. Nobody stole the money. The early number was never real - it was a guess wearing a dollar sign.

The federal government does something most owners never do: it audits its own predictions. The FTA's 2020 Predicted-versus-Actual study covers 29 transit projects - rail, BRT, streetcar - that opened between 2007 and 2015, with the cost estimate recorded at every gate. I pulled the table and ran the numbers.

At the early gate, before scope and schedule were pinned down, the median project missed its estimate by 18 percent, and two projects in three blew through the early number by more than 10. Hartford's CTfastrak ran +520 percent. Pittsburgh's North Shore Connector, +367. San Diego County's Sprinter, +123. The Silver Line to Dulles doubled.

Same projects. Same owners. Different moment: at final funding agreement - scope locked, schedule built - the median miss was about one percent, and only two of the 29 ran more than ten percent over.

FTA's own cause table says why. On the projects that missed early, schedule change shows up ten times - the most-cited driver alongside scope. A schedule that slips drags cost with it: inflation on every late month of materials and labor, professional services that don't go home, overhead that keeps billing. The early estimates didn't fail as arithmetic. They failed as bets on schedules that didn't exist yet.

And these bets sit open a long time. The median project here took seven years from entering engineering to opening day. The worst took fifteen. An early estimate is a claim about a market a decade away, priced before anyone has sequenced the work.

The takeaway isn't that estimators are bad. It's the opposite - the same people nailed the number once the schedule underneath it was real. An estimate is only as good as the logic it stands on. If the schedule isn't built yet, the number is a placeholder. Treat it like one.

Methodology

Data: FTA, Predicted Versus Actual Impacts of Capital Investment Grants Projects (2020), Table 2.1-1 - 29 CIG projects opened for revenue service 2007-2015, with capital cost estimates at entry into Preliminary Engineering/Project Development, at Full Funding Grant Agreement, and actual cost at completion, all in year-of-expenditure dollars (so early-vs-actual comparisons aren't an inflation artifact). Error = (actual - estimate) / estimate. Our recount from the published table finds 19 projects more than 10 percent over their PE-stage estimate; FTA's text states 17 - the difference is two borderline cases and doesn't change the finding. Our FFGA-stage results reproduce FTA's exactly: two projects (Portland's WES commuter rail, Norfolk's Tide) more than 10 percent over; two others finished more than 10 percent under. Development duration = opening year minus PE-entry year. Schedule-change driver counts from FTA Table 2.2-1. Transit projects, federal program - patterns generalize to large capital work, but thresholds are program-specific.