Data Collection, Forecasting, and Analysis
The data base
Credible forecasts start with defensible inputs: traffic counts expanded to AADT with seasonal, axle, and day-of-week factors; screenline and cordon counts; household travel surveys and the NHTS; transit APC ridership; Census/ACS demographics; land use, employment, and building-permit data; and increasingly probe, GPS, and mobile-device data. Sampling error, expansion factors, and nonresponse bias are fair game.
Modeling sequence
Know the four-step model - trip generation, trip distribution (gravity model), mode choice (logit), and assignment (user equilibrium) - and its alternatives: activity-based and tour-based models, microsimulation, and sketch-planning tools. Distinguish calibration (adjusting parameters so the model matches local observations) from validation (testing the calibrated model against an independent data set it was not fit to). Then know scenario and sensitivity testing, elasticity methods, and before-and-after evaluation.
Traps
Watch for unit confusion (person trips versus vehicle trips; daily versus peak hour), treating one short count as an AADT, claiming that a more complex model is automatically better, and forecasts that ignore induced demand, land use feedback, or demographic change.
Rule of thumb - model acceptability
Use the GEH statistic, GEH = √[2(M−C)²/(M+C)], comparing modeled (M) and counted (C) flows. A common screening criterion is GEH < 5 for individual link volumes and GEH < 1 for screenline totals; a link failing GEH should be rechecked before the model is accepted.
Worked example - elasticity
Parking price rises 10% and the own-price elasticity of vehicle demand is −0.2. Percentage change in volume ≈ −0.2 × 10% = −2%, so 50,000 daily trips become about 49,000.