Parametric cost estimation builds a statistical model of cost as a function of one or more cost drivers, measurable project characteristics that historically predict cost well. The simplest form is a linear regression:
where are the cost drivers (size, throughput, weight, square footage, headcount, complexity score) and the ‘s are fitted from a historical dataset.
The classic software example is COCOMO (Constructive Cost Model): cost from lines of code, team experience, schedule pressure, and similar drivers, with coefficients fitted from decades of historical project data. Chemical plants use throughput, pressure, and process type. Buildings: floor area, number of storeys, building class.
Parametric models sit between unit cost estimation (one driver, simple ratio) and definitive estimates (bottom-up bills of materials). Good for budgetary-level accuracy when:
- You have enough historical data to fit a reasonable model.
- The new project falls within the range of the historical projects on each driver. Extrapolating outside the fitted range is risky.
- The drivers actually capture the variation in cost. Adding more drivers helps only if they explain variance in the historical data.
Accuracy is middle of the road: better than pure unit-cost scaling because it accounts for several factors, worse than detailed bottom-up estimates because it stays aggregate-level. How good it is depends on how well the historical fit predicts new projects.
Power-sizing model is a one-driver parametric model with a non-linear form (cost scales as size to a fractional power), the workhorse for engineering economies of scale.