An alternative approach called direct sub-metering requires putting an energy meter on every appliance over an extended period of time. While effective, it is a much more expensive undertaking that requires recruitment and field personnel as well as incentives for households to offset the invasiveness. A third approach simply takes the engineering characteristics of the appliance and extrapolates consumption based on an assumed usage pattern, but this assumed usage may not actually be representative of the behavioral range of the population. Because it takes actual behavior data into account, and is not prohibitively expensive, CDA splits the difference between the two approaches and offers a cheap yet robust avenue for modeling end-use consumption.