Using synthetic data to explore the usefulness of prism's parameters at inferring causes of changes in normalized annual consumption

PRISM's linear regression of energy per day vs heating degree days (HDD) per day is widely used for the weather normalization of energy use in buildings. While Normalized Annual Consumption (NAC) is PRISM's most robust result, changes in the regression parameters intercept, slope, and reference temperature are often cited to infer the physical changes in the buildings under consideration. Using DOE-2 simulations of energy use, we show that such use of parameters other than NAC is inadvisable in a mild climate, due to significant nonlinearity in the relation between energy use and heating degree days. PRISM's unique feature is its adjustment of the reference temperature for HDD calculations to maximize the linearity of energy/day vs HDD/day, so that each house or group of houses under examination has its own reference temperature. This method of reference temperature adjustment, in a mild climate such as San Francisco's, results in significant misstatement of the actual reference temperature, base-level consumption, and heat part parameters. The distortions are not consistently biased. Further, the standard errors of the parameter estimates sometimes provide insufficient warning of the problems attendant on the use of changes in these parameters to explain physical changes in the building.