Frank Martin

A local electric utility company that supplies many thousands of households with electricity is revising its billing scheme. One of the key factors in the new scheme will be the "peak electrical usage" of a household--the highest rate of electricity consumption during the billing period. Accurately measuring peak usage, however, requires constant monitoring of a household's electricity usage and is expensive; the company can only measure the peak usage on a sample of area households. Many things are already known about each household--its monthly electric bill, the approximate age of the dwelling, etc. The utility wants to use the relationship between these quantities and peak usage to: a) find a useful predictive relationship between these known quantities and the peak usage, and b) choose which households to sample in order to optimize the process of estimating the peak usage. As member of the Statistical Consulting Service, I spend much of my time working on problems such as the one outlined above--problems which range from race discrimination (is there statistical evidence that discrimination is occurring?) to assessing the specific causes of lake acidification.