| ORNL Aggregate Weather-Adjusted Evaluation Model Perhaps the most notable feature of the natural gas or electric billing data for the low-income households we examined is its tendency to be extremely noisy. Behavioral changes, household vacancies, equipment breakdowns, and service shut-offs due to nonpayment frequently overwhelm weather-related correlations with fuel usage. The amount of noise in the billing data is more typical of the social sciences than of problems from physics or engineering. Because of the high level of noise in the billing data, we have found that using the PRISM, which relies on linear models of weather adjustment, leads to high model failure rates at the individual household level. Excluding large numbers of homes from the statistical analysis due to model failures is likely to introduce sample bias, making the measurement of representative Program impacts difficult because so many homes are eliminated. Circumstances that cause energy consumption to vary in ways that PRISM's weather-adjustment model is not designed to handle occur frequently in the low-income households. ORNL developed the aggregate weather-adjusted evaluation model to overcome this problem by performing weather-normalization on large groups of houses. The ORNL aggregate model makes weather-normalization possible for a larger, more complete, and more representative sample of households. As a result, the aggregate model reduces sample bias by making it possible to evaluate savings in a more inclusive group of Program homes. When nearly all homes, both weatherized and control, are included in an analysis, the overall Program effect, rather than individual household savings, becomes the focus of program assessment. The ORNL aggregate weather-adjusted evaluation model was used to analyze billing records for 895 gas heated houses and 2,379 electrically heated or cooled houses. For each group, about one-third was weatherized houses and two-thirds were control houses. The analysis showed that weatherized houses had a normalized annual savings of about 8% with a 95% confidence interval of 2% to 14 %. For the electric billing data, normalized annual savings of weatherized houses were about 5.6% with a 95% confidence interval from 2% to 10%. The results also indicated that for weatherized houses, cooling energy consumption increased while heating energy consumption decreased to give the net savings reported above. Contact:
Lance
McCold |