Household, sociodemographic, building and land cover factors affecting residential summer electricity consumption: A systematic statistical study in Phoenix, AZ

Abstract

Understanding determinants of residential electricity consumption is crucial for urban sustainability efforts for planners and policy makers to develop targeted strategies to lower energy use, reduce greenhouse gas emissions, and to increase community resilience. This study presents a systematic approach to build an interpretable multivariate linear model, addressing challenges like outlier detection, multicollinearity, non-normality, and heteroscedasticity. Using 2019 summer residential electricity data for 426 census tracts in Phoenix and 30 variables, the approach involves (1) addressing multicollinearity and regression outliers through Variance Inflation Factor and studentized residual analysis, (2) comparing an automatic variable selection method with Ridge, Lasso, and Elastic Net regression, (3) evaluating the final model, and (4) interpreting variable effects. Critical findings reveal multicollinearity in land cover and racial variables, while 21 census tracts on the urban periphery exhibit outliers with unique features. Variable selection demonstrates the significance of household and building information in influencing residential electricity consumption. Household variables alone account for 84 % of electricity usage variation. Incorporating building information and land cover variables reduces errors by 35 % and 26 % respectively, emphasizing the significance of including household characteristics as predictors or control variables when modeling electricity consumption. A final model with 93 % explanatory power enables precise predictions.