2025 Orange County Water Demand Projection Model
There are additional expectations for model coefficients affiliated with the explanatory variables that helped in model development. Estimated coefficients needed to be rational (both in signs and magnitudes), aligning with Hazen’s experience and yielding as high of an explanatory power as practically possible. Orange County water use varies widely by geography (agency) and time. A panel regression modeling approach is best suited to explicitly account for variables that have historically varied highly across geography but not time (for example, median income, persons per household, and housing density). The panel approach enhances forecast accuracy for each agency by simultaneously fitting consumption data from all member agencies. The larger sample size enables more robust fits for agencies with data gaps. All agencies still receive their own unique statistical equations and outputs, although some coefficients may apply to all (or be the same across) agencies. Because consumption data was not consistently available back to 2010 for all member agencies, the dataset represents what is called an “unbalanced panel” (not all retail agencies have the same number of observations for the same time periods), requiring these special techniques to account for missing information and potential cross- sectional biases. 3.2 Model Development The development of econometric models is an iterative process as outlined in Section 3.2.1. Model fitting results in a set of explanatory variables used to forecast water use. In Section 3.3 , model fits and performance are organized by water-use sector.
3.2.1
Model Fitting
Table 3-1 outlines the model fitting process.
Table 3-1: Iterative Process for Developing Econometric Models
Model Fitting Procedure
Description
Conduct necessary pre-processing calculations prior to model fitting: Geographical processing of driver units. Calculate per-unit use. Calculate natural logarithms of per-unit use and appropriate explanatory variables. Calculate departures from normal conditions for appropriate explanatory variables (i.e., economic trend and weather). Calculate any index, “dummy”, or interacted parameters (e.g., seasonal cycle, geography, drought severity). Smooth monthly and bimonthly data to adjust for irregular billing cycles. Use statistical estimation software (e.g., R, SAS, EViews) to fit linear regression equations to per unit use with the initially selected explanatory variables. Check measures of fit (such as R 2 ) and coefficient values for reasonable magnitude, direction/sign, and significance.
Pre-process model input data
Fit regression models for each sector
Examine coefficient estimates and measure of fit
3-4
Appendix G - 41
Powered by FlippingBook