Table 2.

Summary of past efforts at calibration and validation

Reference	Calibration & Validation	Scale & Resolution	Forecasting interval (predictive error)
(Rai and Robinson, 2015)	Yes	Austin, TX households	Q3 2013–Q4 2014 (5.5%)
(Zhang et al., 2016)	Yes	San Diego County, CA households	05/11–04/13 (∼10%a)
(Yu et al., 2018)	Yes	All U.S. census tracts	N/A
(Williams et al., 2020)	No	Germany, Japan, Arizona, California, Massachusetts	N/Ab
(Müller and Trutnevyte, 2020)	Yes	Switzerland, 143 districts	1 year <12%c

^aThe authors do not provide this figure, but the in the final prediction time period, the average predicted value appears around 125, whereas the true value falls nearer to 139.

^bThe authors do present predictive results but do not assess their accuracy.

^cEstimated based on the best performing model presented by the authors in Figure 9 on page 10. The authors also present 5-year forecasts, but do not give a percentage error.