Using built environment characteristics to predict walking for exercise

Table 4 Holdout validation and replication of models using the built environment to predict walking for exercise

	Logistic model of walking or not		Linear model of walking time
	Training set, C-statistic (95% CI)	Validation set, C-statistic (95% CI)	Training set, variation explained	Validation set, variation explained
Models from Table 3, training set β estimates	0.58 (0.55, 0.62)	0.46 (0.41, 0.51)	4.1 %	0.0 %
Restricted to King County	0.61 (0.57, 0.66)	0.45 (0.38, 0.52)	5.6 %	0.2 %
WBC components, training set β estimates	0.56 (0.53, 0.60	0.46 (0.41, 0.51)	1.9 %	0.3 %
Restricted to King County	0.58 (0.53, 0.62)	0.44 (0.38, 0.51)	3.2 %	0.3 %
	Training and validation sets, C-statistic (95% CI)		Training and validation sets, variation explained
WBC components, WBC β estimates	0.50 (0.48, 0.53)		0.1 %
King County only	0.51 (0.47, 0.55)		0.1 %

Notes: All models were run for the entire HVH population and restricted to King County residents; C-statistic indicates area under the receiver operating characteristic curve; CI indicates confidence interval; WBC indicates Walkable and Bikeable Communities Study

ISSN: 1476-072X