Table 7.

Statistics on linear models predicting daily well-being from activity measures. Whereas the models provide an improvement overall, there is a range in the ability to model individuals. The P values are for permutation tests, checking whether user lift is greater than 0, that is, whether models are significantly more accurate than always predicting each individual to be at their most frequent state.

Problem (model)	Well-being measure	Average user lift	Minimum user lift	Maximum user lift	P value
Good or bad day (penalized logistic regression)	Mood (Prediction error)	5.44%	−21.74%	35.00%	.001
	Energy (Prediction error)	4.92%	−22.73%	39.39%	.008
Daily average (linear regression with elastic net)	Mood (RMSEa)	0.026	−0.232	0.48	.08
	Energy (RMSE)	0.048	−0.169	0.575	.01

^aRMSE: root-mean-square error.