|
Abstract:
|
We introduce the study design of a Stanford led project using Apple watches, in which the goal is to promote physical activity by pushing sequences of notifications on the Apple Watch. We also identify issues of the analysis of adaptive sequential decision making that arise from this study that can be generalized to other mHealth studies in the area. In the Apple Watch study, a focal issue is whether the watch provides useful covariates and thus being able to test whether the variables obtained by wearable devices are truly informative plays an important role in decision-making. The validity of hypothesis testing for the model coefficients however, is predicated on the correct specification of either the policy or value models. We propose a robust test statistic that is valid when either the model of the algorithm or the auxiliary model for the variable collected from the wearable is correctly specified.
|
