EcoSta 2021: Start Registration
View Submission - EcoSta2021
A0444
Title: Bounding local average treatment effects under exclusion-restriction violations in mobile health interventions Authors:  Andrew Spieker - Vanderbilt University Medical Center (United States) [presenting]
Abstract: In the modern era, mobile health interventions are of increasing interest. In the context of randomized trials based on text-message interventions, emphasis has been placed on providing patients with opportunities for engagement and responses to the text-messages. One goal in such settings is to evaluate the effect of engagement with the intervention on the outcome of interest. As engagement is a post-randomization exposure that can be thought of as analogous to compliance in a randomized drug study, instrumental variable approaches appear at first to be a reasonable way to address this goal. However, in pragmatic text-message based interventions, violations to the exclusion-restriction assumption are almost certainly to be expected. In this sense, the goal is truly to tease apart the effect of the intervention itself and the effect of engagement with the intervention. We develop a sensitivity analysis approach to place sharp bounds on the local average treatment effect under different levels of engagement. We further illustrate this approach to accommodate a wide range of scenarios and various levels of departures from the exclusion-restriction assumption.