Primary Submission Category: Instrumental Variables
A Modified Instrumental Variable Approach for Modeling Engagement in Mobile Health
Authors: Alexis Fleming, Andrew Spieker,
Presenting Author: Alexis Fleming*
Engagement with mobile health interventions has gained increasing attention as mobile devices become more widely accessible, with identifying thresholds for effective engagement being a primary focus. However, in these settings, the exclusion restriction assumption is unreasonable. Previous methodological advances have addressed this challenge by modifying instrumental variable (IV) approaches through sensitivity analyses to understand how engagement influences improvements in outcomes when the exclusion restriction cannot be reasonably assumed. These methods, however, have largely been limited to settings with two treatment arms and a continuous measure of engagement.
We propose a modified instrumental variable (MIV) approach that builds on this prior work and broadens its application to real-world interventions. Our method allows for multiple treatment groups and supports alternative distributions of the engagement variable, thereby expanding the range of settings in which engagement effects can be assessed. Under correct specification of a sensitivity parameter, the class of local average treatment effects (LATEs) of interest becomes identifiable.
Estimation proceeds through two-stage least squares to recover a weighted average of the LATEs over a specified range of the sensitivity parameter. We evaluate the proposed approach using simulation studies and demonstrate its ability to recover LATEs across varying combinations of treatment and engagement variable types.