Hybrid experimental designs (HEDs) combine interventions that adapt on different timescales—slow (e.g., every few weeks) and fast (e.g., every few hours). A common HED integrates the sequential, multiple assignment, randomized trial (SMART) with the micro-randomized trial (MRT), enabling researchers to study how digital just-in-time components interact with human-delivered support. Until recently, methods for analyzing these designs treated one component as a moderator of the other, which cannot capture true synergistic effects. In a new paper with Mengbing Li and Inbal Nahum-Shani, we formalize causal estimands and propose a data-analytic method for hybrid SMART-MRTs. See the paper: Evaluating time-varying treatment effects in hybrid SMART-MRT designs.

Micro-randomized trials (MRTs) are designed to evaluate the effectiveness of mobile health (mHealth) interventions delivered via smartphones. In practice, the assumptions required for MRTs are often difficult to satisfy: randomization probabilities can be uncertain, observations are frequently incomplete, and prespecifying features from high-dimensional contexts for linear working models is also challenging.

To address these issues, the doubly robust weighted centered least squares (DR-WCLS) framework provides a flexible procedure for variable selection and inference. The methods incorporate supervised learning algorithms and enable valid inference on time-varying causal effects in longitudinal settings. The DR_WCLS_LASSO R package implements post-selection inference with LASSO in this framework. A detailed tutorial is available at whd-lab.github.io/DR_WCLS_LASSO.