OBJECTIVES: Most approaches for analyzing treatments with staggered timing do not allow a treatment’s effect to be isolated from effects of confounders associated with treatment timing (ie, early vs late adopter) and outcome. Two-way fixed effects (TWFE) models produce biased or difficult to interpret estimates when effects vary with timing or duration. A potential solution involves using difference-in-differences with inverse probability of treatment weights (IPTW) to adjust for confounding across cohorts, but IPTWs lead to biased estimates in cross-sectional evaluations comparing multiple treatments. An alternative , vector-based kernel weighting (VBKW), produces estimates that are less biased and more efficient than IPTW in cross-sectional evaluations. A natural extension of VBKW is to longitudinal studies in which treatment groups are defined by time of treatment receipt, but the degree to which it reduces bias and improves efficiency over other estimators in longitudinal applications has not yet been explored.

METHODS: Using simulations, we compared VBKW and TWFE-based estimates' bias. We conducted simulations with 300 observations over 20 time periods (500 replications). We specified a static true treatment effect and induced heterogeneity through an observable factor associated with timing and outcomes. We conducted simulations with mild and moderate heterogeneity (confounder coefficient=0.03,0.1) and in scenarios with an even split of observations across cohorts (never, early[time2], late[time10]) or relatively few never treated observations.

RESULTS: In all simulations, VBKW produced estimates with lower absolute mean relative bias (AMRB) than TWFE. Reductions in bias with VBKW were greater when the “never treated” group was small (AMRB, mild heterogeneity:VBKW=0.21%,TWFE=7.42%; moderate heterogeneity:VBKW=0.70%,TWFE=24.74%).

CONCLUSIONS: In analyses involving staggered treatment timing, VBKW produces less biased estimates than TWFE. Future comparisons will include IPTWs, optimal weighting methods, and a broader range of analytic scenarios. Identifying best practices to analyzing staggered treatments is critical to improving the rigor of evidence used to support complex interventions that cannot be randomized.