Estimating Effects of Time-Varying Special Education Regimes Using Targeted Maximum Likelihood
Abstract
Longitudinal treatments are common in social science research but are often reduced to ever-versus-never exposure indicators that obscure their time-varying nature. This study shows how causal questions can be clarified by defining treatments through explicit intervention regimes rather than collapsed summaries. We introduce this framework by linking the longitudinal setting to single-time-point causal inference and extending it to time-varying exposures, sequential identification assumptions, and estimation using targeted maximum likelihood estimation (TMLE). We also address practical challenges, including missing data and positivity violations. Using data from the Early Childhood Longitudinal Study, Kindergarten Class of 1998-1999 (ECLS-K), we examine the effect of special education participation on 5th-grade mathematics achievement. Comparing static and dynamic regimes, we find meaningful differences: always-treated versus never-treated contrasts yield the largest negative estimates, while rule-based dynamic regimes produce substantially smaller effects. These results highlight the importance of well-defined interventions for interpreting longitudinal treatment effects.
Related
- Estimating the marginal likelihood with Integrated nested Laplace approximation (INLA)
- Neural ODE enhanced linear mixed effect models for estimating complex association patterns of time-varying covariates with the marker trajectory
- Neural ODE enhanced linear mixed effect models for estimating complex association patterns of time-varying covariates with the marker trajectory Supplementary Materials
- Estimating Treatment Effects in Complex Disease Dynamics: Immortal Time, Varying Hazards, and Multiple Time Scales
- Multifidelity Computer Model Emulation Via Diffusion Model Steering and Targeted Maximum Likelihood
BibTeX
@misc{keller2026estimating,
title = {{Estimating Effects of Time-Varying Special Education Regimes Using Targeted Maximum Likelihood}},
author = {Bryan Keller and Lindsey Liu and Yingqi Huan},
howpublished = {PsyArXiv},
year = {2026},
month = sep,
doi = {10.31234/osf.io/fs7tw_v2},
url = {https://osf.io/preprints/psyarxiv/fs7tw_v2/},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Estimating Effects of Time-Varying Special Education Regimes Using Targeted Maximum Likelihood
AU - Bryan Keller
AU - Lindsey Liu
AU - Yingqi Huan
PY - 2026
DA - 2026/09/17
PB - PsyArXiv
DO - 10.31234/osf.io/fs7tw_v2
UR - https://osf.io/preprints/psyarxiv/fs7tw_v2/
AB - Longitudinal treatments are common in social science research but are often reduced to ever-versus-never exposure indicators that obscure their time-varying nature. This study shows how causal questions can be clarified by defining treatments through explicit intervention regimes rather than collapsed summaries. We introduce this framework by linking the longitudinal setting to single-time-point causal inference and extending it to time-varying exposures, sequential identification assumptions, and estimation using targeted maximum likelihood estimation (TMLE). We also address practical challenges, including missing data and positivity violations. Using data from the Early Childhood Longitudinal Study, Kindergarten Class of 1998-1999 (ECLS-K), we examine the effect of special education participation on 5th-grade mathematics achievement. Comparing static and dynamic regimes, we find meaningful differences: always-treated versus never-treated contrasts yield the largest negative estimates, while rule-based dynamic regimes produce substantially smaller effects. These results highlight the importance of well-defined interventions for interpreting longitudinal treatment effects.
ER -