Pacific D
Inference Without Randomization or Ignorability: A Stability-controlled Quasi-experiment on the Prevention of Tuberculosis (307919)
Chad Hazlett, UCLA Departments of Statistics and Political ScienceWerner Maokola, Kilimanjaro Christian Medical University College
*David Amichai Wulf, UCLA Department of Statistics
Keywords: Causal Inference; Non-randomized Experiments; Observational Studies; Clinical Trials
Learning the effect of a new treatment through a randomized trial is often problematic due to ethical concerns, administrative constraints, or because those who are eligible for and consent to participation in a randomized trial may differ widely from those who take the treatment once generally available. We discuss a recently proposed identification strategy, SCQE, that replaces the need for a randomized control group with an assumption on how the average non-treatment potential outcome for successive cohorts vary over time. This single assumption - without ignorability - allows identification of the average treatment effect on the treated. Where this assumption is uncertain, a sensitivity analysis may yield informative results. We apply SCQE to study the effectiveness of isoniazid preventive therapy (IPT) in reducing incidence of tuberculosis among HIV positive patients in Tanzania. We also provide suggestions for standard error calculation. The SCQE has broad applicability and will sometimes lead to definitive claims about effectiveness. In this case, it usefully aids in protecting against over-confidence in claims that IPT was effective.