Think globally, act globally: An epidemiologist's perspective on instrumental variable estimation

Area of Science:

• Epidemiology
• Econometrics
• Biostatistics

Background:

• Instrumental variable (IV) methods are widely used in epidemiologic research.
• The common focus on local average treatment effect (LATE) over global average treatment effect (ATE) has been a source of confusion and debate across disciplines.
• Previous discussions have highlighted concerns regarding the utility of LATE for clinical practice and policy due to its focus on a subset of the population (compliers).

Purpose of the Study:

• To critically evaluate the limitations of the local average treatment effect (LATE) as an estimand in epidemiologic and public health research.
• To question the tenability of the monotonicity assumption, often required for LATE identification.
• To propose alternative methods that recenter the focus on the global average treatment effect (ATE) for greater population relevance.

Main Methods:

• Critical review and discussion of existing literature on instrumental variable (IV) methods, specifically focusing on LATE and ATE.
• Analysis of the assumptions underlying LATE, including the monotonicity condition.
• Exploration of alternative effect measures beyond LATE.

Main Results:

• The relevance of LATE as a primary effect measure in epidemiology is questioned.
• The monotonicity assumption is deemed not plausible in many common epidemiologic settings.
• The limitations of LATE hinder its applicability for informing clinical practice and policy decisions.

Conclusions:

• The exclusive focus on LATE in IV analyses may not be optimal for public health and clinical decision-making.
• Alternative approaches that estimate the global average treatment effect (ATE) are needed to provide more broadly applicable insights.
• Revisiting and refining IV methods to prioritize ATE estimation is essential for advancing epidemiologic research and its translation into practice.