The challenging interpretation of instrumental variable estimates under monotonicity

Area of Science:

• Epidemiology
• Biostatistics
• Causal Inference

Background:

• Instrumental variable (IV) methods estimate local causal effects in specific subgroups.
• These local effects may not be suitable for clinical or policy decisions.
• Interpreting IV effects is complicated by non-causal instruments, common in epidemiology.

Purpose of the Study:

• To investigate the interpretation of local causal effects from IV estimates.
• To address challenges arising from non-causal instruments in IV analysis.
• To guide researchers in posing and answering questions about local effect estimates.

Main Methods:

• Developed a framework for evaluating IV estimates under monotonicity for both causal and non-causal instruments.
• Presented results to help investigators assess the subgroup, size, characteristics, and sensitivity of local effect estimates.
• Provided formal proofs to support the interpretation of IV estimates.

Main Results:

• Common approaches for interpreting IV local effects are often only valid for causal instruments.
• The validity of interpreting IV estimates as local effects depends on the instrument's causality.
• Sensitivity of effect estimates to deviations from monotonicity can be quantified.

Conclusions:

• Transparent reporting of IV results requires careful consideration of instrument causality.
• Understanding the subgroup and characteristics of local effects is crucial for decision-making.
• The presented framework enhances the utility of IV estimates in informing policy and clinical practice.