Simple graphical rules for assessing selection bias in general-population and selected-sample treatment effects

Area of Science:

• Causal inference
• Epidemiology
• Biostatistics

Background:

• Selection bias can compromise the validity of analyses using selected samples.
• Understanding bias is crucial for accurate causal effect estimation in observational studies.

Purpose of the Study:

• To develop simple graphical rules for assessing selection bias in selected-sample analyses.
• To determine if covariate adjustment can mitigate selection bias.
• To introduce and provide graphical rules for the 'net treatment difference' estimand.

Main Methods:

• Utilizing single-world intervention graphs to represent causal structures.
• Developing graphical criteria to identify and assess different types of selection bias.
• Decomposing bias into 'internal bias' and 'net-external bias'.

Main Results:

• Provided clear graphical rules to check for unbiasedness of selected-sample analyses relative to general population and selected-sample ATE.
• Established rules for covariate adjustment to eliminate selection bias.
• Introduced graphical rules for the 'net treatment difference' when treatment affects selection.

Conclusions:

• Graphical rules offer a straightforward method for evaluating selection bias in selected samples.
• The decomposition of bias provides conceptual clarity on bias mechanisms.
• This framework enhances the understanding and handling of selection bias in causal research.