Definition
Selection bias occurs when the process that determines which observations, assets, studies or results are included is related to the quantity being measured. The selected sample can then differ systematically from the target population. Correcting it requires understanding the selection mechanism, not merely increasing the selected sample size.
How it is calculated
Selection bias is illustrated when the expected outcome among selected cases S=1 differs from the target-population expectation.
| Symbol | Meaning | Unit |
|---|---|---|
| E | expectation | outcome units |
| Y | outcome | outcome units |
| S | selection indicator | binary |
Interpretation
Studying only survivors, winners or published results is like measuring a race after discarding the runners who dropped out. The remaining average may be real for the selected group but not for the population you meant to describe. Selection is part of the data-generating story.
Assumptions and conventions
-
The target population and selection rule are defined.
-
The selection mechanism can be examined or modeled for a correction.
The compared expectations have Y units; S is dimensionless.
Variants
- The target population and selection rule are defined.
Limitations and common misreadings
-
A large selected sample can still be biased.
-
Selection bias is broader than survivorship bias.
References
- James J. Heckman. (1979). Sample Selection Bias as a Specification Error. Econometrica. Sections 1–2. doi:10.2307/1912352. Source

