Attrition
The loss of outcome data for some experimental subjects (missing outcomes). Attrition is benign when outcomes are missing independently of potential outcomes (MIPO) — randomization is preserved among observed units and group-mean comparisons stay unbiased. It becomes a threat when attrition is systematically related to potential outcomes, because the observed treatment and control groups are no longer comparable even though assignment was random — a selection problem that reopens the Fundamental-Problem-of-Causal-Inference. Remedies: inverse-probability reweighting when attrition is ignorable given covariates (MIPO conditional on , an Ignorability-type assumption for the missingness), worst-case/extreme-value bounds (and trimming bounds) that bracket the effect without such assumptions, and double sampling — gathering outcomes from a random subset of the missing.
Relied on by
RCTs and any study with incomplete outcome measurement; a core internal-validity threat.