Home Statistical Dictionary Attrition (Study)
Attrition (Study)
The progressive loss of participants from a longitudinal study or panel over successive waves of data collection, a form of nonresponse specific to studies that follow the same units over time.
In Plain English
Attrition is what nonresponse looks like specifically in a study that tracks the same people over time, a longitudinal survey, a clinical trial, a panel study. People who were in the study at the start drop out along the way, they move, lose interest, become impossible to reach, or in a clinical trial, sometimes stop participating because of side effects or because the treatment isn't working for them. Like ordinary nonresponse, attrition only becomes a serious bias problem if the people who drop out are systematically different, on the outcome being studied, from the people who stay in, but because attrition often has a specific cause tied to the study itself (side effects, discouragement, relocation for a new job), it's frequently more suspect than one-time survey nonresponse.
Definition
Attrition is the cumulative loss of participants from a longitudinal or panel study across successive waves of data collection, distinct from cross-sectional nonresponse in that it specifically concerns participants who were successfully enrolled and observed at an earlier wave but fail to be observed at a later one. Attrition threatens the validity of longitudinal analysis when it is differential, that is, correlated with the outcome or treatment being studied, in a clinical trial, participants who experience adverse side effects may be more likely to drop out, and analyzing only those who remain (a form of survivorship bias specific to longitudinal designs) can bias treatment-effect estimates; common remedies include intent-to-treat analysis (which analyzes participants according to their original group assignment regardless of attrition, preserving randomization's protection against bias), multiple imputation of missing follow-up outcomes under a stated missingness assumption, and inverse-probability-of-attrition weighting, which reweights remaining participants to compensate for the observed characteristics of those who dropped out.
Properties
- Attrition only seriously biases a study's conclusions when it is differential, correlated with the outcome or with treatment assignment, random attrition unrelated to the outcome reduces statistical power (a smaller effective sample) but does not necessarily bias point estimates, which is why distinguishing differential from non-differential attrition is a standard step in evaluating a longitudinal study's validity.
- In randomized clinical trials, differential attrition tied to treatment side effects or perceived lack of efficacy is a well-known and serious threat to validity, intent-to-treat analysis is the standard defense, since it analyzes all randomized participants according to their assigned group regardless of whether they completed the study, preserving the unbiased comparison that randomization was designed to guarantee.
- Attrition rates and patterns are typically reported explicitly in longitudinal study documentation (often via a CONSORT-style flow diagram in clinical trials), and researchers routinely compare the observed characteristics of those who dropped out against those who remained as a diagnostic check for whether attrition is likely to be differential.
At a Glance