Recall Bias

Recall bias is differential measurement error in self-reported history. Memory is reconstructive, and people who have experienced an outcome search harder for an explanation and report candidate exposures more completely than those who have not. Where misclassification is non-differential, affecting both groups equally, the usual consequence for a binary exposure is dilution towards the null. Where it depends on outcome status the estimate can move either way, so nobody can argue the true effect must be at least as large as the observed one.

The textbook setting is the case-control study of congenital malformation, where mothers of affected infants recall pregnancy medication and illness far more thoroughly than mothers of healthy infants, generating associations prospective records fail to reproduce. Checked against doubly labelled water, self-reported energy intake is under-reported substantially, and more so at higher body weight, so the error tracks the very characteristic under study.

The practical consequence is a preference for exposure data recorded before the outcome was known: dispensing records, prospective diaries, laboratory measurements. It is the strongest argument for prospective cohorts over case-control designs when exposure history is the hard part, and inside a trial it argues for capturing symptoms at scheduled visits rather than at the end.

Where it bites hardest here is user-reported peptide experience. In forum threads and vendor surveys, someone who felt a benefit recalls the compound and the timing confidently while someone who felt nothing forgets or never posts, and the dataset can appear to show a dose-response that is entirely a memory artefact. Retrospective symptom scoring collected after participants have guessed their allocation carries the same defect inside a trial.