Neyman's bias

From RationalWiki
Jump to navigation Jump to search
Cogito ergo sum
Logic and rhetoric
Icon logic.svg
Key articles
General logic
Bad logic
A late look at those exposed (or affected) early will miss fatal and other short episodes, plus mild or ‘silent’ cases and cases in which evidence of exposure disappears with disease onset.
—David Sackett[1]

Neyman's bias (also known as prevalence-incidence bias) is a form of selection bias that ensues due to the timing of when cases are incorporated into a study.[1] Case-control studies tend to be the most vulnerable to this bias.[1] Essentially, the bias comes about whenever sick or healthy patients are excluded from a study, hence the bias occurs in two forms:

  • Patients who have died (or are in severe condition) are excluded, consequently making conditions seem less severe.
  • Patients who have recovered are excluded, consequently making conditions seem more severe.

The more time between exposure and investigation shall subsequently affect the chances of death or recovery, which can aggravate this bias as more cases are excluded from the study being conducted.[1] When time goes by, patients will die. Conversely, other patients will recover during this same time interval. The people conducting the study may never know which groups of people are being excluded, thus it becomes impossible to adjust the study in order to mitigate bias; accordingly, there shall be either overestimated or underestimated results.


If a case-control study investigates the flu (caused by influenza viruses), but only incorporates cases who have been admitted to a hospital, those who died preceding admission won't be included. Because the sample won't include those fatal cases, the study will be inaccurate.


For studies to be accurate, careful selection of study samples is imperative for an adequate grasp of certain diseases and their causes. Incident cases (newer cases) are favored over prevalent cases (pre-existing cases, which tend to be sicker with more progressed diseases), therefore, it's best to use incident cases instead of prevalent cases.[1] This is why Neyman's bias is also called prevalence-incidence bias.

See also[edit]