Table 1 Missing data mechanisms and their examples
Missing data mechanism | Description | Example |
|---|---|---|
Missing completely at random (MCAR) | The probability that values are missing is unrelated to either the specific missing values that should have been obtained or the set of observed values. | Missing data due to equipment failure. |
Missing at random (MAR) | The probability that values are missing depends on the set of observed values but is further unrelated to the specific missing values that should have been obtained. | Missing data of blood glucose can be said to be MAR given age, if younger patients are less (or more) likely to have their blood glucose measured than older patients. |
Missing not at random (MNAR) | The probability that values are missing is related to the specific missing values that should have been obtained, in addition to the ones actually obtained. | If there are data that are unobtained but can influence the missingness of blood glucose, missing data of blood glucose cannot be said to be MAR but said to be MNAR. |