Table 3.
Mechanisms of Missing Data in Clinical Research
| Mechanism | Definition | Example | Prevalence |
|---|---|---|---|
| Missing Completely at Random (MCAR) | If the likelihood of being missing is not related to either the value of the missing variable or to the values of any other variables in the data set. | A set of samples are “lost” due to lab error or an instrument is wrongly calibrated for a day on which random sample of subjects were measured. | Almost never occurs. |
| Missing at Random (MAR) | If the likelihood of missing data can be completely explained by other variables in the analysis. | The probability of missing data on ADL can be explained by cognition, comorbidity, and living arrangements. | Other data can sometimes provide a good prediction, but missingness is rarely completely explained. |
| Missing Not at Random (MNAR) | If missing values are not randomly distributed across participants, and the probability of being missing cannot be predicted from the other variables. | The probability of missing data on CESD is related to cognitive status, which was never measured. | Most missing data. |