missingvalue

A missing value is a data point where a value is not observed for a variable in a dataset. Missing values are commonly represented by symbols such as null, NaN, NA, or blank fields and can occur in any data type.

Missing data are categorized by the mechanism that produced them: missing completely at random (MCAR), missing at random (MAR), and missing not at random (MNAR). MCAR means missingness is unrelated to observed or unobserved data; MAR means it relates to observed data; MNAR means it relates to unobserved data. Understanding the mechanism helps inform appropriate handling and analysis strategies.

Missing values can bias analyses, reduce statistical power, and distort distributions. The impact depends on the amount of missing data, the mechanism of missingness, and the analytic method used. In some contexts, a small amount of MCAR data may have limited effect, while MNAR data can lead to serious bias if not properly addressed.

Common strategies for handling missing values include deletion (listwise or pairwise), single imputation (mean, median, or mode substitution; or a constant value), and more advanced methods such as multiple imputation, regression imputation, or k-nearest neighbors imputation. Model-based approaches include maximum likelihood and expectation-maximization; time-series data may use interpolation or forward/backward fill.

Best practice involves documenting the extent and mechanism of missingness, reporting results with and without imputation when appropriate, and using analyses that accommodate missing data or imputations. Software in R, Python, SAS, and Stata provides multiple methods for handling missing values.