regressioimputation

Regressioimputation, commonly referred to as regression imputation, is a statistical method for handling missing data by filling in missing values with predictions derived from a regression model. The approach uses observed data to estimate the relationship between the variable with missing values and a set of other variables that are available. A regression model is fitted with the target variable as the dependent variable and the predictors as the independent variables. The model is then used to predict the missing entries. In single regression imputation, the predicted values serve as substitutes for the missing data. For binary or categorical targets, logistic or multinomial regression can be employed.

Key assumptions include that the data are missing at random (MAR) or missing completely at random, and that the relationships captured by the model are valid for the missing cases. Advantages of this method include its simplicity and its ability to leverage information contained in the observed relationships among variables. It is also flexible, accommodating different types of outcomes with appropriate regression models.

However, regression imputation has notable limitations. It typically underestimates the uncertainty associated with missing values because it does not incorporate prediction error, which can bias standard errors and confidence intervals. It can also distort relationships if the regression model is misspecified or if the missingness mechanism depends on unobserved factors. To address these drawbacks, many practitioners use multiple imputation based on regression models, creating several plausible imputed datasets and combining results to reflect uncertainty.

Regressioimputation often serves as a baseline method in exploratory analyses or as a component within more advanced imputation strategies.