# Missing data imputation methods: an overview

Learn how missing data imputation methods fill gaps in records, from mean and regression imputation to hot deck and multiple imputation, and when to delete.

Missing data imputation methods are statistical techniques that replace a gap in a record with an estimated value, so a dataset can be analyzed or compared without dropping the record entirely. This overview explains the main families of imputation, how each one works, what bias it can introduce, and how the choice of method depends on why the data is missing. It also covers deletion versus imputation, the role of comparable records and donor records, and the basic checks that show whether a filled value is sound. 

## Simple single-value imputation methods 

Mean imputation replaces a gap with the average of the observed values in that column, while median imputation uses the middle value and suits skewed data. Both are fast and easy to explain, yet they shrink variation and can distort relationships between columns, which is a form of imputation bias. These statistical methods work as a first pass when a column has few gaps and you only need a rough estimate. A filled value from either method is a placeholder, not a measurement. 

## Imputation from comparable records 

Hot deck imputation fills a gap with a value taken from a donor record that resembles the incomplete one, and nearest neighbor imputation does the same by choosing the closest record in a defined distance. Regression imputation instead fits a model on complete cases and predicts the missing value from the other columns. All three rely on comparable records being genuinely alike on the factors that matter, so the match rules need care. Imputing missing values this way preserves more structure than a single mean, but a poor match adds error. 

## Deletion versus imputation and uncertainty 

Deletion versus imputation is the first decision, because removing incomplete records is simple but can throw away information and skew the sample when gaps are not random. Multiple imputation answers the uncertainty problem by producing several filled datasets and pooling the results, so the spread between them reflects doubt about the estimate. Handling gaps in listing data or incomplete specifications often needs this wider view, since a filled field can change a shortlist comparison. Matching a record to comparable listings without recording the uncertainty can hide a weak fill. 

## Questions about missing data imputation 

### What does imputing missing data mean? 

Imputing missing data means estimating a value for a gap and placing it in the record so the dataset is complete. Imputation is not recovery of the true value, so the estimate carries some error. The method you pick decides how much structure the filled value keeps and how much bias it adds. 

### How do I calculate or estimate missing values? 

You estimate a missing value by choosing a rule that uses the observed information around it, such as a column average, a matched donor record or a regression prediction. The calculation itself is simple once the rule is fixed, and the harder part is checking that the rule fits the way the data went missing. Recording which method produced each value keeps the estimate auditable. 

### Which imputation method suits incomplete listing data? 

The right method for incomplete listing data depends on which fields are empty and how the gaps arose. A missing specification may be filled from comparable records with hot deck or nearest neighbor imputation, while a numeric field with few gaps may take a median. Troubleshooting incomplete listing data starts with naming the pattern of gaps before any value is estimated. 

### What are estimation methods for incomplete data in a survey? 

Estimation methods for incomplete data in a survey cover deletion, single-value imputation and multiple imputation, each with trade-offs. A survey with a large share of gaps benefits from multiple imputation, because it shows the uncertainty around each estimate. A survey with scattered gaps in one column may need only a simple fill, provided the bias is checked.
