The multiple imputation method: A case study involving secondary data analysis

Aim To illustrate with the example of a secondary data analysis study the use of the multiple imputation method to replace missing data. Background Most large public datasets have missing data, which need to be handled by researchers conducting secondary data analysis studies. Multiple imputation is a technique widely used to replace missing values while preserving the sample size and sampling variability of the data. Data source The 2004 National Sample Survey of Registered Nurses. Review methods The authors created a model to impute missing values using the chained equation method. They used imputation diagnostics procedures and conducted regression analysis of imputed data to determine the differences between the log hourly wages of internationally educated and US-educated registered nurses. Discussion The authors used multiple imputation procedures to replace missing values in a large dataset with 29,059 observations. Five multiple imputed datasets were created. Imputation diagnostics using time series and density plots showed that imputation was successful. The authors also present an example of the use of multiple imputed datasets to conduct regression analysis to answer a substantive research question. Conclusion Multiple imputation is a powerful technique for imputing missing values in large datasets while preserving the sample size and variance of the data. Even though the chained equation method involves complex statistical computations, recent innovations in software and computation have made it possible for researchers to conduct this technique on large datasets. Implications for research/practice The authors recommend nurse researchers use multiple imputation methods for handling missing data to improve the statistical power and external validity of their studies.