Comparisons of imputation methods on different types of survey research data: A continuous variable.

Missing data problems are commonly unavoidable and affect the outcome of many studies. The insufficiency of data resulted in inaccurate results and predictions in many statistical analyses. In survey studies, datasets with missing values require some imputation method to continue with reliable statistical analyses. However, the many imputation methods available are confusing. Thus, this study aims to compile the characteristics of missing values in survey data, mapping it to suggested imputation methods. In addition, the performances of five missing data imputation methods which are mean imputation, median imputation, deterministic regression imputation, stochastic regression imputation, and predictive mean matching (PMM), were compared. Two survey datasets were used in this study, and the performance of the five compared methods was evaluated using root-mean-square error (RMSE). Results indicated that for deterministic regression imputation performed the best (RMSE = 0.3674863) and the predictive mean matching imputation (RMSE = 0.3780853) performed the least for survey data "Malaysian Perception on Rising Cost of Living". However, for the second survey dataset "A Retrospective International Study on Factors Associated with Injury, Discomfort, and Pain Perception among Cyclists" resulted in the versa, the predictive mean matching imputation (RMSE = 0.4223341) performed the best, and deterministic regression imputation performed the least (RMSE = 0.3780853). In conclusion, the selection of imputation methods should be based on the type of variable and the unique features of the datasets. [ABSTRACT FROM AUTHOR]