Close-Knit-Regression: An Efficient Technique in Estimating Missing Completely at Random Data

Abstract

The study aimed at using the Close-Knit Regression (CKR) technique to approximate values absent because of the missing completely at random mechanism. Bivariate datasets were generated and simulated for MCAR mechanism at low (10%) and high (60%) rates. The CKR method was used and compared alongside other single imputation techniques like mean imputation, simple regression and K- Nearest Neighbors (K-NN). The difference between parameter estimates like mean, correlation coefficient (r), maximum, minimum and standard deviation which were gotten using predicted data and those using the original data as well as assessment of error rates like mean absolute error (MAE) and root mean square error (RMSE) were used as metrics in deciding the efficiency of the techniques. Results showed that the CKR technique was the best from those considered, with its estimated data having parameter estimates closer to that of the original whilst having the least error rates at 10% (MAE of 0.01 and RMSE of 0.047) and 60% (MAE of 0.021 and RMSE of 0.073) in comparison to other methods, CKR technique is a suitable single imputation technique which produces estimates close to the original data and parameters with low error rates when data are MCAR.