Abstract:
The objective of this research is the comparison on the accuracy of regression-coefficient estimation of multiple regression with errors in independent variables. This research compares three multiple-regression-coefficient estimation methods, Ordinary Least Squares method (OLS), Total Least Squares method (TLS), and Bayesian Latent Variable Regression method (BLVR). The measurement for making decision is Average Mean Squares Error (AMSE) and Ratio of Different Average Mean Squares Error (RDAMSE). As for the case study, we specify beta (1, 1, , 1) for OLS and TLS methods. However, in the case of BLVR method, the distribution of beta is assumed to be normal distribution with mean equal to 1 and standard deviation equal to 0.5, the numbers of independent variables are equivalent to 3, 5 and 7, and the sample sizes are 30, 50, 70, 100 and 150. All independent variables have errors which random error in each independent variable (e[subscript ~xj]) has the average of normal distribution equivalent to 0 and standard deviations are equal to 0.1, 0.3, 0.5, 0.7 and 1.0, respectively. The distribution of error in the dependent variable (epsilon[subscript ~y]) is normal distribution with mean equal to 0 and standard deviations equal to 0.1, 0.3, 0.5, 0.7 and 1.0, respectively. The data for this research is simulated by using the Monte Carlo simulation technique with 500 repetitions for each case. The results of this research are as follows: According to the comparison of AMSE from three referred methods, we found that when the standard deviation of the errors in independent variables is equal to 1 and the standard deviations of the error in dependent variable are equal to 0.1, 0.3 and 0.5, OLS method would give the lowest AMSE. Conversely, when standard deviations of error in dependent variable are equal to 0.7 and 1.0, OLS method is the best method for large sample size while BLVR method is the most appropriate for small sample size. Additionally, in case of the standard deviation of errors in independent variables is equal to 0.3, OLS method would give the lowest AMSE when the number of independent variables is equal to 3 and with large sample size. As well, when the numbers of independent variables are equal to 5 and 7, AMSE value from BLVR method is the lowest. Furthermore, when the standard deviations of errors in independent variables are equal to 0.5, 0.7, and 1.0, from the result, the lowest AMSE is from BLVR method, while TLS and OLS are the second and the third lowest, respectively. From all methods, the factors that effect to AMSE are sample size, the number of independent variables, the standard deviation of errors in independent variables and the standard deviation of error in dependent variable.