Dampak Missing Data terhadap Reliabilitas dan Akurasi Hasil Analisis Kasus Structural Equation Modeling

Abstract

Missing data is a problem that often become an obstacle for researcher in data analysis. Missing data can cause bias in a model which can lead to inaccuracies of estimation parameter and wrong conclusion (Chen et al. 2016). General countermeasure used to resolve the adverse of missing data is by removing certain respondent with missing data from dataset. However, wasting data on fillen entry will reduce research efficiency. Another method used to resolve missing data is by imputing data using various approaches, including serial mean, linear interpolation, linear trend. Using Structural Equation Modeling (SEM) analysis, the researcher can observe internal relationship of a large number of variable which can represent the level of data reability in estimating the real data. Missing data estimation method in the SEM produces different estimated parameter coefficient values even when using the same simulation characteristic data. The estimation results show that the linear trend method is optimal in resolving the adverse of missing data in missing precentages of 5%, 10% and 15%. Serial mean method optimal in missing precentages of 5% and 10%. Linear interpolation method optimal in missing precentages 5%. Listwise deletion method is not optimal enough to resolve the adverse of missing data in missing precentages of 5%, 10%, 15% and 20%. The result of goodness of fit model using chi-square test, p-value, RMSEA and the reability level using construct reliability (CR) and variance extracted (VE) shows that increasing the percentage of missing data will cause a decrease in the efficiency of estimation. But more than that, the compatibility of the missing data pattern with the estimated value of each method used will affect the level of efficiency and optimization of the estimation. The result of accuracy and precision level using MAPE shows that in resolving the adverse of missing data the serial mean, linear interpolation, and linear trend methods with a missing precentage 5% are feasible in estimating the parameters of the control SEM model. Meanwhile the listwise deletion method for each missing precentage is not accurate to estimate the parameters of the control SEM model.

Missing data is a problem that often become an obstacle for researcher in data analysis. Missing data can cause bias in a model which can lead to inaccuracies of estimation parameter and wrong conclusion (Chen et al. 2016). General countermeasure used to resolve the adverse of missing data is by removing certain respondent with missing data from dataset. However, wasting data on fillen entry will reduce research efficiency. Another method used to resolve missing data is by imputing data using various approaches, including serial mean, linear interpolation, linear trend. Using Structural Equation Modeling (SEM) analysis, the researcher can observe internal relationship of a large number of variable which can represent the level of data reability in estimating the real data. Missing data estimation method in the SEM produces different estimated parameter coefficient values even when using the same simulation characteristic data. The estimation results show that the linear trend method is optimal in resolving the adverse of missing data in missing precentages of 5%, 10% and 15%. Serial mean method optimal in missing precentages of 5% and 10%. Linear interpolation method optimal in missing precentages 5%. Listwise deletion method is not optimal enough to resolve the adverse of missing data in missing precentages of 5%, 10%, 15% and 20%. The result of goodness of fit model using chi-square test, p-value, RMSEA and the reability level using construct reliability (CR) and variance extracted (VE) shows that increasing the percentage of missing data will cause a decrease in the efficiency of estimation. But more than that, the compatibility of the missing data pattern with the estimated value of each method used will affect the level of efficiency and optimization of the estimation. The result of accuracy and precision level using MAPE shows that in resolving the adverse of missing data the serial mean, linear interpolation, and linear trend methods with a missing precentage 5% are feasible in estimating the parameters of the control SEM model. Meanwhile the listwise deletion method for each missing precentage is not accurate to estimate the parameters of the control SEM model