Pengaruh Pendugaan Data Hilang terhadap Hasil Analisis Model Persamaan Struktural (Studi Kasus: Indeks Pembangunan Manusia Pulau Jawa)

Abstract

Data hilang merupakan salah satu faktor yang dapat menurunkan kualitas data, karena dapat menyebabkan bias pada pendugaan parameter model sehingga kesimpulan yang didapatkan menjadi tidak tepat (Chen et al. 2012). Upaya untuk mengatasi data hilang dapat ditempuh melalui imputasi dengan menggunakan beberapa metode, antara lain linear interpolation, linear trend, dan serial mean. Dalam penelitian ini, Analisis Structural Equation Modeling Partial Least Square (SEM-PLS) digunakan untuk melihat dampak adanya data hilang terhadap hubungan kausalitas antar peubah yang direpresentasikan melalui tingkat validitas dan reliabilitas data dalam pendugaan koefisien model. Hasil evaluasi model menunjukkan metode linear interpolation saat persentase data hilang 5% dan 10%, metode linear trend dan serial mean saat persentase data hilang 5% menghasilkan model yang memenuhi semua kriteria dalam evaluasi model. Hasil dari akurasi dan ketepatan menggunakan MAPE menunjukkan metode linear interpolation saat persentase data hilang 5%, 10%, dan 15%, metode linear trend saat persentase data hilang 5% dan 10%, metode serial mean saat persentase data hilang 5% memberikan hasil yang baik dalam menduga nilai koefisien model.

Missing data is one of factors that can reduce data quality, because it can cause biased in estimating model parameters so that conclusions obtained are incorrect (Chen et al. 2012). Efforts to overcome missing data can be achieved through imputation using several methods, including linear interpolation, linear trend, and serial mean. Structural Equation Modeling Partial Least Square (SEM-PLS) analysis was used to see impact of missing data on causal relationships between variable which can represent level of validity and reliability of data in estimating model coefficients. Results of model evaluation show linear interpolation method when the percentage of missing data is 5% and 10%, serial mean and linear trend methods when the percentage of missing data is 5% produces models that meet all the criteria in model evaluation. Results of the accuracy and precision using MAPE show that linear interpolation method when percentage of missing data is 5%, 10%, and 15%, linear trend method when percentage of missing data is 5% and 10%, serial mean method when percentage of missing data is 5% give good result in estimating value of the model coefficients.