Principal Component Analysis (PCA) untuk Mengatasi Multikolinieritas terhadap Faktor Angka Kejadian Pneumonia Balita di Jawa Timur Tahun 2014

• Main Authors: Fita Mega Kusuma, Arief Wibowo
• Format: Article
• Language: English
• Published: Universitas Airlangga 2018-10-01
• Series: Jurnal Biometrika dan Kependudukan
• Subjects: PCA (Principal Component Analysis), VIF/Tolerance, multicollinearity, pneumonia children under five year
• Online Access: https://e-journal.unair.ac.id/JBK/article/view/4833
• ISSN: 2302-707X; 2540-8828

Correlation between independent variables in multiple linear regression model called multicollinearity. One of the assumptions of multiple linear regression free from multicollinearity problem. Principal Component Analysis (PCA) method in this study aims to overcome the existence of multicollinearity in multiple linear regression and know the dominant factor to the research. PCA method has the advantage of clearing the correlation without losing the original variable. Case study a risk factor that affects the incidence of pneumonia infants in East Java 2014. This non reactive research because uses publication data of health profil of East Java. Result of this research multicollinearity problem in research data when detected by VIF/tolerance method. Variable of vitamin A coverage, measles immunization coverage and health service coverage are the variables that observed multicollinearity. A multicollinearity solution produces (F1) or new variable(coverage of vitamin A, immunization measles and health service), reduction of three variables that multicollinearity to not multicollinearity with VIF value of 1.608 < 10. Results of this study also proves the weakness of PCA method in analyzing the significance. PCA method shows the most influencing factors on the incidence of pneumonia of children under five year. Dominant factor in this research coverage of infant health services covering, coverage of vitamin A and coverage of measles immunization. Coverage factor of health services has a coefficient matrix value of 0.890 or 89% more influential than other factor.