Estimasi Parameter dan Pengujian Hipotesis Pada Model Geographically Weighted Bivariate Zero-Inflated Negative Binomial Regression

Sari, Henni Jovita (2024) Estimasi Parameter dan Pengujian Hipotesis Pada Model Geographically Weighted Bivariate Zero-Inflated Negative Binomial Regression. Masters thesis, Institut Teknologi Sepuluh Nopember.

[thumbnail of 6003221016_Master_Thesis.pdf] Text
6003221016_Master_Thesis.pdf

Download (3MB)

Abstract

Geographically Weighted Bivariate Zero-Inflated Negative Binomial Regression (GWBZINBR) adalah salah satu metode yang dapat digunakan untuk menganalisis variabel respon cacahan yang mengalami overdispersi diakibatkan adanya excess zero. GWBZINBR merupakan pengembangan regresi bivariate Negative Binomial dengan adanya excess zero dan mengakomodasi heterogenitas spasial antar observasi, dimana setiap observasi akan diberikan pembobot Adaptive Gaussian Kernel. Estimasi parameter model GWBZINBR dilakukan dengan menggunakan Maximum Likelihood Estimation (MLE) dengan iterasi numerik Berndt-Hall-Hall-Hausman (BHHH), sedangakan pengujian hipotesis menggunakan Maximum Likelihood Ratio Test (MLRT). Selanjutnya, metode GWBZINBR diaplikasikan pada kasus jumlah kematian ibu hamil dan jumlah kematian ibu nifas di Kabupaten Cilacap dan Kabupaten Kebumen tahun 2021 dengan menggunakan 4 variabel prediktor yang diduga berpengaruh terhadap jumlah kematian ibu hamil dan jumlah kematian ibu nifas di Kabupaten Cilacap dan Kabupaten Kebumen tahun 2021. Pada penelitian ini diketahui bahwa GWBZINBR lebih baik daripada BZINBR yang ditunjukkan dengan nilai Akaike Information Criterion Corrected (AICc) dan Sum Square Error (SSE) pada model GWBZINBR lebih kecil dibandingkan dengan Bivariate Zero-Inflated Negative Binomial Regression (BZINBR). Pemodelan GWBZINBR menghasilkan 3 kelompok kecamatan berdasarkan kesamaan variabel yang signifikan terhadap jumlah kematian ibu hamil maupun jumlah kematian ibu nifas, namun untuk zero-inflated state tidak membentuk kelompok karena seluruh variabel prediktor berpengaruh signifikan terhadap jumlah kematian ibu hamil dan jumlah kematian ibu nifas. Persentase ibu hamil yang mendapatkan tablet tambah darah berpengaruh signifikan pada semua kecamatan berdasarkan jumlah kematian ibu hamil dan jumlah kematian ibu nifas.
===============================================================================================
Geographically Weighted Bivariate Zero-Inflated Negative Binomial Regression (GWBZINBR) is one of the methods that can be used to analyze response variables with overdispersion due to excess zero. GWBZINBR is a development of bivariate Negative Binomial regression with excess zero and accommodates spatial heterogeneity between observations, where each observation will be given an Adaptive Gaussian Kernel weight. Parameter estimation of the GWBZINBR model uses Maximum Likelihood Estimation (MLE) with Berndt-Hall-Hausman (BHHH) numerical iteration, while hypothesis testing uses Maximum Likelihood Ratio Test (MLRT). Furthermore, the GWBZINBR method is applied to the case of the number of deaths of pregnant women and the number of deaths of postpartum women in Cilacap and Kebumen in 2021 using 4 predictor variables that are thought to affect the number maternal mortality and the number postpartum mortality in Cilacap and Kebumen in 2021. In this study, it is known that GWBZINBR is better than Bivariate Zero-Inflated Negative Binomial Regression (BZINBR), which is indicated by the value of Akaike Information Criterion Corrected (AICc) and Sum Square Error (SSE) of the GWBZINBR model is smaller than BZINBR. GWBZINBR modeling produces 3 groups of sub-districts based on the similarity of variables that are significant to the number of maternal mortality and the number of postpartum mortality, but the zero-inflated state does not form groups because all predictor variables have a significant effect. The percentage of pregnant women who received blood supplement tablets was significant in all sub-districts based on the number of maternal mortality and the number of postpartum mortality.

Item Type: Thesis (Masters)
Uncontrolled Keywords: BHHH, Excess Zero, GWBZINBR, Maternal Mortality, MLE, MLRT, Kematian Ibu
Subjects: H Social Sciences > HA Statistics > HA31.7 Estimation
Divisions: Faculty of Science and Data Analytics (SCIENTICS) > Statistics > 49101-(S2) Master Thesis
Depositing User: Henni Jovita Sari
Date Deposited: 07 Aug 2024 22:27
Last Modified: 07 Aug 2024 22:27
URI: http://repository.its.ac.id/id/eprint/114207

Actions (login required)

View Item View Item