ANALISIS REGRESI LOGISTIK EKSAK PADA PENANGANAN SAMPEL KECIL

• 主要作者: ISNAWATI, 090013943 M
• 格式: Theses and Dissertations NonPeerReviewed
• 語言: Indonesian; Indonesian
• 出版: 2002
• 主題: QA319-329 Functional Analysis
• 在線閱讀: http://repository.unair.ac.id/35785/1/gdlhub-gdl-s2-2006-isnawati-708-tkm_04-03.pdf; http://repository.unair.ac.id/35785/12/35785.pdf; http://repository.unair.ac.id/35785/; http://lib.unair.ac.id
• 機構: Universitas Airlangga

Correlation between a categorical response variable and one or several predictor variables involving large samples is analyzed using logistic regression method. However, if the sample size is small or the data is sparse, the relevance of conventional (asymptotic) logistic regression method to use in such correlation analysis will be questioned. This is a statistical study for application of exact logistic regression in the condition of small sample size and sparse data. Exact logistic regression analysis was done to cases with sample size of 10, 20, 29, and 55 taken from results of random sampling data obtained by Immunization. The dependent variable was immunization status, and the independent variable was people's exposure to information, education, occupation, living children, knowledge, attitude and participation. The data were analyzed with Logxact Turbo program. Results showed that parameter estimation and hypothesis test using exact test provided better solution compared to conventional (asymptotic) regression test such as likelihood ratio, Wald, and score tests. Exact test also provided correlation type or model with sample size of 20, 30, and 55 and wider confidence interval compared to asymptotic inference type. Probability between those intervals had, therefore, larger parameter of population. For tests involving numerous independent variables, exact logistic regression method also provided better solution compared to three conventional (asymptotic) logistic regression tests. Exact logistic regression method should be used in statistical test with small sample size or sparse data and tests that use table with sparse value in each of its cells.