碩士 / 國立東華大學 / 應用數學系 / 107 / We examine four goodness of fit tests for assessing the goodness of fit of ordinal logistic regression model with adjacent logit link: Cheng's test ($W_{g}$), Pulkstenis and Roinson test (PR), Fagerland and Hosmer test ($C_{g}$) and Osius and Rojek test ($SD_{\lambda}$). The properties of $C_{g}$ and PR tests have previously been investigated for the proportional odds model and the constrained adjacent category logistic model, whereas the W_g test has been investigated for the multinomial logistic models with nominal response. The SD_{lambda} test was proposed as a general method for assessing the lack of fit of any polytomous regression models; the finite sample properties of this test was however not examined yet in settings of ordinal regression models with adjacent logit link. Here, we extend the W_g, PR and C_g tests to unconstrained adjacent category logistic models. Through simulation studies, we compare all four tests based on their performance in type I error rate and power rate to detect different types of lack of fit. The utility of these tests in real applications is presented by the analysis of a data set arising from adolescent placement study.
| Identifer | oai:union.ndltd.org:TW/107NDHU5507005 |
| Date | January 2019 |
| Creators | Hsing-Cheng Pan, 潘星丞 |
| Contributors | Wei-Hsiung Chao, 趙維雄 |
| Source Sets | National Digital Library of Theses and Dissertations in Taiwan |
| Language | en_US |
| Detected Language | English |
| Type | 學位論文 ; thesis |
| Format | 73 |
Page generated in 0.0117 seconds