博士 / 國立陽明大學 / 公共衛生研究所 / 100 / In a binary model relating a response variable Y to a risk factor X, it may be necessary to consider an extraneous effect from Z that is related to X, or to both X and Y. In such an instance, Z is called an extraneous variable. The proposed method of values deviated from fitted values (VDFV) of a fractional polynomial may reduce the estimation bias, especially when the relationship between the covariates (e.g., X and Z) is nonlinear. The generation of fitted values may be based on non-diseased subjects only, or on both non-diseased and diseased subjects (pooled data), depending on whether Z is unrelated (pattern I) or related (pattern II) to Y. Our simulation-based study revealed that VDFV-p (using pooled data) is reliable with less bias and a smaller mean square error (MSE) in pattern I, and that VDFV-c (using non-diseased data) shows less bias in pattern II. The improvement using the VDFV method is more apparent when the relationship between the main covariate (X) and the extraneous variable (Z) is nonlinear, which is a common occurrence in real applications. Note that an extremely large MSE is never observed in VDFV-p or VDFV-c, though this is a common issue related to a small sample size or sparse data in logistic regression. Two fetal studies are described in this paper, one for pattern I and the other for pattern II.
| Identifer | oai:union.ndltd.org:TW/100YM005058025 |
| Date | January 2012 |
| Creators | Yang-Wen Yang, 楊雅雯 |
| Contributors | Chong-Yau Fu, 傅瓊瑤 |
| Source Sets | National Digital Library of Theses and Dissertations in Taiwan |
| Language | en_US |
| Detected Language | English |
| Type | 學位論文 ; thesis |
| Format | 67 |
Page generated in 0.0508 seconds