碩士 / 國立陽明大學 / 公共衛生研究所 / 95 / Previous studies have shown that when the number of events is low the parameter estimates of the logistical regression analysis are unreliable. Applying propensity score method is shown to improve the performances of the logistic regression with sparse events. However, the impact of this phenomenon on correlated data has not yet been explored. The purposes of this study are to examine the effect of sparse events on the estimates of logistic regression models with correlated data. The performances of the conditional logistic regression(CLR), the generalized linear mixed model (GLMM), and the generalized estimating equation (GEE) models with and without propensity score(PS) are compared.
We conducted Monte Carlo simulation studies to compare the estimates by conditional logistic regression, generalized linear mixed model, and generalized estimating equation in cluster data. Data generated with different number of clusters and cluster sizes, different magnitude of the intra-correlation within subjects, and varying percents of events will be compared.
Our results showed that when the data contained approximately less than 10% events, the point estimates by conditional logistic regression with propensity score (CLR+PS) had least bias, were more efficient, and the 95% coverage proportion was close to 95%, while the results by CLR and those by GLMM were similar in most situations. Given the same size of total observations, the 95% confidence intervals calculated by generalized estimating equation are getting worse as the number of clusters decreased. On the other hand, when the number of events was more than 10%, the parameter estimates by the models with propensity score were more biased than those without PS.
| Identifer | oai:union.ndltd.org:TW/095YM005058030 |
| Date | January 2007 |
| Creators | Chung-Kai Huang, 黃忠凱 |
| Contributors | I-Feng Lin, 林逸芬 |
| Source Sets | National Digital Library of Theses and Dissertations in Taiwan |
| Language | zh-TW |
| Detected Language | English |
| Type | 學位論文 ; thesis |
| Format | 85 |
Page generated in 0.0111 seconds