中文摘要：

目的：比较logistic回归分析中估计回归系数可信区间及假设检验的三种方法。方法：从理论上介绍Wald、Bootstrap和轮廓似然三种方法，比较其应用前提。应用Monte-Carol模拟方法，分别产生自变量为二分类变量和连续型变量的二分类logistic分析数据，比较三种方法的准确性和稳定性。结果：无论自变量为连续型或二分类变量，Bootstrap法和轮廓似然法的检验效能均高于Wald法，该差异在小样本情形下更为明显。但Bootstrap方法的稳定性较差，对Ⅰ类错误的控制也欠佳。结论：轮廓似然方法最为稳健，能严格控制I类错误率，且检验效能优于wald方法，值得推荐。

英文摘要：

Objective： To compare the statistical performances of three statistical methods in estimating confidence interval of regression coefficients and hypothesis testing in a logistic regression model. Methods： We introduced three methods, including the Wald-based, Bootstrap and Profile-likelihood method, and com- pared their assumptions. Then we generated logistic data regressed on a dichotomous-variable and a continu- ous-variable, respectively, using the Monte-Carlo method. Analyzing the simulated data using three different methods, we compared their accuracy and stability. Results： Whatever type of the variable is, the power of Bootstrap method and Profile-likelihood method are both higher than the Wald-based method, particularly in the case of small sample sizes. The Bootstrap method is relatively unstable and performs poorer in controlling the type I error. Conclusion. Profile-likelihood method can strictly control the type I error and always have a more efficient power than the Wald-based method, which should be strongly recommended.