定量变量的分级方法对logistic模型影响的研究

A Study of the Influence of Dividing Methods of Quantitative Variable on Logistic Model

目的获得logistic回归模型将定量自变量转化为等级变量的正确划分方法。方法 R语言编程模拟计算三种分布下不同分级方法所得ROC曲线下面积、logistic回归AIC信息量,并比较优劣。结果服从不同的分布的自变量,其分级方法不同,拟合效果也不同。结论特定分布下,可得到相对固定的较优分级方法。非特定分布下,通过计算各分级方法的ROC曲线下面积以及AIC信息量,并比较指标优劣可获得较优分级方法。

Objective Through Logistic regression analysis simulation, to obtain correctly dividing method that transform Quantitative variable into Categorical variable under different circumstances. And finally forming systemic, comprehensive dividing suggestion through tables or graphs which facilitate interpreting. Methods Calculating the areas under the ROC curve （AUC） and AIC of different dividing methods under three kinds of distribution by the R programming language simulation. Then comparing methods＇ advantages and disadvantages. Results Variables with different distribution should be subject to dif- ferent dividing methods, which can affect the model fit. Condusion Under the specific distribution, we can obtain the best and fixed dividing method. While for other distributions, AUC and AIC of different dividing methods should be calculated. We can choose the better one by comparing the two indexes.