如何在逻辑模型预测（使用lsmeans）后获得绝对差异

Question

我正在使用逻辑模型来获得具有三个独立预测变量的二元结果。调整模型后，我想用边际均值（lsmeans）估计我的结果（y）的预测，并对比这些预测。但是，我注意到，如果我使用的模型是逻辑模型，则对比度是以相对比例（比率）而不是绝对比例（差异）完成的。有没有人知道我如何才能获得从逻辑模型中获得的预测对比（使用lsmeans）的绝对差异？

以下是一个例子：

library(lsmeans)

#### Creating my Data
y<- rbinom(10000, 1, .02)
a<- sample(c(2005:2014), 10000, replace = T, prob = c(0.10023452, 0.09867466,
                                                       0.09869976, 0.09880913,
                                                       0.09903145, 0.10275898,
                                                       0.10470253, 0.09909241, 
                                                       0.09889877, 0.09909779))
b<- rbinom(10000, 1, 0.57)
c<- sample(c(1:4), 10000, replace = T, prob = c(0.61969154, 0.12735996, 0.16385892, 0.08908958))


df<-data.frame(cbind(y, a, b, c))
df$a<-factor(df$a)
df$b<-factor(df$b)
df$c<-factor(df$c)

#### Logistic Model

m1<-glm(y~a*b*c, family = quasibinomial(link = "logit"), data = df)
summary(m1)

#### LSMEANS prediction
m1.ls<-lsmeans(m1, "a", by=c("b", "c"), infer=c(TRUE, TRUE), level=.95, 
    data=df)

#### Checking my predictions in the log odds scale
summary(m1.ls)

#### Checking my predictions in the probability scale
summary(m1.ls, type = "response")

m1.ctr.odds<-summary(contrast(m1.ls, infer=c(TRUE, TRUE), by=c("b", "c"), 
                           level=.95, method="revpairwise"))

m1.ctr.prob<-summary(contrast(m1.ls, infer=c(TRUE, TRUE), by=c("b", "c"), 
                           level=.95, method="revpairwise"), type = "response")


#### Contrasts are in the log odds ratio scale. NOTE this are RATIOS of log odds
m1.ctr.odds

#### Contrasts are in the probability scale. NOTE this are RATIOS of probabilities
m1.ctr.prob #### I would like these contrasts to be absolute differences (substractions) 

最好，

→