如何从 elrm 汇总输出中提取系数

Question

我使用包 elrm 对我的数据集进行了精确的逻辑回归

我将它与普通的逻辑回归进行比较。

我能够在普通逻辑回归上运行引导程序，我提取的感兴趣的统计数据是估计系数和 p 值。

但是，我无法运行我的 elrm 引导程序，因为我无法从输出中提取所需的系数。

使用我的数据，摘要会打印出来：

Results:
   estimate p-value p-value_se mc_size
M  0.15116 0.06594    0.00443   49000

95% Confidence Intervals for Parameters

     lower     upper
M 0.00156155 0.3647232

我想提取 M 估计值和 p 值，以便在进行引导时提取这些统计数据。我尝试了一堆不同的组合来尝试提取值，但它们不起作用。

summary(model)$coefficient 
summary(model)$Results
summary(model)$estimate

所有这些只是再次吐出摘要。

有谁知道是否可以从 elrm 摘要中提取？

任何帮助将不胜感激。

Answer 1

看起来 p 值和置信区间是作为模型对象本身的一部分存储的。 summary.elrm 函数只是为您提取和格式化一些信息。我展示了我用来解决这个问题的以下步骤，但摘要版本是分别为 object$p.values 和 object$coeffs.ci 索引到您的模型对象本身。

#Example from help page
data(utiDat) 
uti.elrm <- elrm(uti/n~age+current+dia+oc+pastyr+vi+vic+vicl+vis,
                 interest=~dia,r=4,iter=5000,burnIn=1000,dataset=utiDat)
#Look at summary
summary(uti.elrm)
#Let's examine the structure of the summary object to see what's in there, i.e.
#what can we extract?
str(summary(uti.elrm))
#Hmm, doesn't look like anything of interest. Let's look at the source code itself
summary.elrm
#Looks like the p.values are stored in the actual model object iself and the summary function
#just formats them for us. The relevant part of the summary code for the p-value is:
#-----
  #inferences = as.data.frame(cbind(round(as.numeric(object$coeffs),5), 
  #                                 round(as.numeric(object$p.values), 5), 
  #                                 round(as.numeric(object$p.values.se),5),
  #                                 object$mc.size)
  #                           )
  #results = data.frame(row.names = names(object$coeffs), inferences)
  #names(results) = c("estimate", "p-value", "p-value_se", "mc_size")
#So, it looks like we can grab the p.values directly from "object"
> uti.elrm$p.values
dia 
0.02225
#And the confidence intervals are also in the object, located here in the summary code:
#-----
  #cat(object$ci.level, "% Confidence Intervals for Parameters\n", 
  #    sep = "")
  #cat("\n")
  #print(object$coeffs.ci)
#So we can extract them thusly:
> uti.elrm$coeffs.ci
lower upper
dia 0.1256211   Inf

Answer 2

library(elrm)
data(drugDat)
drug.elrm <- elrm(formula=recovered/n~sex+treatment,interest=~sex+treatment,r=4, 
    iter=100000,burnIn=1000,dataset=drugDat)

> summary.elrm(drug.elrm)

Call:
[[1]]
elrm(formula = recovered/n ~ sex + treatment, interest = ~sex + 
    treatment, r = 4, iter = 1e+05, dataset = drugDat, burnIn = 1000)


Results:
          estimate p-value p-value_se mc_size
joint           NA 0.14886    0.00173   99000
sex        0.27092 0.69385    0.01204    2649
treatment  0.76739 0.07226    0.00314   13160

95% Confidence Intervals for Parameters

               lower    upper
sex       -0.6217756 1.212499
treatment -0.1216884 1.852346

# If you look at the summary function, it simply outputs formatted results to
# the screen. So instead, we can just work with the original drug.elrm object

names(drug.elrm)
# shows you everything in this object

# to see the p-values
drug.elrm$p.values.se
      joint         sex   treatment 
0.001734482 0.012039701 0.003143006 

# to get the p-value for joint
drug.elrm$p.values.se[[1]]

# now for the CI
drug.elrm$coeffs.ci
               lower    upper
sex       -0.6217756 1.212499
treatment -0.1216884 1.852346

> drug.elrm$coeffs.ci[[1]]
[1] -0.6217756 -0.1216884
> drug.elrm$coeffs.ci[[1]][1]
[1] -0.6217756
>