7

所以，我试图比较两个模型，fit1 和 fit2。

最初，我只是在做 anova(fit1,fit2)，这产生了我理解的输出（包括 p 值）。

然而，当我将我的模型从基于 lm() 的模型切换到基于 glm() 的模型时，anova(fit1,fit2) 现在产生了残差自由度、残差偏差和 Df 偏差，我无法解释它们（资源解释这些指标似乎很少）。我希望为两个模型之间的比较提取 p 值，但由于某种原因 anova(fit1,fit2, test='Chisq') 不起作用。有什么建议么？

我意识到，根据我的 glms 中的链接功能，卡方可能不是最合适的测试，但我在适当的上下文中使用了“F”，也有类似的失望。

这个问题其他人熟悉吗？建议？非常感谢！

例子：

make_and_compare_models <- function(fitness_trait_name, data_frame_name, vector_for_multiple_regression, predictor_for_single_regression, fam){
        fit1<-glm(formula=as.formula(paste(fitness_trait_name,"~", paste(vector_for_multiple_regression, sep="+"))), family=fam, data=data_frame_name)
        print ("summary fit 1")
        print(summary(fit1))
        fit2<- glm(data=data_frame_name, formula=as.formula(paste(fitness_trait_name,"~",predictor_for_single_regression)), family=fam)

        print("summary fit 2")
        print(summary(fit2))
        print("model comparison stats:")
        mod_test<-anova(fit2,fit1)

        ##suggestion #1
        print(anova(fit2,fit1, test="Chisq"))

        #suggestion #2
        print ("significance:")
    print (1-pchisq( abs(mod_test$Deviance[2]),df=abs(mod_test$Df[2])))

        }


data<-structure(list(ID = c(1L, 2L, 4L, 7L, 9L, 10L, 12L, 13L, 14L, 
15L, 16L, 17L, 18L, 20L, 21L, 22L, 23L, 24L, 25L, 27L, 28L, 29L, 
31L, 34L, 37L, 38L, 39L, 40L, 41L, 43L, 44L, 45L, 46L, 47L, 48L, 
49L, 52L, 55L, 56L, 59L, 60L, 61L, 62L, 63L, 65L, 66L, 67L, 68L, 
69L, 71L), QnWeight_initial = c(158L, 165L, 137L, 150L, 153L, 
137L, 158L, 163L, 159L, 151L, 145L, 144L, 157L, 144L, 133L, 148L, 
151L, 151L, 147L, 158L, 178L, 164L, 134L, 151L, 148L, 142L, 127L, 
179L, 162L, 150L, 151L, 153L, 163L, 155L, 163L, 170L, 149L, 165L, 
128L, 134L, 145L, 147L, 148L, 160L, 131L, 155L, 169L, 143L, 123L, 
151L), Survived_eclosion = c(0L, 1L, 0L, 1L, 1L, 1L, 1L, 1L, 
1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 0L, 1L, 
1L, 1L, 1L, 0L, 1L, 1L, 0L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 
1L, 1L, 0L, 1L, 1L, 1L, 1L, 1L, 1L, 1L), Days_wrkr_eclosion_minus20 = c(NA, 
1L, NA, 3L, 0L, 2L, 0L, 1L, 0L, 0L, 0L, 1L, NA, 0L, 7L, 1L, 0L, 
1L, 0L, 1L, 2L, 2L, NA, 2L, 3L, 2L, 2L, NA, 0L, 1L, NA, NA, 0L, 
0L, 0L, 0L, 3L, 3L, 3L, 1L, 0L, 2L, NA, 1L, 0L, 1L, 1L, 3L, 1L, 
2L), MLH = c(0.5, 0.666666667, 0.555555556, 0.25, 1, 0.5, 0.333333333, 
0.7, 0.5, 0.7, 0.5, 0.666666667, 0.375, 0.4, 0.5, 0.333333333, 
0.4, 0.375, 0.3, 0.5, 0.3, 0.2, 0.4, 0.875, 0.6, 0.4, 0.222222222, 
0.222222222, 0.6, 0.6, 0.3, 0.4, 0.714285714, 0.4, 0.3, 0.6, 
0.4, 0.7, 0.625, 0.555555556, 0.25, 0.5, 0.5, 0.6, 0.25, 0.428571429, 
0.3, 0.25, 0.375, 0.555555556), Acon5 = c(0.35387674, 0.35387674, 
0.35387674, 0.35387674, 0.35387674, 0.35387674, 0.35387674, 0, 
0, 1, 0, 1, 0.35387674, 0, 0, 0.35387674, 1, 1, 0, 0, 0, 1, 0, 
0.35387674, 0, 1, 1, 1, 0, 0, 0, 1, 0, 1, 0, 0, 0, 1, 1, 1, 0, 
0, 0, 1, 0, 0, 0, 1, 0, 0.35387674), Baez = c(1, 1, 1, 0.467836257, 
1, 1, 0, 0, 1, 1, 0, 0.467836257, 1, 0, 0, 0, 0, 1, 0, 0, 0, 
0, 0, 0.467836257, 1, 1, 0, 0, 1, 1, 0, 0, 1, 0, 0, 1, 0, 1, 
1, 0, 0, 1, 1, 0, 0, 1, 0, 1, 1, 1), C294 = c(0, 1, 0, 0, 1, 
0.582542694, 0, 1, 1, 0, 0, 1, 0, 0, 0, 0, 1, 1, 0, 1, 0, 0, 
0, 1, 1, 0, 0, 0.582542694, 1, 1, 0, 0, 1, 0, 1, 1, 0, 1, 1, 
1, 1, 1, 1, 1, 1, 1, 1, 0, 1, 1), C316 = c(1, 1, 0, 0, 0.519685039, 
0.519685039, 0, 1, 1, 0, 1, 1, 0, 1, 1, 1, 1, 0.519685039, 0, 
1, 0, 1, 0, 1, 0, 0, 0, 0, 1, 0, 0, 0, 0.519685039, 1, 0, 1, 
1, 0, 0.519685039, 1, 0.519685039, 1, 1, 1, 0.519685039, 0.519685039, 
0, 0.519685039, 0.519685039, 0), i_120_PigTail = c(1, 1, 0, 1, 
0.631236443, 0.631236443, 1, 1, 1, 1, 1, 0, 0.631236443, 1, 1, 
1, 0, 0.631236443, 1, 1, 1, 0, 0, 1, 1, 1, 0.631236443, 0, 1, 
1, 0, 1, 0.631236443, 1, 0, 1, 0, 0, 1, 0.631236443, 0.631236443, 
0, 1, 0, 0.631236443, 0.631236443, 1, 0.631236443, 0.631236443, 
1), i129 = c(0L, 1L, 1L, 0L, 1L, 0L, 1L, 1L, 0L, 1L, 0L, 0L, 
1L, 0L, 0L, 0L, 0L, 0L, 1L, 0L, 0L, 0L, 1L, 1L, 0L, 0L, 0L, 0L, 
0L, 0L, 1L, 0L, 1L, 0L, 0L, 0L, 1L, 0L, 0L, 0L, 0L, 1L, 0L, 0L, 
0L, 0L, 0L, 0L, 0L, 0L), Jackstraw_PigTail = c(0L, 1L, 1L, 0L, 
1L, 1L, 0L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 0L, 0L, 0L, 0L, 0L, 
1L, 0L, 1L, 1L, 1L, 1L, 0L, 1L, 1L, 1L, 1L, 1L, 0L, 1L, 0L, 1L, 
0L, 1L, 1L, 0L, 1L, 1L, 0L, 0L, 0L, 0L, 0L, 0L, 0L, 0L), Neil_Young = c(0.529636711, 
0, 1, 0, 0.529636711, 0.529636711, 1, 1, 0, 1, 1, 1, 0, 0, 1, 
1, 1, 0, 0, 1, 1, 0, 1, 1, 0, 0, 1, 0, 1, 1, 0, 0, 1, 0, 1, 0, 
1, 1, 0, 1, 0, 0, 0, 1, 0, 1, 1, 0, 1, 1), Ramble = c(0, 0, 0, 
0, 0.215163934, 0.215163934, 0, 0, 0, 0, 0, 1, 0, 0, 1, 0, 0, 
0, 0, 1, 0, 0, 1, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0.215163934, 0, 
0, 1, 0, 1, 0, 0, 0, 0, 1, 1, 1, 0.215163934, 0, 0, 0, 0), Sol_18 = c(1, 
0, 1, 1, 1, 0, 0, 1, 0, 1, 1, 0, 0, 1, 0, 0, 0, 0, 1, 0, 0, 0, 
0, 1, 1, 0, 0, 0, 0, 1, 1, 1, 1, 0, 1, 0, 1, 1, 0.404669261, 
1, 0, 0, 0, 1, 0, 0, 0, 0, 0, 1)), .Names = c("ID", "QnWeight_initial", 
"Survived_eclosion", "Days_wrkr_eclosion_minus20", "MLH", "Acon5", 
"Baez", "C294", "C316", "i_120_PigTail", "i129", "Jackstraw_PigTail", 
"Neil_Young", "Ramble", "Sol_18"), class = "data.frame", row.names = c(NA, 
-50L))

make_and_compare_models("QnWeight_initial", data, c("Acon5","Baez","C294","C316","i_120_PigTail","i129","Jackstraw_PigTail","Neil_Young","Ramble","Sol_18"), "MLH", "gaussian")
4

2 回答 2

8

“更大”或更复杂的模型与嵌套或“简化”模型之间的偏差差异（渐近地）分布为卡方变量，这两个模型的自由度不同。因此，您将提取偏差估计和自由度的差异，并将其与 pchisq(deviance, diff(df)) 进行比较。“p 值”只是 1 减去该值。

> 1-pchisq(3.84,1)
[1] 0.05004352

如果您在 glm 帮助页面中运行第一个示例，然后添加一个没有“处理”变量的简化模型，您会得到：

glm.D93.o <- glm(counts ~ outcome, family=poisson())
 anova.res <-anova(glm.D93, glm.D93.o)
 anova.res
#------------
Analysis of Deviance Table

Model 1: counts ~ outcome + treatment
Model 2: counts ~ outcome
  Resid. Df Resid. Dev Df    Deviance
1         4     5.1291               
2         6     5.1291 -2 -2.6645e-15
#---------------
 str(anova.res)
Classes ‘anova’ and 'data.frame':   2 obs. of  4 variables:
 $ Resid. Df : num  4 6
 $ Resid. Dev: num  5.13 5.13
 $ Df        : num  NA -2
 $ Deviance  : num  NA -2.66e-15
 - attr(*, "heading")= chr  "Analysis of Deviance Table\n" "Model 1: counts ~ outcome + treatment\nModel 2: counts ~ outcome"

因此，在查看了事物如何存储在对象本身中之后，这给出了“结果”的 p 值：

 1-pchisq( abs(anova.res$Deviance[2]), abs(anova.res$Df[2]))
[1] 1

这将是治疗+结果模型与仅治疗模型的相应程序：

> glm.D93.t <- glm(counts ~ treatment, family=poisson())
> anova.res2 <-anova(glm.D93, glm.D93.t)
> 1-pchisq( abs(anova.res2$Deviance[2]), abs(anova.res2$Df[2]))
[1] 0.06547071
于 2012-11-05T18:37:29.220 回答
1

如果您的 2 个模型是嵌套的，那么您可以使用 2 个模型的偏差变化来查看包含额外参数的模型是否产生了改进的拟合。如果模型 1 包含k参数，而模型 2 包含这些相同k的参数加上一个附加参数，则偏差的变化遵循（近似）具有自由度的m卡方分布。m您可以使用此检验统计量来查看模型 2 是否是对模型 1 的改进。

如果您是该领域的新手，我强烈建议您阅读有关 GLM 的介绍性文本

于 2012-11-05T18:21:06.127 回答