R语言GLM模型summary未显示Deviance Residuals的问题咨询

要让glm模型的summary()输出默认包含**偏差残差（Deviance Residuals）**的五数统计（最小值、第一四分位数、中位数、第三四分位数、最大值），可以通过两种方式实现：

方法一：覆盖默认的print.summary.glm函数 #

直接修改R中默认的glm摘要打印函数，让它在输出中自动加入偏差残差部分：

# 定义新的print.summary.glm函数，覆盖默认行为
print.summary.glm <- function (x, digits = max(3, getOption("digits") - 3), symbolic.cor = x$symbolic.cor,
          signif.stars = getOption("show.signif.stars"), ...)
{
  # 打印模型调用语句
  cat("\nCall:\n", paste(deparse(x$call), sep = "\n", collapse = "\n"),
      "\n\n", sep = "")
  
  # 新增：计算并打印偏差残差统计
  dev_resid <- resid(x$model, type = "deviance")
  cat("Deviance Residuals: \n")
  print(summary(dev_resid)[c("Min.", "1st Qu.", "Median", "3rd Qu.", "Max.")],
        digits = digits)
  cat("\n")
  
  # 以下为原默认函数的其余逻辑，保持不变
  if (length(x$aliased) == 0) {
    cat("No Coefficients\n")
  }
  else {
    if (!is.null(x$coefficients)) {
      coefs <- x$coefficients
      if (is.matrix(coefs)) {
        printCoefmat(coefs, digits = digits, signif.stars = signif.stars,
                     na.print = "NA", ...)
      }
      else {
        cat("Coefficients:\n")
        print(coefs, digits = digits)
      }
    }
    if (length(x$aliased) > 0) {
      cat("\n(Dispersion parameter for ", x$family$family,
          " family taken to be ", format(x$dispersion), ")\n\n",
          sep = "")
      df <- x$df
      if (df[2] > 0) {
        cat("    Null deviance: ", format(signif(x$null.deviance,
                                                digits)), "  on ", df[1], "  degrees of freedom\n",
            sep = "")
        cat("Residual deviance: ", format(signif(x$deviance,
                                                 digits)), "  on ", df[2], "  degrees of freedom\n",
            sep = "")
      }
      else {
        cat("Residual deviance: ", format(signif(x$deviance,
                                                 digits)), "  on ", df[2], "  degrees of freedom\n",
            sep = "")
        cat("(Null deviance undefined for the multinomial family)\n",
            sep = "")
      }
      cat("AIC: ", format(signif(x$aic, digits)), "\n\n", sep = "")
      if (!is.null(x$iter)) {
        cat("Number of ", x$method, " iterations: ", x$iter,
            "\n", sep = "")
      }
      if (!is.null(x$na.action)) {
        cat("\n", x$na.action, "\n", sep = "")
      }
      if (!is.null(x$correlation) && symbolic.cor) {
        cat("\nCorrelation of Coefficients:\n")
        tmp <- x$correlation
        diag(tmp) <- NA
        print(tmp, digits = digits, na.print = "")
      }
      else if (!is.null(x$correlation)) {
        cat("\nCorrelation matrix not shown by default, as it is symmetric.\n")
        cat("Use print(x, correlation=TRUE) to see it.\n")
      }
    }
  }
  invisible(x)
}

使用方式 #

运行上述函数后，直接调用summary(model)即可得到包含偏差残差的输出：

model <- glm(am ~ disp + hp, data=mtcars, family=binomial)
summary(model)

方法二：自定义打印函数（不修改全局默认行为） #

如果不想覆盖全局的print.summary.glm函数，可以写一个自定义函数来组合输出：

# 自定义打印函数，包含偏差残差
print_glm_with_devresid <- function(model_summary) {
  # 打印模型调用语句
  cat("\nCall:\n", paste(deparse(model_summary$call), sep = "\n", collapse = "\n"), "\n\n", sep = "")
  
  # 计算并打印偏差残差统计
  dev_resid <- resid(model_summary$model, type = "deviance")
  cat("Deviance Residuals: \n")
  print(summary(dev_resid)[c("Min.", "1st Qu.", "Median", "3rd Qu.", "Max.")], 
        digits = max(3, getOption("digits") - 3))
  cat("\n")
  
  # 打印系数表
  if (!is.null(model_summary$coefficients)) {
    coefs <- model_summary$coefficients
    if (is.matrix(coefs)) {
      printCoefmat(coefs, digits = max(3, getOption("digits") - 3), 
                   signif.stars = getOption("show.signif.stars"), na.print = "NA")
    } else {
      cat("Coefficients:\n")
      print(coefs, digits = max(3, getOption("digits") - 3))
    }
  }
  
  # 打印分散参数、偏差值、AIC等信息
  cat("\n(Dispersion parameter for ", model_summary$family$family, 
      " family taken to be ", format(model_summary$dispersion), ")\n\n", sep = "")
  df <- model_summary$df
  if (df[2] > 0) {
    cat("    Null deviance: ", format(signif(model_summary$null.deviance, max(3, getOption("digits") - 3))), 
        "  on ", df[1], "  degrees of freedom\n", sep = "")
    cat("Residual deviance: ", format(signif(model_summary$deviance, max(3, getOption("digits") - 3))), 
        "  on ", df[2], "  degrees of freedom\n", sep = "")
  } else {
    cat("Residual deviance: ", format(signif(model_summary$deviance, max(3, getOption("digits") - 3))), 
        "  on ", df[2], "  degrees of freedom\n", sep = "")
    cat("(Null deviance undefined for the multinomial family)\n", sep = "")
  }
  cat("AIC: ", format(signif(model_summary$aic, max(3, getOption("digits") - 3))), "\n\n", sep = "")
  
  # 打印迭代次数
  if (!is.null(model_summary$iter)) {
    cat("Number of ", model_summary$method, " iterations: ", model_summary$iter, "\n", sep = "")
  }
  
  invisible(model_summary)
}

使用方式 #

先生成模型摘要，再调用自定义函数：

model <- glm(am ~ disp + hp, data=mtcars, family=binomial)
mod_sum <- summary(model)
print_glm_with_devresid(mod_sum)

内容的提问来源于stack exchange，提问作者Nauru