构建含两个协变量的Logistic模型遍历变量时公式报错求助

1. 公式构建错误 #

你构建公式的代码冗余且未正确生成公式对象，导致glm无法识别合法的模型公式。原代码嵌套的paste操作虽看似生成了正确字符串，但直接传递字符串给formula参数易出问题，且collapse="+"完全多余（循环中每次仅处理单个变量）。

修正公式的两种简洁方法： #

方法一：转换字符串为公式对象

修改循环内的公式构建代码：

formula <- as.formula(paste("dependent_variable ~", variable))

方法二：用reformulate直接生成公式

reformulate是R中专门用于生成模型公式的函数，无需手动拼接字符串：

formula <- reformulate(variable, response = "dependent_variable")

2. 隐藏的 Data 类型问题 #

你的df中variable1、variable2、variable4、variable5均为字符型数值，glm会默认将其当作分类变量处理，这不符合逻辑回归对数值型预测变量的要求，会导致模型结果完全偏离预期。必须先将这些变量转换为数值型：

# 转换字符型数值为数值型
df$variable1 <- as.numeric(df$variable1)
df$variable2 <- as.numeric(df$variable2)
df$variable4 <- as.numeric(df$variable4)
df$variable5 <- as.numeric(df$variable5)

# 分类变量转为因子（规范操作，避免glm自动处理）
df$variable3 <- as.factor(df$variable3)
df$dependent_variable <- as.factor(df$dependent_variable)

完整修正后的代码 #

# 原始数据
df <- data.frame(
  dependent_variable = c("A", "A","A","B","B"),
  variable1 = c("3","5","2","1","6"),
  variable2 = c("4","2","3","4","0"),
  variable3 = c("b","c","b","a","c"),
  variable4 = c("13","6","20","8","10"),
  variable5 = c("5","5","2","3","1"),
  variable6 = c("group1","group1","group2","group2","group2"),
  variable7 = c("1","2","1","3","2")
)

# 修正数据类型
df$variable1 <- as.numeric(df$variable1)
df$variable2 <- as.numeric(df$variable2)
df$variable4 <- as.numeric(df$variable4)
df$variable5 <- as.numeric(df$variable5)
df$variable3 <- as.factor(df$variable3)
df$dependent_variable <- as.factor(df$dependent_variable)

# 初始化结果存储框
results <- data.frame(variable = character(), coefficient = numeric(), p_value = numeric(), AIC = numeric(), stringsAsFactors = FALSE)

# 预测变量列表
predictor_variables <- c("variable1", "variable2", "variable3", "variable4", "variable5")

# 循环拟合模型
for (variable in predictor_variables) {
  # 生成合法公式
  formula <- reformulate(variable, response = "dependent_variable")
  
  # 拟合逻辑回归
  model <- glm(formula, data = df, family = "binomial")
  
  # 提取模型结果
  coefs <- coef(model)
  p_vals <- summary(model)$coefficients[, "Pr(>|z|)"]
  aic_val <- AIC(model)
  
  # 存储结果（注：若因子变量有多个水平，当前仅提取第二个系数，需按需调整逻辑）
  results <- rbind(results, data.frame(
    variable = variable,
    coefficient = coefs[2],
    p_value = p_vals[2],
    AIC = aic_val,
    stringsAsFactors = FALSE
  ))
}

print(results)

补充提示 #

• 当variable3这类多分类因子作为预测变量时，模型会生成多个系数（对应参考水平以外的每个类别），当前代码仅提取第二个系数，若需完整记录所有因子水平的结果，需调整结果存储逻辑。
• 循环中用rbind拼接数据框效率较低，若预测变量数量较多，建议先初始化固定大小的数据框或用列表存储结果后再合并。

内容的提问来源于stack exchange，提问作者Rafael Bravo