除了使用 glm 之外,还有其他方法可以在 R 中拟合泊松模型吗?

Is there an alternative way to fit Poisson model in R besides using glm?

当我尝试在 R 中的给定数据集中拟合泊松模型时出现错误。我正在努力理解错误的原因。

library(COUNT)  # Titanic dataset
data("titanic")
library(tidyverse)

# Number of missing values
titanic %>%
  map_int(~sum(is.na(.)))


# Fit the Poisson regression model

poifit <- glm(survived ~ class, family = poisson, data = titanic)


titanic2 <- titanic %>%
  mutate(across(.cols = everything(), ~as.factor(.)))

poifit2 <- glm(survived ~ class, family = poisson, data = titanic2)

我收到错误:

Warning in Ops.factor(y, 0) : ‘<’ not meaningful for factors
Error in if (any(y < 0)) stop("negative values not allowed for the 'Poisson' family") : 
  missing value where TRUE/FALSE needed

您可能会感到困惑。您无法将泊松拟合到分类响应中。在将 survival "yes"/"no" 转换为 0/1 之后,您 可以 将泊松分布拟合到二进制数据,但这实际上没有意义:

glm(as.numeric(survived=="no") ~ class, family = poisson, data = titanic)

明智的做法(可能)是 cross-tabulate 并使用值,例如

cc <- as.data.frame(table(titanic))
glm(Freq ~ ., data = cc, family = poisson)