Running multiple glm models on mixed data with purrr

问题

Suppose we have a toy data set:

library(tidyverse)
library(purrr)

tbl <- tibble(a = rep(c(0, 1), each = 5),
              b = rep(c(0, 1), times = 5),
              c = runif(10),
              d = rexp(10)) %>%
    mutate_at(vars(1,2), as.factor)

where a is a dependent variable and b:d are independent variables. The idea is to run glm model for each independent variable:

• glm(a ~ b, data = tbl, family = "binomial")
• glm(a ~ c, data = tbl, family = "binomial")
• glm(a ~ d, data = tbl, family = "binomial")

My initial attempt goes as follows:

tbl %>%
    pivot_longer(2:4, names_to = "key", values_to = "val") %>%
    group_split(key) %>%
    map(~ glm(a ~ val, data = .x, family = "binomial"))

This resulted in an error because data types of b and c (or d) are not the same.

Error: No common type for `b` <factor<dec08>> and `c` <double>.

I wonder how to address this issue.

回答1:

Without reshaping, we can use map to apply glm to various independent variables and use reformulate to create the formula like :

purrr::map(names(tbl)[-1],~glm(reformulate(.x, 'a'),data = tbl,family = 'binomial'))

#[[1]]

#Call:  glm(formula = reformulate(.x, "a"), family = "binomial", data = tbl)

#Coefficients:
#(Intercept)           b1  
#    -0.4055       0.8109  

#Degrees of Freedom: 9 Total (i.e. Null);  8 Residual
#Null Deviance:     13.86 
#Residual Deviance: 13.46   AIC: 17.46

#...
#...

来源：https://stackoverflow.com/questions/60598466/running-multiple-glm-models-on-mixed-data-with-purrr