因变量是分类变量的每个级别的回归系列

Question

我想测试一下女性对出院日的影响。 为此，我想运行一系列回归，如果星期一是出院日，则因变量 = 1，否则 = 0。 接下来，如果是星期二，model 将为 =1，否则为 =0……等等。此时的星期几存储在一个名为wkday的分类变量中。

例如，我将如何在for-loop中使用tidymodel for 快速完成此操作？ 这是我到目前为止...

# libraries:
library(tidyr)
library(dplyr)

# create dataset:
id <- seq(1:1000)
wkdays <- c("Monday", "Tuesday", "Wednesday", "Thursday", "Friday", "Saturday", "Sunday")
wkday <- sample(wkdays,1000, replace=T)
female <- sample(0:1, 1000, replace = T)

dta <- data.frame(id=id, wkday=wkday, female=female)

dta$mon <- ifelse(dta$wkday=="Monday",1,0)
dta$tues <- ifelse(dta$wkday=="Tuesday",1,0)
dta$wed <- ifelse(dta$wkday=="Wednesday",1,0)
dta$thurs <- ifelse(dta$wkday=="Thursday",1,0)
dta$fri <- ifelse(dta$wkday=="Friday",1,0)
dta$sat <- ifelse(dta$wkday=="Saturday",1,0)
dta$sun <- ifelse(dta$wkday=="Sunday",1,0)



# Model:
mon <- glm(mon ~ female, data=dta, family = "binomial")
tues <- glm(tues ~ female, data=dta, family = "binomial")

.
.
.

summary(mon)
summary(tues)

Answer 1

也许像下面这样的东西回答了这个问题。
首先，不需要一个一个地手动创建假人， model.matrix就是为此而生的。

library(tidyr)
library(dplyr)
library(purrr)
library(broom)

# create dataset:
set.seed(2021)
id <- 1:1000
wkdays <- c("Monday", "Tuesday", "Wednesday", "Thursday", "Friday", "Saturday", "Sunday")
wkday <- sample(wkdays,1000, replace=T)
female <- sample(0:1, 1000, replace = T)

dta <- data.frame(id=id, wkday=wkday, female=female)

tmp <- model.matrix(~0 + wkday, dta)
colnames(tmp) <- sub("wkday", "", colnames(tmp))

现在是模型。

cbind(dta, tmp) %>% 
  select(-wkday) %>%
  pivot_longer(
    cols = -c(id, female),
    names_to = "wkday",
    values_to = "dummy"
  ) %>% 
  group_by(wkday) %>%
  do(tidy(lm(dummy ~ female, data = .)))
## A tibble: 14 x 6
## Groups:   wkday [7]
#   wkday     term        estimate std.error statistic  p.value
#   <chr>     <chr>          <dbl>     <dbl>     <dbl>    <dbl>
# 1 Friday    (Intercept)  0.128      0.0155     8.26  4.55e-16
# 2 Friday    female       0.0226     0.0219     1.03  3.03e- 1
# 3 Monday    (Intercept)  0.152      0.0163     9.30  8.52e-20
# 4 Monday    female       0.0126     0.0231     0.547 5.84e- 1
# 5 Saturday  (Intercept)  0.170      0.0163    10.4   3.78e-24
# 6 Saturday  female      -0.0234     0.0231    -1.01  3.12e- 1
# 7 Sunday    (Intercept)  0.138      0.0156     8.82  4.88e-18
# 8 Sunday    female       0.00857    0.0221     0.388 6.98e- 1
# 9 Thursday  (Intercept)  0.128      0.0157     8.14  1.13e-15
#10 Thursday  female       0.0326     0.0222     1.47  1.43e- 1
#11 Tuesday   (Intercept)  0.138      0.0145     9.49  1.63e-20
#12 Tuesday   female      -0.0355     0.0205    -1.73  8.41e- 2
#13 Wednesday (Intercept)  0.148      0.0155     9.55  9.66e-21
#14 Wednesday female      -0.0174     0.0219    -0.797 4.26e- 1

---

没有最终 output 中的截距（但带有截距的模型）：

cbind(dta, tmp) %>% 
  select(-wkday) %>%
  pivot_longer(
    cols = -c(id, female),
    names_to = "wkday",
    values_to = "dummy"
  ) %>% 
  group_by(wkday) %>%
  do(tidy(lm(dummy ~ female, data = .))) %>%
  filter(term != "(Intercept)")
## A tibble: 7 x 6
## Groups:   wkday [7]
#  wkday     term   estimate std.error statistic p.value
#  <chr>     <chr>     <dbl>     <dbl>     <dbl>   <dbl>
#1 Friday    female  0.0226     0.0219     1.03   0.303 
#2 Monday    female  0.0126     0.0231     0.547  0.584 
#3 Saturday  female -0.0234     0.0231    -1.01   0.312 
#4 Sunday    female  0.00857    0.0221     0.388  0.698 
#5 Thursday  female  0.0326     0.0222     1.47   0.143 
#6 Tuesday   female -0.0355     0.0205    -1.73   0.0841
#7 Wednesday female -0.0174     0.0219    -0.797  0.426 

---

这个比较简单，不需要创建tmp ，可以在 pipe 中创建。 过滤掉拦截是可选的。

dta%>%
  bind_cols(
    model.matrix(~0 + wkday, dta) %>% as.data.frame
  ) %>%
  select(-wkday) %>% 
  pivot_longer(
    cols = -c(id, female),
    names_to = "wkday",
    values_to = "dummy"
  ) %>% 
  mutate(wkday = sub("^wkday", "", wkday)) %>% 
  group_by(wkday) %>%
  do(tidy(lm(dummy ~ female, data = .)))