CODEWITHSUNDEEP
rtidyversepredict
Amc
Amc

Reputation: 159

Predicted values of a poisson in R

I'm trying to find the predicted values of car accidents according to age and sex and finally adjusted to population.

My data is (df):

df <- dplyr::tibble(
  city = c("a", "a", "b", "b", "c", "c"), 
  sex = c(1,0,1,0,1,0),
  age = c(1,2,1,2,1,2),
  population = c(100, 123, 189, 234, 221, 435),
  accidents = c(87, 98, 79, 43,45,65)
)

My code:

library(tidyverse)
library(ggeffects)

poisson<-glm(accidents~sex+age,family="poisson",data=df)

df<-df%>%
  mutate(acc_pred=predict(poisson))

Output:

city    sex age population  accidents   acc_pred
a   1   1   100 87  4.36
a   0   2   123 98  4.43
b   1   1   189 79  4.21
b   0   2   234 43  4.25
c   1   1   221 45  4.26
c   0   2   435 65  3.93

What am I doing wrong?

Upvotes: 0

Views: 483

Answers (1)

Allan Cameron
Allan Cameron

Reputation: 174506

A Poisson glm uses a log link function, and by default the predict.glm method returns the predictions without applying the inverse link function. You either need to use type = "response" inside predict, which will call the inverse link function on the predictions to give you predictions in the same units as your input data, or equivalently, since the inverse link function is essentially just exp, you can exponentiate the results of predict.

So you can do either:

df %>%
  mutate(acc_pred=predict(poisson, type = 'response'))
#>   city sex age population accidents acc_pred
#> 1    a   1   1        100        87 70.33333
#> 2    a   0   2        123        98 68.66667
#> 3    b   1   1        189        79 70.33333
#> 4    b   0   2        234        43 68.66667
#> 5    c   1   1        221        45 70.33333
#> 6    c   0   2        435        65 68.66667

Or

df %>%
  mutate(acc_pred = exp(predict(poisson)))
#>   city sex age population accidents acc_pred
#> 1    a   1   1        100        87 70.33333
#> 2    a   0   2        123        98 68.66667
#> 3    b   1   1        189        79 70.33333
#> 4    b   0   2        234        43 68.66667
#> 5    c   1   1        221        45 70.33333
#> 6    c   0   2        435        65 68.66667

Upvotes: 5

Related Questions