GLM with multiple categorical predictors

C

Comfort Eagle

New Member
#1
#1
Hello internet,

lets say I want to predict an outcome using multiple categorical predictors. Please consider the follow reproducible R-code:

Code: 
library(dplyr)
library(psych)

set.seed(42)
n <- 500

df <- data.frame(
  loneliness = round(runif(n, 0, 1)),
  pet = factor(round(runif(n, 0, 1))),
  addiction = factor(round(runif(n, 0, 2))),
  sex = factor(round(runif(n, 0, 1)))
)

levels(df$pet) <- c('dog', 'cat')
levels(df$addiction) <- c('none', 'low', 'high')
levels(df$sex) <- c('male', 'female')

df_coded <- data.frame(
  loneliness = df$loneliness,
  psych::dummy.code(df$pet) %>% as_tibble() %>%  setNames(paste0('pet_', names(.))),
  psych::dummy.code(df$addiction) %>% as_tibble() %>%  setNames(paste0('addiction_', names(.))),
  psych::dummy.code(df$sex) %>% as_tibble() %>%  setNames(paste0('sex_', names(.)))
)

model_formula <- as.formula(
  paste0(
    'loneliness ~ ',
    paste0(colnames(df_coded)[-1], collapse = ' + ')
  )
)

model_fit <- glm(formula = model_formula, data = df_coded, family = binomial)
print(summary(model_fit))
As you can see, I wish to predict loneliness (no/yes) by what kind of pet (dog/cat) the subject has, by if the subject has a certain level of addiction (none/low/high) and by the subjects biological sex (male/female). So far, so good. I've dummy-coded each of these predictors, resulting in one variable for each predictor level to use them in my linear model. I know that one dummy-variable is usually omitted because it represent the baseline and expressed by the intercept. However, I find this problematic in the case of multiple categorical predictors since the intercept would then represent the intersection between one level of each categorical predictor (e.g. intercept = pet_dog + addiction_none + sex_male). I do not want my intercept to represent this interaction, in fact, I do not care about any intersect at all since i merely wish to evaluate the impact of each category on the outcome. How is this usually achieved?

Thanks a lot!
 
Dason

Dason

Ambassador to the humans
#4
#4
So if you don't care about it... why do you care about it? It literally is going to represent a predicted value for some combination of your factors. If you don't have anything specific that you want it to represent in mind why does this particular representation bother you?
 
You must log in or register to reply here.