lm-and-glm.R

library(tidyverse)
library(palmerpenguins)
library(broom)

# Get rid of missing values
penguins <- penguins %>% drop_na(sex)

# Scatterplot
ggplot(penguins, aes(x = bill_length_mm, y = body_mass_g)) +
  geom_point() + 
  geom_smooth(method = "lm")

# Model with one slider (bill length)
model_simple <- lm(body_mass_g ~ bill_length_mm, data = penguins)
tidy(model_simple)

# A 1 mm increase in bill length is associated with an 86.8 g increase in body mass, on average

# Model with one switch (species)
model_2 <- lm(body_mass_g ~ bill_length_mm + species + flipper_length_mm, data = penguins)
results <- tidy(model_2, conf.int = TRUE)
results

# Holding bill length and flipper length constant, Chinstrap penguins are 732 g lighter than Adelie penguins on average

# Coefficient plot for flipper length and bill length
results %>% 
  filter(term != "(Intercept)", !str_detect(term, "species")) %>% 
  ggplot(aes(x = estimate, y = term)) +
  geom_pointrange(aes(xmin = conf.low, xmax = conf.high)) +
  geom_vline(xintercept = 0)

# Coefficient plot for species
results %>% 
  filter(term != "(Intercept)", str_detect(term, "species")) %>% 
  ggplot(aes(x = estimate, y = term)) +
  geom_pointrange(aes(xmin = conf.low, xmax = conf.high)) +
  geom_vline(xintercept = 0)


# Make a binary column
penguins <- penguins %>% 
  mutate(is_gentoo = species == "Gentoo")

# Logistic regression
model_logit <- glm(is_gentoo ~ bill_length_mm + body_mass_g, 
                   data = penguins,
                   family = binomial())
tidy(model_logit, exponentiate = TRUE, conf.int = TRUE)

# When exponentiated, all these coefficients are centered around 1 and are percents. So a 1 mm increase in bill length makes it 9% more likely to be a Gentoo