Using residuals dsitributions to evaluate model

residqq.R

library(ggplot2)
library(patchwork)
library(rethinking)

data(Howell1)

# linear model and linear log model

linmodel <- lm(height ~ weight, data = Howell1)
logmodel <- lm(height ~ log(weight), data = Howell1)

# fit charts

g1 <- ggplot(data = Howell1, aes(x = weight, y = height)) +
  geom_point(size = 0.8) +
  geom_jitter(color = "blue") +
  geom_smooth(method = "lm", color = "red") + 
  theme_minimal() +
  labs(x = "Weight", y = "Height")
  
g2 <- ggplot(data = Howell1, aes(x = weight, y = height)) +
  geom_point(size = 0.8) +
  geom_jitter(color = "blue") +
  geom_smooth(formula = "y ~ log(x)", method = "lm", color = "red") + 
  theme_minimal() +
  labs(x = "Weight", y = "Height")

# density/hist

g3 <- ggplot() +
  geom_histogram(aes(x = linmodel$residuals, y=..density..), fill = "lightblue", color = "pink") +
  geom_density(aes(x = linmodel$residuals), alpha=.2, fill="#FF6666", inherit.aes = FALSE) +
  theme_minimal() +
  labs(x = "Residual", y = "Density")

g4 <- ggplot() +
  geom_histogram(aes(x = logmodel$residuals, y=..density..), fill = "lightblue", color = "pink") +
  geom_density(aes(x = logmodel$residuals), alpha=.2, fill="#FF6666", inherit.aes = FALSE) +
  theme_minimal() +
  labs(x = "Residual", y = "Density")

# qqplots

g5 <- ggplot() +
  stat_qq(aes(sample = linmodel$residuals), color = "blue", size = 1.5) +
  stat_qq_line(aes(sample = linmodel$residuals), color = "red") +
  theme_minimal() +
  labs(x = "Theoretical", y = "Sample")

g6 <- ggplot() +
  stat_qq(aes(sample = logmodel$residuals), color = "blue", size = 1.5) +
  stat_qq_line(aes(sample = logmodel$residuals), color = "red") +
  theme_minimal() +
  labs(x = "Theoretical", y = "Sample")

# combine

(g1 | g2) /
(g3 | g4) /
(g5 | g6)