The datasaurus dozen is a collection of datasets that demonstrate how summary statistics can be misleading. This gist shows how to access the data and visualize it.

datasaurus_example.R

# If you don't have devtools, install that first. If you have Windows and R version 4.0 (or greater), see: https://cran.r-project.org/bin/windows/Rtools/
devtools::install_github("lockedata/datasauRus")

# Load the datasaurus dozen and our favorite suite of data processing tools
library(datasauRus)
library(tidyverse)

# Quick look at the contents of the datasaurus_dozen data frame
summary(datasaurus_dozen)
unique(datasaurus_dozen$dataset)

# Scatter plots of the "slant up" and high lines" datasets
filter(datasaurus_dozen, dataset == "slant_up") %>% 
  ggplot(aes(x, y)) +
  geom_point() +
  theme_classic()
filter(datasaurus_dozen, dataset == "high_lines") %>% 
  ggplot(aes(x, y)) +
  geom_point() +
  theme_classic()

# Scatter plot of the legend, the myth, the datasaurus
filter(datasaurus_dozen, dataset == "dino") %>% 
  ggplot(aes(x, y)) +
  geom_point() +
  theme_classic()

# Scatter plots of the full collection
ggplot(datasaurus_dozen, aes(x, y, color = dataset)) +
  geom_point() +
  facet_wrap(~ dataset) +
  theme_classic() +
  theme(legend.position = "none")

# This is what makes the datasaurus dozen so special: despite *looking* so distinctive to human eyes, they all
# have the same summary statistics (mean, standard deviation, and correlation coefficient). A reminder that we
# need to be discerning when we look at summary statistics, whether in a research article or in training a 
# machine learning algorithm. Looking at the raw data, or using algorithms that look at the raw data, can be the
# difference between useful inference and misleading nonsense.
datasaurus_dozen %>% 
  group_by(dataset) %>% 
  summarize(mean(x),
            mean(y),
            sd(x),
            sd(y),
            cor(x, y))