Simple linear regression with tidyverse

broom
tidymodels
Learn simple linear regression with tidyverse in R. Practical tutorial with examples.
Published

September 13, 2023

Introduction

Simple linear regression models the relationship between two continuous variables by fitting a straight line through data points. This technique helps predict one variable based on another and is fundamental for understanding correlations in your data.

Getting Started

library(tidyverse)
library(palmerpenguins)

Example 1: Basic Usage

The Problem

We want to understand if there’s a linear relationship between penguin flipper length and body mass. This will help us predict body mass based on flipper measurements.

Step 1: Explore the Data

First, let’s examine our dataset and visualize the relationship.

penguins |>
  select(flipper_length_mm, body_mass_g) |>
  head(10)

This shows us the structure of our two key variables.

Step 2: Create a Scatter Plot

Visualizing data helps identify linear patterns before modeling.

penguins |>
  ggplot(aes(x = flipper_length_mm, y = body_mass_g)) +
  geom_point() +
  labs(title = "Penguin Flipper Length vs Body Mass")

The scatter plot reveals a clear positive linear relationship between flipper length and body mass.

Step 3: Fit the Linear Model

Now we’ll create our regression model using the lm() function.

model <- penguins |>
  lm(body_mass_g ~ flipper_length_mm, data = _)

summary(model)

The model summary shows coefficients, R-squared value, and statistical significance of our relationship.

Step 4: Add Regression Line to Plot

Visual confirmation helps validate our model’s fit to the data.

penguins |>
  ggplot(aes(x = flipper_length_mm, y = body_mass_g)) +
  geom_point() +
  geom_smooth(method = "lm", se = TRUE) +
  labs(title = "Linear Regression: Flipper Length vs Body Mass")

Scatter plot with linear regression line and confidence band for penguin flipper length vs body mass in R

The regression line with confidence bands shows our model fits the data well.

Example 2: Practical Application

The Problem

A marine biologist needs to estimate penguin body mass when only flipper measurements are available in field studies. We’ll build a predictive model and evaluate its accuracy using the mtcars dataset for comparison.

Step 1: Prepare Training Data

We’ll use complete cases and create a clean dataset for modeling.

clean_penguins <- penguins |>
  filter(!is_na(flipper_length_mm), !is_na(body_mass_g)) |>
  select(flipper_length_mm, body_mass_g)

nrow(clean_penguins)

This ensures we have complete data for accurate model training.

Step 2: Build and Extract Model Coefficients

Creating a model with easily interpretable coefficients for field predictions.

field_model <- lm(body_mass_g ~ flipper_length_mm, data = clean_penguins)

coefficients <- field_model |>
  broom::tidy()

coefficients

The coefficients table provides the intercept and slope needed for manual calculations in the field.

Step 3: Make Predictions on New Data

Testing our model’s predictive capability with hypothetical flipper measurements.

new_measurements <- tibble(flipper_length_mm = c(190, 200, 210, 220))

predictions <- new_measurements |>
  mutate(predicted_mass = predict(field_model, newdata = new_measurements))

predictions

These predictions show how body mass increases with flipper length, giving field researchers estimation guidelines.

Step 4: Evaluate Model Performance

Assessing model quality using standard regression diagnostics.

model_stats <- field_model |>
  broom::glance()

model_stats |>
  select(r.squared, adj.r.squared, p.value)

High R-squared values and low p-values indicate our model explains the relationship well and is statistically significant.

Summary

  • Simple linear regression with tidyverse uses lm() combined with pipe operators for clean, readable code
  • Always visualize your data first with scatter plots to identify linear relationships before modeling
  • Use geom_smooth(method = "lm") to add regression lines to ggplot visualizations
  • Extract model information using broom::tidy() and broom::glance() for tidy data frames

  • Make predictions on new data using predict() with your fitted model object