3. Inferential Statistics in R

Inferential statistics let you draw conclusions about a population based on sample data. This tutorial covers the most common hypothesis tests and when to use each one.

Setup Link to heading

library(tidyverse)
data(mtcars)
mtcars$cyl <- factor(mtcars$cyl)

The Logic of Hypothesis Testing Link to heading

  1. State null hypothesis (H₀) and alternative hypothesis (H₁)
  2. Choose a significance level (α, typically 0.05)
  3. Compute the test statistic
  4. Determine the p-value
  5. Reject or fail to reject H₀

Remember: A p-value is NOT the probability that H₀ is true. It is the probability of observing data at least as extreme as yours, assuming H₀ is true.

One-Sample t-Test Link to heading

Is the population mean different from a specific value?

# Test whether mean MPG differs from 20
t.test(mtcars$mpg, mu = 20)

Two-Sample t-Test Link to heading

Do two groups differ in their means?

# Compare MPG between 4-cylinder and 8-cylinder cars
four_cyl <- mtcars$mpg[mtcars$cyl == "4"]
eight_cyl <- mtcars$mpg[mtcars$cyl == "8"]

# Independent samples t-test (Welch's, default)
t.test(four_cyl, eight_cyl)

# Equal variance assumed (Student's t-test)
t.test(four_cyl, eight_cyl, var.equal = TRUE)

Checking Assumptions Link to heading

# Normality
shapiro.test(four_cyl)
shapiro.test(eight_cyl)

# Equal variances (Levene's test)
library(car)
leveneTest(mpg ~ cyl, data = mtcars %>% filter(cyl %in% c("4", "8")))

Paired t-Test Link to heading

# Example with simulated before/after data
set.seed(42)
before <- rnorm(30, mean = 100, sd = 15)
after <- before + rnorm(30, mean = 5, sd = 8)

t.test(before, after, paired = TRUE)

One-Way ANOVA Link to heading

Compare means across 3+ groups.

# Does MPG differ across cylinder groups?
aov_model <- aov(mpg ~ cyl, data = mtcars)
summary(aov_model)

Post-Hoc Tests Link to heading

# Tukey's HSD for pairwise comparisons
TukeyHSD(aov_model)

# Visualise
plot(TukeyHSD(aov_model))

Checking ANOVA Assumptions Link to heading

# Residual plots
par(mfrow = c(2, 2))
plot(aov_model)

# Levene's test for homogeneity of variances
leveneTest(mpg ~ cyl, data = mtcars)

Two-Way ANOVA Link to heading

mtcars$gear <- factor(mtcars$gear)

aov_two <- aov(mpg ~ cyl * gear, data = mtcars)
summary(aov_two)

# Interaction plot
interaction.plot(mtcars$cyl, mtcars$gear, mtcars$mpg,
                 col = 1:3, lwd = 2, type = "b",
                 xlab = "Cylinders", ylab = "Mean MPG",
                 trace.label = "Gears")

Chi-Square Test Link to heading

Test association between two categorical variables.

# Are cylinders and transmission type (am) related?
tab <- table(mtcars$cyl, mtcars$am)
tab

chisq.test(tab)

Non-Parametric Alternatives Link to heading

When normality assumptions are violated:

Parametric TestNon-Parametric AlternativeR Function
One-sample tWilcoxon signed-rankwilcox.test(x, mu = ...)
Two-sample tMann-Whitney Uwilcox.test(x, y)
Paired tWilcoxon signed-rankwilcox.test(x, y, paired = TRUE)
One-way ANOVAKruskal-Walliskruskal.test()
# Kruskal-Wallis test
kruskal.test(mpg ~ cyl, data = mtcars)

# Mann-Whitney U test
wilcox.test(four_cyl, eight_cyl)

Effect Sizes Link to heading

# Cohen's d for t-test
library(effectsize)
# install.packages("effectsize")
cohens_d(four_cyl, eight_cyl)

# Eta-squared for ANOVA
eta_squared(aov_model)

Confidence Intervals Link to heading

# CI for a mean
t.test(mtcars$mpg)$conf.int

# CI for a proportion
prop.test(sum(mtcars$am), nrow(mtcars))

Key Takeaways Link to heading

  • Always check assumptions before choosing a test.
  • Use Welch’s t-test (the default) unless you have good reason to assume equal variances.
  • Report effect sizes alongside p-values — statistical significance ≠ practical significance.
  • When assumptions are violated, use non-parametric alternatives.

Previous: ← Data Visualisation Next: Regression & Modelling →