How to Perform Multilevel Modeling in R (With Example)

Image by Editor | Midjourney

Multilevel modeling (MLM), or hierarchical linear modeling (HLM), is used to analyze grouped data. For example, students within schools or patients within hospitals. Regular regression methods don’t work well because they don’t consider the relationships within the groups, which can lead to wrong results.

MLM helps researchers study patterns at different levels, looking at both differences within and between groups. This makes it useful for complex data. In R, you can use packages like lme4 and lmerTest to perform MLM.

This article will show you how to perform multilevel modeling in R.

Installing and Loading Required Packages

To perform multilevel modeling in R, you will need the following packages:

lme4: Provides functions to fit linear mixed-effects models.
lmerTest: Adds p-values and statistical tests to lme4 models.
ggplot2: For visualizing results, such as plotting random effects.

Run the following commands to install and load the necessary packages:

# Install required packages
install.packages(c("lme4", "lmerTest", "ggplot2"))

# Load the libraries
library(lme4)
library(lmerTest)
library(ggplot2)

Loading Data

Here, we will use the Exam dataset from the mlmRev package. The dataset contains exam scores for students. These students are nested within schools. It also includes variables like exam scores and socioeconomic status.

# Install the package
install.packages("mlmRev")

# Load the package
library(mlmRev)

# Load the Exam dataset
data("Exam")

# View the first few rows of the dataset
head(Exam)

The dataset contains the following key variables:

normexam: Students’ exam scores (dependent variable).
school: Identifier for schools (grouping variable).
sex: Students’ gender (predictor).

Fitting a Random Intercept Model

A random intercept model gives each group its own starting point. The effect of the predictors is the same for all groups. This model explains group differences while keeping the relationship between predictors and the outcome constant.

The formula used in the lmer() function specifies the relationship:

Dependent variable (normexam): Students’ exam scores.
Fixed effect predictor (sex): A categorical variable.
Random intercept ((1 | school)): Allows the intercept to vary by school.

# Random intercept model
model <- lmer(normexam ~ sex + (1 | school), data = Exam)

# Display the model summary
summary(model)

Key output:

Fixed Effects: Shows the average effect of gender on exam scores across all schools.
Random Effects: Displays the variability in intercepts across schools.

Fitting a Random-Slope Model

A random-slope model lets the effect of a predictor change across groups. Each group can have its own slope. This means the relationship between the predictor and the outcome can be different for each group. It helps understand more complicated patterns in the data.

# Random intercept and slope model
model_slope <- lmer(normexam ~ sex + (sex | school), data = Exam)

# Display the summary
summary(model_slope)

Here, (sex | school) specifies that both the intercept and the slope of gender can vary across schools.

Comparing Models

Use the anova() function to compare the random intercept model and the random slope model. This helps determine if adding random slopes improves the model fit.

# Compare models with ANOVA
anova(model, model_slope)

If the p-value is small, the random slope model improves the fit and should be preferred.

Visualizing Results

Random effects provide insight into how much variability exists between groups. You can visualize the random intercepts using a histogram.

# Load ggplot2 for visualization
library(ggplot2)

# Extract random effects
random_effects <- ranef(model)$school

# Convert to data frame
random_effects_df <- as.data.frame(random_effects)

# Inspect the structure of the data frame
str(random_effects_df) 
# Plot random intercepts
ggplot(random_effects_df, aes(x = `(Intercept)`)) +
    geom_histogram(binwidth = 0.1, fill = "blue", color = "black") +
    labs(title = "Random Intercepts for Schools", x = "Random Intercept", y = "Frequency")

Conclusion

Multilevel modeling in R helps you analyze data with a group structure. Using the lme4 package, you can make models that look at both fixed effects (predictors) and random effects (group differences). This example showed how to build and understand random intercept and random slope models.

With these tools, you can better understand your grouped data.