[linear regression, model, lm, prediction, model evaluation, linear inferences]


In the social sciences, regresion analysis is a popular tool to estimate relationships between a dependent variable and one or more independent variables. It is a way to find trends in data, quantify the impact of input variables, and make predictions for unseen data.

In this building block, we illustrate how to estimate a model, identify outliers, plot a trend line, and make predictions.


Estimate Model

Linear regression (lm) is suitable for a response variable that is numeric. For logical values (e.g., did a customer churn: yes/no), you need to estimate a logistic regression model (glm). The code sample below estimates a model, checks the model assumptions, and shows the regression coefficients.

  • Model transformations can be incorporated into the formula, for example: formula = log(y) ~ I(x^2).
  • The coefficients (coefficients(mdl)), predictions for the original data set (fitted(mdl)), and residuals (residuals(mdl)) can be directly derived from the model object.
  • A concrete example on how to evaluate model assumptions (mean residuals is 0, residuals are normally distributed, homskedascticiy) can be found here.

# estimate linear regression model
# to estimate a logistic regression model use:
# glm(formula = y ~ x, data = data, family = binomial)
mdl <- lm(formula = y ~ x, data = data)

# check model assumptions
  which = 1:3,
  nrow = 1,
  ncol = 3

# show regression coefficients

Identify Outliers

Compute the leverage of your data records and influence on mdl to identify potential outliers.

leverage_influence <- mdl %>%
    augment() %>%
    select(y, x, leverage = .hat, cooks_dist = .cooksd) %>%
    arrange(desc(cooks_dist)) %>%

Plot Trend Line

Plot a scatter plot of two numeric variables and add a linear trend line on top of it.


ggplot(data = data, aes(x, y)) +
geom_points() +
geom_smooth(method = "lm", se = FALSE)

Make Predictions

Given a linear regression model (mdl), make predictions for unseen input data (explanatory_data). Note that for multiple linear regression models, you need to pass an explanatory_data object with multiple columns.

explanatory_data <- c(..., ..., ...)

prediction_data <- explanatory_data %>%
    y = predict(
      type = "response"

# See the result

Export Model Output

You can export your model output using stargazer. This package will create a nicely-formatted regression table for you in a variety of formats. You can learn more about it here.

Convert regression coefficients of mdl_1 and mdl_2 into a HTML file that can be copied into a paper.


stargazer(mdl_1, mdl_2,
          title = "Figure 1",
          column.labels = c("Model 1", "Model 2"),

This tutorial outlines how to run, evaluate, and export regression model results for the cars dataset. In particular, it analyzes the relationship between a car’s speed and the stop distance.

A trend plot in R.