In UBC-MDS/linregasm: Package for testing linear and multiple linear regression assumptions

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>"
)

Example usage

Here we will demonstrate how to use linregasm in a project to check linear regression assumptions:

library(linregasm)

Define Sample Data

We will use the Iris dataset from the data package to demonstrate the functionality of lrasm.linregasm`.

data(iris)
head(iris)

In our example, we want to investigate the linear relationship between the response Petal.width versus the predictors: Sepal.Length, Sepal.Width, and petal width. We start by defining the dataframe and a formula to relate the variables within a linear regression model:

formula = "Petal.Width ~ Sepal.Length + Sepal.Width + Petal.Length"
iris_df = iris

Test for Homoscedasticity of Residuals

We can evaluate the homoscedasticity of our residuals by passing our data and formula into the function homoscedasticity(). Note that 'homoscedasticity' returns two objects, the first one is a plot of residuals vs predicted values, while the second is the p value of the correlation between residuals vs predicted values:

hsc_results <- linregasm::homoscedasticity(iris_df, formula)

hsc_results

Test for Normality of Residuals

We can evaluate the normality of our residuals by passing our data and formula into the function normality(). Note that normality() returns two objects, the first one is a p-value from the Shapiro Wilk test, while the second is a string containing either the word Pass or Fail depending on the results of the test. The function will also print a summary statement.

norm_results <- normality(iris_df, formula)

norm_results

After applying the Shapiro Wilks test for normality of the residuals, the regression assumption of normality has passed.

Test for Multicollinearity

We can evaluate the multicollinearity assumption of our data set by passing our data, formula, and the threshold for the VIF value. multicollinearity() takes these arguments and returns a data frame with VIF scores for each feature along with the statement advising the user if the multicollinearity assumption is violated.

mult_results <- linregasm::multicollinearity(iris_df, formula)

mult_results

In this case, since all of our VIF scores were below the threshold of 10, we can conclude that our data satisfies the multicollinearity assumption.

UBC-MDS/linregasm documentation built on Feb. 6, 2022, 7 a.m.