Wize Business Analytics Textbook > Correlation and Regression

Objectives of Regression (Prediction, not Causality)

What Regression Is For: Prediction and Description, Not Proof of Cause

Regression can be useful without being causal

A regression summarizes how a response variable changes on average across observed values of an explanatory variable.


Its primary predictive use is conditional: for a new case similar to the observed cases and within the observed range, use the fitted relationship to estimate the response.


This can support staffing, inventory, forecasting, triage, and many other decisions even when the explanatory variable is not a cause.

Prediction language differs from intervention language

A predictive statement compares cases that already differ in the explanatory variable: cases with larger observed x have a different predicted mean y. A causal statement asks what would happen if the same unit were changed from one value of x to another. Those questions agree only when the data-generating design justifies a causal interpretation.


Observational predictors are often associated with other variables that influence the outcome. Temperature can jointly affect two summer activities; student choices can reflect motivation or available time; marketing spend can be planned in anticipation of demand.


A regression coefficient may make an excellent forecast because it captures those patterns, while still failing to estimate the isolated effect of an intervention.

Design determines the causal claim

Random assignment of an exposure or treatment, with credible implementation and adequate follow-up, makes treatment groups comparable in expectation and supports causal estimation in the assigned range.


In nonrandomized data, a causal claim requires a defensible identification strategy that addresses relevant confounding, selection, measurement, and timing. Statistical significance or a narrow confidence interval does not repair a biased comparison.

Choose the regression for the prediction target

The least-squares line predicting y from x minimizes vertical residuals and is not generally the algebraic inverse of the line predicting x from y, which minimizes horizontal residuals.


To predict a response, fit or use the regression with that response on the left-hand side. Reversing the fitted equation can lead to systematically different predictions unless the relationship is perfectly linear without scatter.


Keep the claim at the evidence level. A line fitted to past sales can forecast future sales in comparable conditions. It does not alone prove that increasing an advertising budget, requiring tutoring, or changing a policy will produce the slope-sized change. A transparent answer separates the numerical prediction from the additional causal question.

City records for 40 summer weeks show a strong positive linear relationship between weekly ice-cream sales and the number of swimming-pool rescues. Which use of the regression of rescues on ice-cream sales is supported by these data?

Choose an answer

Create a free account to get started

Unlock full course notes by signing up

  • Video lessons taught by top educators
  • Practice questions with guided solutions
  • Written review notes for reference
  • Expert answers anytime from course tutors

A regression of sale price on floor area for 250 houses sold in one city last year gives a slope of $140 per square foot. Which statement is the best interpretation?

Choose an answer

Create a free account to get started

Unlock full course notes by signing up

  • Video lessons taught by top educators
  • Practice questions with guided solutions
  • Written review notes for reference
  • Expert answers anytime from course tutors

Mark Yourself Question

  1. Grab a piece of paper and try this problem yourself.
  2. When you're done, check the "I have answered this question" box below.
  3. View the solution and report whether you got it right or wrong.

At one school, records for 180 students give the regression line ŷ = 52 + 3.5x, where y is final exam score and x is the number of optional tutoring hours the student chose to attend (observed range 0 to 12 hours). (a) Predict the mean exam score for students who attended 8 hours. (b) The principal proposes requiring 8 hours of tutoring for students who attended none, and claims their scores will rise by 28 points. In two or three sentences, explain whether the regression supports this claim.

Create a free account to get started

Unlock full course notes by signing up

  • Video lessons taught by top educators
  • Practice questions with guided solutions
  • Written review notes for reference
  • Expert answers anytime from course tutors