What Is Simple Regression? Definition and Uses

Simple regression fits a straight line through paired data so that one variable can predict another. Where correlation only reports that two variables move together, regression quantifies the relationship: the slope says how much the output changes for each unit change in the input, and the intercept anchors the line. It is the first modeling tool most practitioners learn, and the gateway to every model after it.

How it works

The fitting method, least squares, chooses the line that minimizes the total squared distance between the data points and the line itself. The fitted equation delivers three things. The slope, the working heart of the model: each additional degree, minute or millimeter of input moves the output this much. A p-value for that slope, testing whether the relationship is distinguishable from noise. And R-squared, the share of the output’s variation the line explains — a measure of how much of the story this one input tells. The residuals — the leftover gaps between the points and the line — are the diagnostic: they should scatter randomly, with no curve, funnel or drift.

One rule keeps regression honest: predict only within the range of the data that built the model. The line knows nothing about conditions it never saw, and real processes rarely stay linear forever.

A worked example

Take an electroplating line where coating thickness drifts from run to run. The team suspects bath temperature and records both across many runs. Regression fits the line and reports a positive slope — each additional degree adds a predictable increment of thickness — with a small p-value and an R-squared indicating temperature explains a substantial share of the drift. Now the team can do two things correlation alone never allows: predict thickness at a proposed setpoint inside the tested range, and calculate how tightly temperature must be controlled to hold thickness within specification. The numbers would be illustrative; the payoff is standard.

• Never extrapolate beyond the data — the line is only trustworthy where it was fed.
• Read the residual plot before the R-squared; patterns in the leftovers mean the model is wrong.
• A fitted slope does not prove causation — confirm with a controlled change before re-engineering the process.
• High R-squared is not the goal; a model can fit history beautifully and predict nothing.
• Outliers and high-leverage points can steer the whole line — investigate them, never silently delete them.

Regression turns a cloud of history into a line you can plan with — inside the range you tested, and nowhere else.

Simple regression closes out the Analyze toolkit of our Green Belt program — $299 for 35 hours, with a 100-question proctored exam and lifetime access to the material. Black Belt training extends it to multiple regression, where several inputs compete to explain a single output.