Simple Linear Regression: A Practical Guide

Simple linear regression fits a straight line through your data to predict one variable from another. Where correlation tells you two variables move together, regression tells you the shape of that relationship in real units: how much y changes for each one-unit increase in x, and what y looks like when x is zero. It's the workhorse of applied quantitative research and the foundation for nearly every model that follows.

This guide covers what regression adds beyond correlation, how to read a slope and intercept, what least squares is actually minimizing, how to interpret r squared and residuals, the four assumptions the model rests on, a complete worked example calculated by hand, and how to report results the way journals expect. If you're not yet comfortable with what a correlation coefficient measures, start with the guide to correlation coefficients and what r actually measures, since regression builds directly on it.

What Regression Adds Beyond Correlation

Correlation and simple linear regression describe the same underlying relationship, but they answer different questions. Three differences matter in practice.

Regression has a direction. Correlation is symmetric: r between study hours and exam score is identical to r between exam score and study hours. Regression is not. You choose one variable as the outcome and one as the predictor, and swapping them produces a different equation. That choice is yours to justify from theory, not something the math decides.

Regression is in real units. A correlation of 0.80 is unitless and tells you the relationship is strong. A slope of 2.4 tells you that each additional hour of study is associated with 2.4 more exam points. For applied questions, the second is far more useful, because it answers "how much" rather than just "how strongly."

Regression predicts. The equation lets you estimate y for a value of x you haven't observed. Correlation offers no such machinery.

What regression does not add is causation. A regression of y on x is still an observational association unless your study design supports a causal claim. The language of "predicting" and "explaining" is statistical convention, not evidence of a causal mechanism.

The Regression Equation

The fitted line is described by two numbers.

ŷ = b ₀ + b ₁x

Here ŷ (read "y-hat") is the predicted value of the outcome, b ₁ is the slope, and b ₀ is the intercept. Your choice of which variable plays which role comes from your research question, and the terminology for those roles is covered in the guide to dependent and independent variables.

Interpreting the slope

The slope is the number that answers your research question. It gives the expected change in y for a one-unit increase in x. A slope of 2.4 in a model predicting exam score from study hours means that each additional hour of study is associated with an expected increase of 2.4 points. A negative slope means y decreases as x increases.

Always state the slope with its units attached. "2.4 points per hour" is interpretable; "the slope was 2.4" is not. Reviewers notice the difference, because the units are what make the finding meaningful to a reader in your field.

Interpreting the intercept

The intercept is the predicted value of y when x equals zero. Sometimes that's meaningful: if x is hours of study, the intercept estimates the expected score for a student who didn't study at all.

Often it isn't. If x is adult height in centimeters, the intercept predicts the outcome for someone zero centimeters tall, which is nonsense. The intercept still has to be there mathematically, because it positions the line vertically, but you don't have to interpret it substantively when zero falls outside the plausible range of your data. Say so rather than inventing a meaning for it.

How Least Squares Chooses the Line

Infinitely many lines could be drawn through a scatter of points. Ordinary least squares picks one by a specific rule: it finds the line that makes the sum of the squared vertical distances from each observed point to the line as small as possible.

Those vertical distances are the residuals, the gap between what actually happened and what the line predicted. The residual for any observation is the observed y minus the predicted ŷ. Points above the line have positive residuals; points below have negative ones.

Squaring the residuals before summing them does two things, and the logic is the same as in the variance and standard deviation formulas. It stops positive and negative gaps from canceling each other out, and it penalizes large misses much more heavily than small ones. A point 10 units off the line contributes 100 to the total, while a point 2 units off contributes only 4. The fitted line is therefore pulled harder by distant observations, which is why a single outlier can visibly tilt a regression line in a small sample.

A Worked Example: Study Hours and Exam Score

Here's the full calculation on the same eight-student dataset used in the correlation guide, so you can see how the two connect. Study hours is the predictor; exam score is the outcome.

Step 1: Compute the slope

The slope is the sum of the cross-products of the deviations divided by the sum of squared x deviations. Both quantities were calculated in the correlation guide: the cross-products sum to 247.00, and the squared x deviations sum to 92.00.

b ₁ = Σ(x − M _x)(y − M _y) / Σ(x − M _x)²

b ₁ = 247.00 / 92.00 = 2.685

Each additional hour of study is associated with an expected increase of about 2.69 exam points.

Step 2: Compute the intercept

The regression line always passes through the point defined by both means, which gives a direct route to the intercept. With M _x = 7.0 and M _y = 79.25:

b ₀ = M _y − b ₁M _x

b ₀ = 79.25 − (2.685 × 7.0) = 79.25 − 18.79 = 60.46

The fitted equation is therefore ŷ = 60.46 + 2.685x. A student who studied zero hours would be predicted to score about 60.5, which is interpretable here because zero hours is a plausible value.

Step 3: Check the fit

The correlation between these two variables is r = 0.989. In simple linear regression, r squared is exactly the square of that correlation:

R² = (0.989)² = 0.979

About 97.9 percent of the variance in exam scores is accounted for by study hours in this sample. That's unusually high because the data was constructed to be clean; real behavioral data rarely comes close.

Step 4: Look at the residuals

The residual column in the table shows the model's misses. They're small (all under 2 points), they're both positive and negative, and they don't grow systematically as x increases. That's what a well-behaved model looks like. Residuals are where you check whether the model is appropriate at all, which brings us to the assumptions.

The Four Assumptions

Simple linear regression rests on four assumptions. Violating them doesn't always invalidate the model, but it changes what you can claim, and reviewers expect to see that you checked. A useful mnemonic is LINE: linearity, independence, normality of residuals, equal variance.

• Linearity. The relationship between x and y is actually a straight line. Check by plotting the data before fitting anything. A curved pattern means a straight line is the wrong model, no matter how significant the slope comes out.
• Independence. Observations don't depend on one another. This is usually a design question rather than something you test after the fact. Repeated measurements of the same person, or observations clustered within schools or clinics, violate it and call for a different model.
• Normality of residuals. The residuals, not the raw variables, should be approximately normally distributed. Check with a histogram or a Q-Q plot of the residuals. This assumption matters most for small samples; it relaxes as n grows.
• Equal variance (homoscedasticity). The spread of the residuals stays roughly constant across the range of x. Plot residuals against predicted values: you want a shapeless band. A funnel or cone shape, where the spread widens as predictions increase, indicates heteroscedasticity and unreliable standard errors.

All four are checked visually more than statistically, which makes plotting essential rather than optional. The guide to choosing the right visualization covers the scatter plots and residual plots that do this work.

Reading Regression Output

Statistical software returns more numbers than most write-ups need. These are the ones that matter and what each one is telling you.

The significance test on the slope uses the same logic as any other hypothesis test: assume no effect, then ask how surprising the observed result would be. If that framework isn't familiar, see the guide to hypothesis testing.

Reporting Regression in APA Format

A complete write-up gives the reader the model, the coefficient, its precision, and the fit. A minimal but sufficient sentence looks like this:

Study hours significantly predicted exam score, b = 2.69, 95% CI [2.29, 3.08], t(6) = 16.65, p < .001. The model explained 97.9% of the variance in exam scores, R² = .98, F(1, 6) = 277.06, p < .001.

A few conventions worth holding to. Report the unstandardized coefficient b with its units, since that's the interpretable quantity. Include the confidence interval, which most journals now expect alongside the p-value. Give the degrees of freedom in parentheses for both t and F. And italicize the statistical symbols ( b, t, p, R², F) while leaving the numbers upright.

Then interpret the slope in plain language in the surrounding text. A reader should learn what 2.69 means in your study's terms without having to work it out themselves.

Common Mistakes

• Reading the slope as causal. Regression estimates association. Unless your design supports causal inference, write "was associated with" rather than "caused" or "led to."
• Skipping the scatter plot. A curved relationship can still produce a significant slope and a respectable r squared. Only the plot reveals that a straight line was the wrong model.
• Interpreting an intercept outside the data range. If x never approaches zero in your sample, the intercept is a mathematical necessity rather than a meaningful prediction. Note that instead of interpreting it.
• Extrapolating beyond the observed range. The fitted line describes the range of x you observed. Predicting exam scores for 40 study hours, when your data tops out at 12, assumes the relationship continues unchanged. It usually doesn't.
• Treating a high r squared as validation. R squared measures fit, not correctness. A model can fit well and still violate its assumptions, omit a crucial variable, or be built on a nonsensical predictor.
• Reporting only the p-value. Significance tells the reader the slope is unlikely to be zero, not how large or important it is. The coefficient and its confidence interval carry that information.
• Ignoring influential outliers. Because least squares squares the residuals, one extreme point can pull the line noticeably. Inspect the residual plot and report whether results hold with and without any influential observation.

Where to Go Next

Simple linear regression handles one predictor. Most real research questions involve several at once, which is where multiple regression comes in, allowing you to estimate the effect of one predictor while holding others constant. When your outcome is binary rather than continuous, such as pass or fail, logistic regression is the appropriate model instead.

For the correlation foundations underneath this article, see the guides to what the correlation coefficient measures and to using correlation coefficients in research papers, which covers coefficient selection and reporting conventions.

When the analysis is written up, the results section is where a subject-matter editor earns their place, because a reviewer who spots a slope described in causal language or a model reported without its assumptions will question everything around it. Editor World's academic editing, dissertation editing, and journal article editing services include review of statistical reporting, and you choose your own editor by field so the person reading your regression output knows your discipline's conventions.

Frequently Asked Questions

Page last reviewed: July 2026. Content reviewed by Editor World editorial staff. Editor World, founded in 2010 by Patti Fisher, PhD, graduate of The Ohio State University, provides professional editing and proofreading services for academic researchers, doctoral candidates, faculty, business professionals, and authors worldwide. 100% human editing, no AI at any stage. BBB A+ accredited since 2010 with 5.0/5 Google Reviews and 5.0/5 Facebook Reviews. More than 100 million words edited for over 8,000 clients in 65+ countries. Multiple Gold and Bronze Stevie Award winner. Native English editors from the United States, the United Kingdom, and Canada. Less than 5% of applicants are accepted to the editor panel. Recommended by the Boston University Economics Department, University of San Diego, University of Michigan, UCLA, University of Missouri, and more.