Regression analysis is a statistical method that measures how one thing changes when another thing changes — for example, how much your exam score goes up for every extra hour you study. If you've ever wondered whether one variable actually predicts another (and by how much), regression is the tool that answers it with real numbers instead of a hunch. Most people meet it while staring at their own thesis data, unsure whether the pattern they think they see is real.
Key Takeaways
- Regression analysis predicts or explains one outcome variable (the dependent variable) using one or more predictor variables (independent variables).
- Simple linear regression uses one predictor; multiple regression uses two or more.
- The output tells you three things: whether the relationship is statistically significant (p), how strong it is (R²), and the size and direction of the effect (the b coefficient).
- An R² of .30 means your predictors explain 30% of the variation in the outcome.
- Regression assumes a roughly straight-line relationship, so always plot your data before trusting the numbers.
What does regression analysis actually do?
Regression analysis draws the "best-fit line" through a cloud of data points so you can describe and predict the relationship between variables. Imagine plotting hours studied on the horizontal axis and exam scores on the vertical axis. The dots scatter around, but they trend upward. Regression finds the single straight line that sits closest to all those dots — and that line becomes a prediction machine.
The line has two useful parts. The slope tells you how much the outcome changes for each one-unit increase in the predictor (e.g., "+4 points per study hour"). The intercept is the predicted outcome when the predictor is zero. Together they let you plug in a value and get a prediction.
In plain terms: regression turns "these two things seem related" into "for every extra X, you get this much more Y, and here's how confident we are."
When should I use regression analysis?
Use regression analysis when your outcome is a continuous number and you want to predict it or explain what drives it. Continuous means a measurable quantity — test scores, blood pressure, income, reaction time — not a category like "yes/no" or "group A/B."
Reach for regression when your research question sounds like one of these:
- Does sleep predict anxiety scores? (one predictor → simple linear regression)
- Do age, income, and education together predict life satisfaction? (several predictors → multiple regression)
- How much of the variation in job performance is explained by these factors?
If your outcome is a category (pass/fail, disease/no disease), you need logistic regression instead. If you only want to know whether two group means differ, a t-test or ANOVA is simpler — see our guide on choosing the right statistical test if you're not sure which camp you're in.
Simple vs. multiple regression: what's the difference?
The core difference is how many predictors you put into the model. Both answer the same style of question; multiple regression just controls for several variables at once so you can isolate each one's unique contribution.
| Feature | Simple Linear Regression | Multiple Regression |
|---|---|---|
| Predictors | One | Two or more |
| Question | Does X predict Y? | Do X₁, X₂, X₃ predict Y? |
| Key benefit | Easy to interpret and plot | Controls for confounders |
| Example | Sleep → anxiety | Sleep, workload, caffeine → anxiety |
| Reported statistic | b, R², p | b per predictor, R², F |
A practical tip: multiple regression is where most real research lives, because outcomes almost never depend on just one thing. Controlling for other variables is what lets you say a predictor matters over and above the others.
A worked example with real numbers
Let's walk through a simple linear regression step by step so the output stops looking like alphabet soup.
The study: We surveyed 60 undergraduates and measured weekly study hours and their final exam score (out of 100). We want to know: do study hours predict exam scores?
The result: We ran the regression and got:
F(1, 58) = 18.42, p < .001, R² = .24, b = 3.10, 95% CI [1.65, 4.55]
Here's what each piece means, in order:
- The relationship is significant. p < .001 is far below the .05 threshold, so the link between study hours and exam scores is very unlikely to be a fluke.
- The effect size: R² = .24 means study hours explain 24% of the variation in exam scores. The other 76% comes from everything else (prior knowledge, sleep, natural ability).
- The slope: b = 3.10 means each extra weekly study hour predicts a 3.1-point increase in exam score, on average.
- The confidence interval [1.65, 4.55] tells us the true increase is plausibly somewhere between 1.65 and 4.55 points — and because it doesn't cross zero, we're confident the effect is real.
So a student studying 5 hours more per week would be predicted to score roughly 15.5 points higher. That's the whole point of regression: it turns a vague "studying helps" into a concrete, defensible number.
What counts as a "good" R²?
There's no universal cutoff — a "good" R² depends entirely on your field. In psychology and social sciences, where human behaviour is messy, an R² of .10 to .30 is common and often meaningful. In tightly controlled physics or engineering, researchers might expect .90 or higher. Cohen's rough benchmarks label R² values of about .02 as small, .13 as medium, and .26 as large for behavioural research. Don't chase a high R² for its own sake — a statistically significant predictor with a modest R² can still be a genuine, publishable finding.
What assumptions does regression need?
Regression only gives trustworthy answers if a few conditions hold, so check these before you report anything:
- Linearity: the relationship should be roughly a straight line (plot it).
- Independence: each observation should be independent of the others.
- Homoscedasticity: the spread of errors should be fairly even across the line.
- Normality of residuals: the leftover errors should be roughly normally distributed.
Skipping these checks is the most common mistake beginners make. A single influential outlier can drag your best-fit line sideways and hand you a misleading slope. Tools like StatRyx run these assumption checks automatically and flag problems before they invalidate your conclusions.
How do I report regression in APA 7 format?
APA 7 wants the test statistic, degrees of freedom, p value, effect size, and — for each predictor — the unstandardised coefficient with its confidence interval. A clean write-up of our example reads:
A simple linear regression showed that study hours significantly predicted exam scores, F(1, 58) = 18.42, p < .001, R² = .24. Each additional study hour was associated with a 3.10-point increase in exam score, b = 3.10, 95% CI [1.65, 4.55].
Note the APA details: statistics are italicised, the p value drops its leading zero, and you report an effect size (R²), not just significance. Getting this notation right is exactly where hand-typed reports go wrong — a strlength StatRyx handles by generating the full APA sentence for you.
Running regression without the headache
You can run regression in SPSS, R, JASP, jamovi, or Stata — but each asks you to know which menu to click or which assumptions to test yourself. SPSS licences run roughly a few hundred dollars a year for students, R demands you write code, and JASP and jamovi are free but require a desktop install. StatRyx is an AI-powered statistical analysis tool that replaces manual SPSS workflows with automated, APA 7-formatted reporting — you upload your