Psychology · Foundations
A correlation A measure of the degree to which two variables move together — the extent of their relationship. Full entry → is a measure of how two variables move together. Two things are correlated when they rise and fall in step — same direction, opposite directions, or with no pattern at all. The tighter the pattern, the stronger the correlation, and a scatterplot A graph in which each dot represents one data point, positioned by its values on two variables. Full entry → is its picture. Correlations are useful for prediction Using the value of one variable to anticipate the value of a related variable, based on how they move together. Full entry → and for generating hypotheses, but a link between two variables is not proof that one causes the other. Correlation is a clue, not a verdict.
Correlations are everywhere in daily life — in news headlines, app reports, health trackers, and product claims — and most people never learn to read them. A sleep app that shows later bedtimes alongside more evening screen time is reporting a correlation, and so is a headline that ties one habit to some outcome. The skill is knowing what the pattern does and does not say: a correlation can point you toward a prediction and spark a question worth studying, but it cannot by itself tell you what causes what. Psychology courses build on this idea constantly — every claim about the relationships between behavior, thought, and biology passes through it.
The APA Dictionary of Psychology defines correlation as the degree of a relationship between two variables. In plainer words, two things are correlated when they move together: a change in one tends to come with a change in the other. Consider a fictional sleep-tracking app called Nightowl, which records two numbers per user each night — minutes of screen time after 9 p.m. and the time that person fell asleep. Sorted by screen time, the pattern jumps out: nights with more evening screen time come with later bedtimes, and nights with little screen time come with earlier ones. The two variables move together, and that co-movement is exactly what a correlation measures. Correlation always describes a pair of variables at a time — this and that, not the whole picture at once.
A correlation has a direction: the story of how the two variables relate. A positive correlation A pattern in which two variables move in the same direction, rising and falling together. Full entry → means they move in the same direction — as one goes up, the other tends to go up too, and as one falls, the other tends to fall. The sleep-app link is positive: more screen time, later bedtime. A negative correlation A pattern in which two variables move in opposite directions: as one increases, the other tends to decrease. Full entry → means they move in opposite directions — as one goes up, the other tends to go down. Original example: a running coach finds that athletes who take more rest days have fewer injuries — rest up, injuries down. A zero correlation A situation in which two variables show no consistent pattern; changes in one carry no information about the other. Full entry → — more precisely, near-zero — means no consistent pattern: the variables do not move together. The same app records shoe size, which has no relationship with bedtime. One caution from OpenStax: a negative correlation is not the same as no correlation. Opposite directions are still a relationship; no relationship means no pattern of any direction.
Direction tells you which way the variables move; strength tells you how reliably. A strong correlation is one where the pattern holds almost every time. A weak one holds sometimes — the variables lean in the same general direction, but exceptions are common. In the Nightowl data, screen time and bedtime move together on most nights but not all: a tired user might scroll for an hour and still fall asleep early. So the pattern is real but not perfect. When researchers put a number on a correlation, it runs from -1 to +1: the closer the number sits to either end, the more strongly the variables are related, and the more predictable one becomes from the other. A number near zero means little or no pattern. The honest note: the arithmetic behind that number is the statistics subject's territory, not this lesson's.
A scatterplot draws the pattern so you can see it. Each dot is one person, one night, or one anything, carrying two pieces of information: left-to-right is the value of one variable Any characteristic or quantity that can take different values across people, events, or time. Full entry →, up-down is the value of the other. Dots that climb from the lower-left corner to the upper-right corner show a positive correlation; dots that slope down from upper-left to lower-right show a negative one. The shape of the whole cloud is the story. When the dots hug a line, the correlation is strong; when they scatter loosely around it, it is weak. If the dots form no shape at all, the variables are essentially unrelated. In one picture, a scatterplot answers two questions: which direction, and how tightly.
Here is the boundary that matters. What a correlation can show: that two variables are linked — that they move together in a consistent way. What it cannot show: why they move together. OpenStax puts it directly: establishing that a relationship exists tells us little about cause and effect. Consider the sleep-app link: it could mean screen time delays sleep, or that people who naturally stay up late reach for their phones while they wait for drowsiness. Or a third factor — a demanding work schedule, a noisy household — could be pushing both at once. The data, taken alone, cannot choose among these stories, because every one of them produces the same pattern. The deeper question — how researchers ever do decide what causes what — is its own lesson. Here, hold the boundary: correlation shows links, not causes.
A correlation earns its keep even when it explains nothing. First, prediction: when two variables move together, knowing one helps you anticipate the other. OpenStax notes that correlations have predictive value — an admissions office might use the link between high-school grades and first-year performance to anticipate which applicants will succeed. Second, hypothesis A candidate explanation, stated clearly enough to be tested with evidence. Full entry → generation: a pattern suggests possible explanations, and each possible explanation is a question worth testing. The Noba Project makes the point with the generosity-and-happiness example: the correlation could mean generosity causes happiness, or happiness causes generosity, or a third variable drives both. A researcher who sees the sleep-app pattern now has a testable question — does cutting evening screen time actually move bedtimes earlier? Testing it is a different method's job, covered in its own lesson.
Correlation is a clue, not a verdict. It tells you where to look, and it never closes the case. The pattern in the data is real and useful — for predicting, for raising questions, for narrowing the possibilities. But the pattern alone cannot prove any explanation, no matter how obvious the story seems. Read correlations the way a careful detective reads a clue: note it, let it point you somewhere, and keep looking for the test that decides between the stories.
Eli explains
Explain it like I’m 10
A correlation is a way of saying that two things move together. Some pairs rise and fall in step — more screen time, later bedtime: that is a positive correlation. Some move in opposite directions — more rest days, fewer injuries: that is a negative one. And some show no pattern at all: near-zero. Correlations also have strength: a strong one holds almost every time, a weak one only sometimes. A scatterplot shows all of this in one picture, one dot per person. Correlations are handy — they let you predict, and they hand you questions to test — but they never say what causes what. The pattern is a clue; the explanation still has to be earned.
Think of a wooden dock floating on the water. When the tide rises, the dock rises; when the tide falls, the dock falls. Watch the dock and you can predict the tide's direction quite well — yet the dock does not cause the tide, and the tide does not lift the dock. Something else, the moon's gravity, moves both. That is what a correlation is: a real, useful, visible link — with the true driver standing off to the side.
The dock and the tide are almost perfectly synchronized — a very strong correlation — while most real correlations are loose, with plenty of exceptions. And in the tide example the hidden cause is famous and easy to name; in real research, third variables are usually invisible and hard to hunt down, which is why the honest lesson ends where it does: a clue, not a verdict.
Priya opens the fictional Nightowl sleep app's anonymized dataset: 10,000 users, two numbers per person per night. She plots evening screen time on one axis and bedtime on the other, one dot per user. The dots form a loose band climbing to the right — positive, moderate strength: plenty of people fit the pattern, plenty do not. Next she plots shoe size against bedtime: the dots go everywhere, no shape at all — near-zero. Then she reads a forum post claiming the app proves that screen time causes late sleep. She stops: the data show a link, not a cause. The pattern can predict — a heavy scroller is more likely to be a late sleeper — and it generates a testable question: does reducing evening screen time move bedtimes earlier? But answering that question needs a method the app's data cannot supply. She writes her summary: "Screen time and bedtime move together in this dataset. That is a correlation — a useful clue, not a verdict."
A correlation measures how two variables move together — same direction, opposite directions, or not at all — and a scatterplot shows the pattern at a glance. It is a clue, not a verdict: useful for prediction and hypothesis generation, never proof of cause.
3 questions here, of 5 in this lesson’s practice set. Answers stay hidden until you check.
"Negative correlation means there is no relationship."
Negative is a direction, not an absence. Variables that move in opposite directions are related — sometimes strongly. No relationship lives near zero, in the middle of the range, where no pattern of any direction shows up.
"The stronger the correlation, the more likely one variable causes the other."
Strength says how tightly the pattern holds, not why it holds. A very tight pattern can still be driven by a hidden third factor, so strength never upgrades a correlation into a cause.
"One exception disproves a correlation."
Correlations describe patterns across many data points, and exceptions are normal. Strength is a matter of how many exceptions there are, not whether any exist.
"If two things are correlated, one must be causing the other."
A link is not proof of a cause. Usually several explanations fit the same pattern — the sleep-app link, for example, has at least three plausible stories. Deciding among them requires other research methods, which this subject covers in separate lessons.
Both are real relationships; they differ only in direction — same direction versus opposite directions. In both cases, knowing one variable helps anticipate the other.
A correlation is a measured link between variables; causation is one variable producing a change in another. A correlation never establishes causation on its own — that question is its own lesson.
Strength is how tightly the pattern holds: a strong correlation shows few exceptions, a weak one shows many. Direction and strength are independent — a strong negative correlation is just as tight as a strong positive one.
EliExplains lessons are original prose written from the open, credible references above. See Copyright & Licensing.
Researched 2026-08-22
Educational content only. It is not medical, legal or professional advice. Found an error? Tell us.
