How to Read a Pearson Correlation Coefficient (r) Without Fooling Yourself

A single number that runs from -1 to 1 gets thrown around in research papers, marketing decks, and homework problems as if it settles arguments. It does not. The Pearson correlation coefficient, written as r, is a useful summary of how two columns of numbers move together, but it is also one of the easiest statistics to misread. This guide walks through what r actually measures, how to interpret its sign and size, why a strong r still does not prove cause, and why you should never trust the number without looking at the picture behind it.

What r Measures, and Why It Sits Between -1 and 1

Pearson r captures the strength and direction of a straight-line relationship between two variables. Feed it paired data — say each student's study hours and their exam score — and it returns one value that can never leave the band from -1 to +1.

The two ends of that band are the extremes:

r = +1 is a perfect positive line: every time X goes up, Y goes up by a proportional amount, with no scatter at all.
r = -1 is a perfect negative line: every time X goes up, Y goes down, again with no scatter.
r = 0 means there is no straight-line relationship at all. The points are a cloud with no upward or downward tilt.

Most real data lands somewhere in between. The reason r is bounded is baked into its formula: it divides the covariance of X and Y by the product of their standard deviations, which normalizes the result onto that fixed scale. That normalization is what makes r comparable across wildly different datasets — a correlation of 0.6 between rainfall and crop yield means the same degree of linear tightness as a 0.6 between ad spend and revenue, even though the raw units share nothing.

Sign Is Direction, Size Is Strength

The single most common mistake I see is conflating the minus sign with weakness. They are two separate pieces of information:

The sign (+ or -) tells you the direction. Positive means the variables rise together; negative means one rises as the other falls.
The magnitude (how close to 1, ignoring the sign) tells you the strength of the linear fit.

So r = -0.9 is a very strong relationship, not a weak one. The minus only says "X up, Y down." Meanwhile a cheerful-looking r = 0.1 is positive but almost meaningless — the points barely tilt. When you judge strength, look at the absolute value and leave the sign out of it.

A widely cited rule of thumb (Evans, 1996), applied to the absolute value of r, gives a rough vocabulary:

below 0.20 — very weak
0.20 to 0.40 — weak
0.40 to 0.60 — moderate
0.60 to 0.80 — strong
0.80 and up — very strong

These bands are conventions, not laws of nature, but they keep your language honest. Calling r = 0.35 a "strong link" is overselling; calling r = 0.85 "a slight trend" is underselling.

A Worked Example: Interpreting r = 0.85

Suppose a sales team pulls twelve months of data: monthly advertising spend (X) against revenue (Y). They paste the pairs into a correlation coefficient calculator and get r = 0.85.

Here is how to read that in full:

Direction: the sign is positive, so higher spend goes with higher revenue. No surprise, but worth confirming — a negative r here would have been a serious red flag.
Strength: 0.85 sits in the "very strong" band. The months line up tightly along an upward trend.
Explained variation: square it. r² = 0.85 × 0.85 ≈ 0.72. That coefficient of determination says roughly 72 percent of the month-to-month swing in revenue tracks with the straight-line fit on ad spend. The other 28 percent comes from everything the model ignores — seasonality, pricing, a competitor's stumble, plain luck.

That last conversion is why r² is so useful in a meeting. "Correlation of 0.85" is abstract; "ad spend accounts for about 72 percent of the revenue swing" is a sentence a finance lead can act on. Just keep the caveat attached: 72 percent of the variation is associated with spend, which is not the same as 72 percent of revenue being caused by it.

Why Correlation Is Not Causation

This is the line everyone has heard and almost everyone still violates. A high r tells you two numbers move in step. It says nothing about why.

The classic example: across a summer, ice cream sales and drowning deaths rise and fall together, easily producing a strong positive r. Ice cream does not drown people. A hidden third variable — hot weather — drives both. More people buy ice cream when it is hot, and more people swim when it is hot, so the two harmless behaviors correlate without one causing the other.

There are three usual suspects whenever a strong correlation tempts you toward a causal story:

A lurking third factor drives both variables (the ice cream case).
Reverse causation — maybe Y causes X, not the other way around.
Coincidence — with enough variable pairs, some will correlate by pure chance.

To earn a causal claim you need a controlled experiment, or at minimum a careful causal argument that rules these out. The r value alone never gets you there, no matter how close to 1 it climbs.

Always Look at the Scatterplot

A scatterplot — X on one axis, Y on the other, one dot per pair — is the cheapest insurance against being fooled by r, because the single number hides things the picture reveals instantly.

Two failure modes matter most:

Outliers. One extreme point can drag r far from where the bulk of the data sits. A cloud of points with essentially no trend can show r near 0.7 because of a single far-flung dot. Remove or question that point and the correlation evaporates.
Curves. Pearson r only measures straight-line strength. A clean, obvious curve — points arching up and then back down — can score r near zero even though the two variables are tightly related. The relationship is real; it just is not linear, and r is blind to it.

Once I started plotting first and reading r second, my interpretations got far more reliable. I had a dataset where two variables showed r = 0.05 and I almost wrote them off as unrelated. The scatterplot showed a crisp U-shape: a strong relationship that Pearson scored as noise because it only sees lines. The number would have sent me the wrong way entirely; the chart fixed it in two seconds. Now the plot is the first thing I check, every time.

Putting It Together

A correlation coefficient is a compact, honest summary if you respect its limits. Read the sign for direction and the magnitude for strength, square it to talk about explained variation, refuse to leap to causation, and let the scatterplot catch the outliers and curves that r cannot. If you want the matching descriptive figures — means, standard deviations, and spread for each column — pair this with a statistics calculator so you see the full shape of your data, not just the one number that connects the two variables.

Treat r as the start of an investigation, not the verdict, and it will rarely steer you wrong.

Made by Toolora · Updated 2026-06-13