Ever looked at a stats problem and thought, "Wait, which of these things about r are actually true?That said, the correlation coefficient r gets tossed around in textbooks, dashboards, and blog posts like everyone agrees on what it means. " You're not alone. They don't.
Here's the thing — most people memorize a couple of rules about r and call it a day. Then they get burned on a test, or worse, in real data work, because they confused correlation with causation or forgot r only catches straight-line relationships. So let's sort the real statements about the correlation coefficient r from the noise.
What Is the Correlation Coefficient r
Picture two variables — say, hours studied and exam score. On the flip side, the correlation coefficient r is a single number that summarizes how tightly those two things move together, and in what direction. It's a built-in scoreboard for linear association.
It always lands between -1 and 1. That's not a suggestion. It's math. An r of 1 means a perfect positive straight-line relationship. An r of -1 means a perfect negative one. Zero means no linear relationship — and notice I said linear. That word does a lot of heavy lifting.
The sign tells you direction, not strength of "good"
A negative r isn't "bad.In real terms, " It just means as one variable goes up, the other tends to go down. Think about it: ice cream sales and coat sales might have a negative r in many cities. Nobody's complaining about the coats.
r is unit-free
You can correlate height in inches with weight in kilograms and get the same r as if both were in centimeters and pounds. That's why r is so portable across studies. It doesn't care about your units.
Why It Matters
Why does this matter? Because most people skip the fine print and then trust r too much. Even so, i've seen marketing reports claim "strong correlation" off an r of 0. Even so, 35. That's not strong. That's a weak murmur Small thing, real impact..
When you can identify the true statements about the correlation coefficient r, you stop falling for three big traps. And you won't say r proves one thing causes another. You won't wave off a curved relationship just because r is near zero. And you'll catch outliers that quietly hijack the value.
In practice, this stuff shows up everywhere — finance, health research, sports analytics, even dating app algorithms. A wrong belief about r can lead to a bad model, a silly headline, or a failed assignment. Real talk: the people who understand r's limits are the ones who don't get fooled by a pretty scatterplot Nothing fancy..
How It Works
So how do you actually tell true from false when someone hands you a list of statements about r? You go back to what r is and isn't. Let's break it down.
r measures linear association only
At its core, the big one. If the true relationship is a curve — like Y = X² — the correlation coefficient r can be close to zero even when the variables are perfectly related in a non-linear way. True statement: r near 0 does not mean no relationship. It means no straight-line relationship That's the part that actually makes a difference. Nothing fancy..
The boundaries are fixed
True: -1 ≤ r ≤ 1. If someone says "r can exceed 1 if the link is really strong," that's false. It can't. The formula normalizes it.
r does not imply causation
Classic, but still missed. True statement: a high |r| does not mean changes in X cause changes in Y. In real terms, there could be a third variable. Or coincidence. Or reverse causation. Knowing this separates analysts from amateurs.
r is symmetric
True: the correlation between X and Y is the same as Y and X. You don't get a different r by swapping the axes. That's not true for regression slopes, by the way, which is a related but different beast.
Outliers distort r
One weird point can drag r from 0.This leads to 2 to 0. 8 or slam it negative. True statement: r is sensitive to unusual observations. That's why you look at the scatterplot before quoting the number.
r² is the share of variance explained
If r = 0.Think about it: 5, then r² = 0. 25. True: about 25% of the variance in one variable is linearly associated with the other. People forget to square it and say "50% explained," which is wrong.
Sample size changes reliability, not the math
A correlation of 0.That's why 9 from 5 data points is shakier than 0. 9 from 5,000. True statement: r itself doesn't tell you if it's statistically significant. You need the sample size or a p-value for that That's the part that actually makes a difference..
Common Mistakes
Here's what most guides get wrong — they list "facts" about r without showing the boundary cases. Let me hit the ones that trip people up constantly Nothing fancy..
First, the "zero r means nothing's going on" mistake. So naturally, turns out, a U-shaped relationship gives r ≈ 0. In real terms, students stare at a perfect curve and swear the variables are unrelated because r said so. They aren't Simple, but easy to overlook..
Second, the causation leap. r is high. Two stocks move together during a crisis. I know it sounds simple — but it's easy to miss in the moment. Doesn't mean one caused the other's move Still holds up..
Third, ignoring outliers. Remove the typo, it's +0.1. Someone runs a correlation on 30 points, one is a typo, and they report r = -0.6. The true statement about r here is that it's not dependable. You must eyeball the data Not complicated — just consistent..
Fourth, thinking stronger |r| always means a better model. No. Here's the thing — if the real world is curved, a linear r of 0. 1 with a good non-linear fit beats an r of 0.7 on the wrong model type. Context wins.
Practical Tips
Want to actually get this right, whether on an exam or in a notebook? Here's what works.
Look at the scatterplot first. Always. Before you trust any r, see the shape. If it's not a cloud roughly along a line, r is incomplete news Simple, but easy to overlook..
Square r when anyone says "explained." If a boss says "70% correlated," ask for r. If they mean r = 0.7, that's 49% of variance, not 70%. Worth knowing And it works..
Use plain language for sign. So negative = opposite. Positive r = same direction. Don't let the minus sign scare you into calling it weak.
Check for outliers by running r with and without suspicious points. And if it swings, say so. That's the honest move.
Remember r is about association, not prediction quality by itself. For prediction, you'll want regression and residuals, not just r.
And on tests — when they ask you to identify the true statements about the correlation coefficient r, scan for these winners: bounded by -1 and 1, measures linear only, symmetric, sign is direction, sensitive to outliers, doesn't imply cause, r² is variance share. Those are your true hits.
FAQ
Can r be greater than 1? No. The correlation coefficient r is always between -1 and 1 inclusive. Anything outside that range means a calculation error or a different statistic That's the whole idea..
If r is 0, does that mean the variables are independent? Not necessarily. It means there's no linear relationship. They could have a strong curved or non-linear association and still show r near 0.
Does a negative r mean a weak relationship? No. The sign is direction only. An r of -0.9 is a very strong negative linear relationship. Strength comes from how close |r| is to 1 Not complicated — just consistent..
Is r affected by changing units? No. Correlation is unit-free. Converting from meters to feet won't change r Simple, but easy to overlook..
Does high correlation mean X causes Y? No. Correlation does not imply causation. A third variable, coincidence, or reverse causation could explain a high r.
The next time a list of "facts" about r lands in front of you, slow down and check the boundaries, the shape of the data, and the words they use. Most false statements about the correlation coefficient r die the moment you ask: is this about a straight line, or just any relationship? Get that right, and you're ahead of most people quoting numbers they don't fully understand Small thing, real impact. That alone is useful..
