Several statistical tools show whether two sets of values are related, how strongly, and in what direction. The most common is the correlation coefficient, a single number between -1 and +1 that captures how closely two variables move together. But correlation is just one option. Scatter plots, regression equations, and heatmaps each reveal different aspects of a relationship between values.
The Correlation Coefficient
The correlation coefficient is the go-to measure for quantifying a relationship between two sets of values. It ranges from -1 to +1, where zero means no linear relationship exists. A positive number means both variables rise together: as one goes up, the other tends to go up too. A negative number means they move in opposite directions.
The closer the coefficient gets to +1 or -1, the stronger the relationship. Here’s a general guide to interpreting the strength:
- 0.90 to 1.00: Very high correlation
- 0.70 to 0.90: High correlation
- 0.50 to 0.70: Moderate correlation
- 0.30 to 0.50: Low correlation
- 0.00 to 0.30: Negligible correlation
The same thresholds apply to negative values. A coefficient of -0.85 is just as strong as +0.85; it simply means the variables move in opposite directions.
There are two main types of correlation coefficient. The Pearson correlation measures strictly linear relationships, where two variables change at a roughly constant rate relative to each other. The Spearman correlation measures monotonic relationships, where two variables tend to move in the same direction but not necessarily at a constant rate. Spearman is also useful when your data involves rankings rather than precise measurements.
Scatter Plots: Seeing the Pattern
A scatter plot is the simplest visual way to spot a relationship between two sets of values. Each data point sits on a grid based on its pair of values, and the overall pattern of dots tells you what’s going on. If the dots climb from lower-left to upper-right, that’s a positive relationship. If they fall from upper-left to lower-right, it’s negative. If the dots look like a random cloud with no clear direction, there’s likely no meaningful relationship.
Scatter plots also reveal things a single number can’t. A correlation coefficient assumes a straight-line pattern, but your data might curve. Two variables could have a strong relationship that follows a U-shape or an exponential curve, and the correlation coefficient would underestimate it or miss it entirely. A scatter plot catches these non-linear patterns immediately.
Outliers, or extreme data points that sit far from the rest, also become obvious on a scatter plot. Even a single outlier can drag a correlation coefficient significantly higher or lower than it should be, especially in small datasets. Plotting your data first helps you decide whether the number you calculate actually reflects reality.
R-Squared: How Much Is Explained
Once you have a correlation coefficient, squaring it gives you a value called R-squared. This tells you the percentage of variation in one variable that’s accounted for by the other. If two variables have a correlation of 0.80, the R-squared is 0.64, meaning about 64% of the variation in one variable can be explained by changes in the other. The remaining 36% comes from other factors.
R-squared is especially useful because it translates an abstract number into something intuitive. Saying “the correlation is 0.50” is less immediately meaningful than saying “25% of the change in one value is tied to the other.” It also makes clear that even moderate correlations leave a lot unexplained.
Regression Lines and Prediction
While correlation tells you that a relationship exists, a regression line models the relationship as an equation you can use for prediction. The basic form is Y = a + bX, where X is the variable you’re using to predict, Y is what you’re predicting, b is the slope (how much Y changes for each unit of X), and a is the starting value when X equals zero.
A regression line drawn through a scatter plot shows the best-fit straight line through your data. The steeper the slope, the more Y changes as X changes. A flat slope means X has little predictive power over Y. Regression goes beyond simply showing that two values are related; it quantifies the relationship precisely enough to make estimates.
Correlation Heatmaps for Multiple Variables
When you’re working with more than two variables, checking them one pair at a time gets tedious. A correlation heatmap displays every possible pair in a color-coded grid. Each cell shows the correlation between two variables, with colors representing strength and direction. Deep reds or blues typically indicate strong positive or negative correlations, while pale colors or white signal weak ones.
Heatmaps are widely used in fields ranging from genetics to finance because they let you scan dozens of relationships at once and quickly spot which pairs of values are most closely linked. Many also include the actual correlation numbers inside each cell for precision.
Statistical Significance and P-Values
Finding a correlation doesn’t automatically mean the relationship is real. With small datasets or noisy data, random chance can produce what looks like a pattern. A p-value helps you judge whether the relationship you found is likely genuine or just a fluke.
The conventional threshold is p < 0.05, meaning there's less than a 5% chance the observed relationship occurred purely by random variation. Some fields set the bar much higher. Genetics research, for instance, often requires p-values below 0.00000001 to account for the sheer number of comparisons being made. A low p-value doesn't tell you the relationship is strong, only that it's unlikely to be imaginary.
Why Correlation Does Not Mean Causation
A strong statistical relationship between two sets of values does not prove that one causes the other. There are two core reasons this matters. The first is the third variable problem: a hidden factor might be driving both variables independently. The classic example is the correlation between ice cream sales and violent crime rates. Both rise in summer, but hot weather is the real cause of each, not ice cream somehow triggering crime.
The second issue is directionality. Even when two variables genuinely influence each other, a correlation alone can’t tell you which one is doing the influencing. If children who play more violent video games also show more aggressive behavior, the relationship could run in either direction, or both could stem from a third factor like the quality of parental attention.
Spurious correlations also crop up purely by coincidence, especially in large datasets where thousands of variable pairs are compared. The only way to establish that one value actually causes changes in another is through controlled experiments or carefully designed studies that rule out alternative explanations.