A bivariate correlation is a statistical measure of how two variables move in relation to each other. It produces a single number, called a correlation coefficient, that falls between -1 and +1. A value of 0 means no relationship exists, while values closer to -1 or +1 indicate a stronger connection. If you’ve ever seen a claim like “income is correlated with education level,” a bivariate correlation is the math behind that statement.
How Two Variables Relate
The word “bivariate” simply means “two variables.” In this type of analysis, you’re looking at how changes in one variable line up with changes in another. Sometimes one variable is treated as the outcome and the other as the explanation, like examining whether hours of exercise predict resting heart rate. Other times, neither variable is assumed to cause the other, and you’re just measuring whether they tend to rise or fall together.
A positive correlation means both variables move in the same direction. As one increases, the other tends to increase too. Height and weight in adults is a classic example. A negative correlation means the variables move in opposite directions: as one goes up, the other tends to go down. Think of outdoor temperature and heating bills. A correlation near zero means knowing the value of one variable tells you essentially nothing about the other.
The Pearson Correlation Coefficient
The most common way to measure bivariate correlation is the Pearson correlation coefficient, represented by the letter “r.” It’s a dimensionless number scaled from -1 to +1. At r = +1, every data point falls on a perfectly straight upward-sloping line. At r = -1, every point falls on a perfectly straight downward-sloping line. At r = 0, there’s no linear pattern at all, and the data points look like a shapeless cloud on a graph.
Pearson correlation specifically measures linear relationships, meaning it detects how well a straight line fits the data. If two variables have a curved relationship (say, job performance increases with stress up to a point, then drops), the Pearson coefficient can underestimate or miss the connection entirely.
For the Pearson coefficient to be reliable, your data needs to meet a few conditions. The relationship between the two variables should be roughly linear, which you can check by plotting the data on a scatterplot. Both variables should be continuous (measured on a numerical scale, not categories like “yes” or “no”). And the data pairs should follow a roughly normal, bell-curve distribution. When these conditions aren’t met, the coefficient can be misleading.
How to Read the Strength of a Correlation
Not all correlations are equally meaningful, and knowing the number alone doesn’t tell you whether it’s “strong” or “weak” without some frame of reference. The most widely cited guidelines come from the statistician Jacob Cohen, who in 1988 suggested that r = 0.10 represents a small correlation, r = 0.30 a medium one, and r = 0.50 a large one.
However, those benchmarks may be too generous. A large-scale review of over 700 published meta-analyses found that the 25th, 50th, and 75th percentiles of real-world correlations landed at roughly 0.10, 0.20, and 0.30. In other words, most correlations researchers actually find in practice are smaller than Cohen’s “medium” threshold. Revised guidelines now suggest treating 0.10 as small, 0.20 as typical, and 0.30 as relatively large, at least for research involving individual differences like personality traits, cognitive abilities, or health behaviors.
This matters because people tend to overestimate how strong correlations should be. A correlation of 0.25 between a parenting style and a child outcome might sound underwhelming, but in context it could be among the stronger effects in that field.
Shared Variance: What R-Squared Tells You
One of the most useful things you can do with a correlation coefficient is square it. The result, called r-squared (r²), tells you the proportion of variation in one variable that’s accounted for by the other. If the correlation between sleep duration and test scores is r = 0.40, then r² = 0.16, meaning about 16% of the variation in test scores can be explained by differences in sleep. The remaining 84% comes from other factors.
This reframes correlations in a way that can be eye-opening. Even a correlation that sounds reasonably strong, like r = 0.50, only explains 25% of the variation. A “medium” correlation of r = 0.30 explains just 9%. Squaring the coefficient is a quick reality check on how much predictive power a relationship actually has.
Visualizing Correlations With Scatterplots
A scatterplot is the standard way to visualize a bivariate correlation. Each dot represents one observation, with one variable on the horizontal axis and the other on the vertical axis. A positive correlation shows dots trending from the lower left to the upper right. A negative correlation shows dots sloping from the upper left to the lower right. No correlation looks like a random scatter with no discernible pattern.
Scatterplots are especially valuable because they reveal things a single number cannot. You might spot outliers pulling the correlation in one direction, or notice a curved pattern that the Pearson coefficient would miss. Before interpreting any correlation coefficient, plotting the data first helps you avoid drawing conclusions from a number that doesn’t reflect the real shape of the relationship.
When to Use Spearman Instead of Pearson
The Pearson coefficient assumes your data is normally distributed and the relationship is linear. When either assumption is violated, the Spearman rank correlation is the go-to alternative. Instead of working with the raw values, Spearman converts each data point to its rank (1st, 2nd, 3rd, and so on) and then calculates the correlation between those ranks.
This makes Spearman useful in several situations: when your data has extreme outliers that would distort a Pearson calculation, when the relationship between variables is consistently increasing or decreasing but not in a straight line, or when you’re working with ordinal data like survey ratings (e.g., satisfaction on a 1-to-5 scale). Like Pearson, Spearman also ranges from -1 to +1, but it detects any monotonic relationship, not just linear ones. If you find a weak Pearson correlation but a strong Spearman correlation for the same data, that’s a signal the relationship exists but follows a curve rather than a straight line.
Statistical Significance vs. Practical Importance
When you run a bivariate correlation, most software will also report a p-value, which indicates the probability of seeing a correlation that strong (or stronger) if no real relationship existed. A p-value below 0.05 is the conventional cutoff for calling a result “statistically significant,” but this label is frequently misunderstood.
Statistical significance does not mean the correlation is large or important. Any effect, no matter how tiny, can produce a small p-value if the sample is large enough. A study with 10,000 participants might find a correlation of r = 0.03 that’s statistically significant, yet that relationship explains less than 0.1% of the variation between the two variables. It’s real in the mathematical sense, but practically meaningless. The size of the correlation and its r-squared value tell you far more about whether the relationship matters than the p-value alone.
Correlation Does Not Mean Causation
This is the most important caveat in all of statistics, and it applies directly to bivariate correlation. Finding that two variables move together does not tell you that one causes the other. There are at least three possible explanations for any observed correlation: variable A causes changes in B, variable B causes changes in A, or some unmeasured third factor drives both.
A classic example: mood and physical health are correlated in adults. But does better mood lead to better health? Does better health improve mood? Or does a third factor, like financial stability or strong social connections, drive improvements in both? The correlation alone cannot answer that question. It can serve as evidence that a relationship exists and is worth investigating further, but establishing causation requires controlled experiments or more advanced statistical methods that go well beyond a simple two-variable analysis.