Correlation Analysis Concept

Direction and Strength of Linear Correlation

Suppose we record two numerical variables for the same observations, such as a student's study time and exam score or a car's age and selling price. A scatter plot displays the paired values and any pattern they form. Correlation analysis summarizes the direction and strength of a linear, or straight-line, pattern.

Correlation answers a descriptive question: how do the two variables vary together? It does not yet tell us why the pattern exists.

The Correlation Coefficient

Pearson's correlation coefficient is denoted by $r$ . The sign of r describes direction, while its distance from zero describes the strength of the linear pattern.

Value	Interpretation
$r>0$	Larger values of one variable tend to occur with larger values of the other.
$r<0$	Larger values of one variable tend to occur with smaller values of the other.
$\lvert r\rvert\approx1$	The points lie close to a straight line.
$r\approx0$	The measured linear association is very weak or zero.

Whenever both variables have nonzero variation and $r$ is defined, the coefficient lies in this interval:

-1\le r\le1

$r=1$ means a perfect positive linear correlation.
$r=-1$ means a perfect negative linear correlation.
$r=0$ means no linear correlation.

A value near zero does not rule out every relationship. Points may follow a clear curve while their linear correlation remains small. The scatter plot and the context must therefore be read together with $r$ .

Swapping Variables and Changing Units

Correlation treats the two variables symmetrically. Swapping the horizontal and vertical variables does not change its value:

r_{xy}=r_{yx}

Changing a positive unit scale also leaves $r$ unchanged. Measuring height in centimeters or meters produces the same coefficient. This differs from regression, where one variable is deliberately used to predict the other.

The Coefficient of Determination

The square of the correlation coefficient, $r^2$ , is called the coefficient of determination. In simple linear regression with an intercept, it gives the proportion of variation in the response that is accounted for by the fitted linear model.

If study time and exam score have $r=0.8$ , then:

r^2=(0.8)^2=0.64

The linear association accounts for $64\%$ of the observed variation in scores in that data set. The remaining $36\%$ is not accounted for by this linear relationship. This percentage describes how well the model fits the data. It cannot show that study time alone caused the score differences.

The coefficient of determination satisfies:

0\le r^2\le1

Correlation and Causal Relationships

A strong correlation does not establish that one variable causes the other. Several explanations can produce the same pattern:

the first variable may influence the second.
the second may influence the first.
a third variable may influence both.
sampling, measurement, or an unusual observation may distort the pattern.

For example, ice cream sales and drowning incidents may both rise during hot weather. The correlation does not mean that buying ice cream causes drowning. Summer weather and increased swimming activity affect both observations.

Use correlation to describe a linear pattern. A causal conclusion needs additional evidence from the study design, temporal order, plausible mechanisms, and alternative explanations.

Check Your Interpretation

Suppose a data set has $r=-0.9$ . This indicates a strong negative linear relationship: larger values of one variable tend to accompany smaller values of the other. It does not, by itself, identify a cause or guarantee that every point lies on the same line.