A Low Correlation Can Hide a Clear Pattern
In “Correlation Requires Quantitative Variables,” you saw that \(r\) summarizes the direction and strength of the linear association between two quantitative variables. That restriction matters: a relationship can be clear and strong without following a straight-line pattern. When a scatterplot curves, \(r\) may be near zero even though the points follow a systematic pattern.
A low \(r\) is therefore not enough to conclude that two variables are unrelated. It says that the data show little linear association. To judge whether there is another kind of relationship, inspect the scatterplot and describe its form, as you practiced in “Recognizing Linear and Nonlinear Form” and “Describing a Scatterplot With DUFS.”
Why a Curve Can Produce \(r=0\)
The correlation \(r\) reflects whether larger values of one variable tend to pair with larger values of the other, or with smaller values, in an overall straight-line pattern. In a curved pattern, the response may rise over part of the range and fall over another part. Those opposite linear tendencies can balance when summarized by one number.
A symmetric U-shaped pattern makes this especially easy to see. On the left side of the curve, \(y\) falls as \(x\) increases toward the center. On the right side, \(y\) rises as \(x\) increases away from the center. The two sides can cancel in the calculation of \(r\), even when the curve is perfectly clear.
Recall from “Calculating \(r\) From Standardized Values” that \(r\) is based on paired deviations from the two means. One equivalent calculation uses \(S_{xy}\), the sum of the products of those paired deviations:
When observations on opposite sides of the center have equal \(y\)-values, their paired deviation products may cancel. Then \(S_{xy}=0\) and \(r=0\). This cancellation describes the linear summary, not the whole shape of the data.
Worked Examples
Worked Example: A Clear Curve With Exactly Zero Correlation
A fictional greenhouse records daily light exposure and a plant-growth score. The explanatory variable \(x\) is the number of hours of light above or below a target amount; negative values mean fewer hours than the target. The response \(y\) is a growth score in points. Here are seven invented observations:
| Light difference, \(x\) (hours) | Growth score, \(y\) (points) |
|---|---|
| -3 | 2 |
| -2 | 6 |
| -1 | 10 |
| 0 | 12 |
| 1 | 10 |
| 2 | 6 |
| 3 | 2 |
Find \(r\), then explain what the number does and does not say about the relationship.
Calculate the means. The \(x\)-values sum to zero, so \(\bar{x}=0\). The \(y\)-values sum to \(48\), so \(\bar{y}=48/7\approx 6.857\) points.
Find the paired-deviation sum. Since \(\bar{x}=0\), \(S_{xy}\) is the sum of \(x(y-\bar{y})\), which equals the sum of \(xy\) because the \(x\)-values sum to zero. The paired products cancel:
For completeness, \(S_{xx}=9+4+1+0+1+4+9=28\). The squared deviations of \(y\) sum to \(S_{yy}=4648/49\approx94.857\). Thus
The scatterplot would show a clear, symmetric inverted-U pattern: the growth score is highest at the target light amount and lower on either side. So \(r=0\) does not mean the variables have no relationship. It means the data have no overall linear association as summarized by \(r\). In context, the observed growth scores are related to light difference in a curved way.
Worked Example: A Near-Zero Correlation With a Curved Pattern
A fictional building manager records the difference between the indoor temperature and a preferred target, \(x\), in degrees, and daily heating-and-cooling energy use, \(y\), in kilowatt-hours. Energy use tends to be higher when the temperature is far below or far above the target. The invented observations are:
| Temperature difference, \(x\) (degrees) | Energy use, \(y\) (kWh) |
|---|---|
| -3 | 10 |
| -2 | 5 |
| -1 | 2 |
| 0 | 1 |
| 1 | 3 |
| 2 | 6 |
| 3 | 11 |
Calculate \(r\) and decide whether describing this association as weak is sufficient.
Calculate the means and sums. The \(x\)-values sum to zero, so \(\bar{x}=0\), and \(S_{xx}=28\). The \(y\)-values sum to \(38\), so \(\bar{y}=38/7\approx5.429\) kWh. As before, because the \(x\)-values sum to zero, \(S_{xy}=\sum xy\):
The sum of the squared \(y\)-values is \(296\). Therefore,
Substitute into the correlation formula:
The correlation is near zero, but the plotted points would still show a pronounced U-shaped tendency: energy use is low near the target temperature and higher toward the extremes. Because the pattern is slightly asymmetric, the correlation is not exactly zero. Its small positive value does not capture the curve well. A complete description should say that the linear association is weak while also identifying the curved pattern in context.
Worked Example: Correcting the Claim “There Is No Relationship”
A fictional greenhouse compares the difference from a preferred humidity level, \(x\), in percentage points, with a plant-health score, \(y\), in points. For nine invented days, the data follow this pattern:
| Humidity difference, \(x\) (percentage points) | Plant-health score, \(y\) (points) |
|---|---|
| -4 | 58 |
| -3 | 70 |
| -2 | 82 |
| -1 | 92 |
| 0 | 98 |
| 1 | 92 |
| 2 | 82 |
| 3 | 70 |
| 4 | 58 |
A student calculates \(r=0\) and writes, “Humidity difference and plant health have no relationship.” Evaluate the statement.
Check the pattern, not just the number. The \(x\)-values are centered at zero, and matching values on either side of zero have the same plant-health score. Their paired-deviation products cancel, so \(S_{xy}=0\) and \(r=0\). But the scores increase toward the preferred humidity level and then decrease as the difference grows in the other direction. That is a clear, symmetric curved pattern.
Improve the conclusion. The student should not say there is no relationship. A more accurate statement is: “For these observed days, the linear association between humidity difference and plant-health score is zero, but the scatterplot shows a strong curved association: scores are highest near the preferred humidity level and lower when humidity differs more.” This distinguishes what \(r\) summarizes from what the scatterplot shows.
How to Interpret a Low \(r\) Carefully
When \(r\) is close to zero, keep the conclusion specific. It is reasonable to say that the observed data show little linear association between the two quantitative variables. It is not reasonable to leap from that statement to “the variables are unrelated” without checking the form of the scatterplot.
A useful sequence is to examine the scatterplot first, then interpret the correlation in light of the pattern. This follows the broader principle from “What the Correlation Coefficient Measures”: \(r\) is a summary of linear association, not a complete description of every feature of bivariate data.
Decide whether the points follow a roughly straight pattern, a curve, or no clear pattern.
Use its sign and magnitude to describe the direction and strength of the linear association, as in “Interpreting a Correlation in Context.”
If the plot curves, say that a low \(r\) does not capture the visible nonlinear pattern. Describe that pattern using the actual variables.
This does not mean \(r\) is useless whenever a relationship is nonlinear. It means its answer addresses a narrower question: how strong and in what direction is the linear association? The scatterplot answers a broader descriptive question about the form of the relationship.
Common Mistakes and AP Exam Tips
- Writing “there is no relationship” because \(r\approx0\). A low correlation is evidence of little linear association, not proof that no pattern exists. Check the scatterplot for a curve or other structure.
- Calling a curved pattern weak because the correlation is small. Strength should be judged relative to the pattern’s form. Points can follow a curve closely even when they do not follow a straight line.
- Reporting only the curve and ignoring \(r\). If a question gives a correlation, state what it says about linear association, then explain whether the scatterplot reveals a pattern \(r\) misses.
- Using the sign of a small \(r\) to describe the whole curve. A small positive \(r\) does not mean the response consistently increases as the explanatory variable increases. Describe the plotted form before making a directional claim.
- Forgetting context. Name both variables and include their units or measurement scales when explaining the observed pattern.
A strong AP response might say: “The correlation is close to zero, so there is little linear association between the hours from the target light exposure and the growth score. However, the scatterplot shows a clear curved pattern, with the highest scores near the target and lower scores farther away. Therefore, it would be incorrect to conclude that the variables have no relationship.”
Check Your Understanding
Use the distinction between linear association and a broader relationship to answer each question.
- A scatterplot has a clear U-shape, but \(r=0.03\). What does the correlation say, and what does it fail to show?
- In a symmetric curved pattern, observations on the left and right of the center have matching response values. How can this lead to \(S_{xy}=0\)?
- A report says two variables are “unrelated” because their correlation is near zero. What additional evidence should you inspect before accepting that claim?
- A scatterplot shows an inverted U-shape and \(r=-0.08\). Is it accurate to say the response generally decreases as the explanatory variable increases? Explain.
- Write one contextual sentence that describes both a near-zero linear association and a visible curved pattern.