Why Small Reporting Errors Matter
A correlation coefficient can be calculated correctly from the numbers entered and still be misleading if those numbers were paired incorrectly. Even with the right pairs, a report can distort the result by dropping the sign, calling the coefficient a percentage, or failing to say which variables it summarizes. Careful work therefore includes both checking the data and communicating what the calculation means.
The Pearson sample correlation coefficient, written \(r\), summarizes the direction and strength of a linear association between two quantitative variables measured on the same observational units. It is calculated from paired observations: each value of one variable must stay matched to the value of the other variable from that same unit.
In practical terms, the formula compares how the two variables vary together with how much each varies on its own. If one variable tends to be above its mean when the other is above its mean, the numerator tends to be positive. If one tends to be above its mean when the other is below its mean, the numerator tends to be negative. The denominator is positive whenever both variables vary, so it does not change the sign.
Error 1: Pairing Values From Different Observations
The correlation formula uses the pairs \((x_i,y_i)\). If a spreadsheet column is sorted independently, or a row is shifted while copying data, the calculation combines values that did not come from the same person, object, or occasion. The arithmetic may still produce a number between \(-1\) and \(1\), but that number summarizes the incorrectly assembled pairs rather than the intended data.
Worked Example: Detecting a Misaligned Column
Suppose a hypothetical class records each student’s hours of study and quiz score. The table gives the correct pairs.
| Student | Study hours, \(x\) | Quiz score, \(y\) |
|---|---|---|
| A | 1 | 50 |
| B | 2 | 60 |
| C | 3 | 70 |
| D | 4 | 80 |
For the correct data, \(\bar{x}=2.5\) hours and \(\bar{y}=65\) points. The cross-products sum to \(({-1.5})(-15)+({-0.5})(-5)+(0.5)(5)+(1.5)(15)=50\). The sums of squared deviations are \(5\) for study hours and \(500\) for quiz scores. Therefore,
In this hypothetical data set, the four correctly paired observations lie exactly on an increasing straight line. Now suppose the scores are shifted one row, producing the incorrect pairs \( (1,60),(2,70),(3,80),(4,50)\). Their means are still \(2.5\) and \(65\), but the cross-products now sum to \(7.5-2.5+7.5-22.5=-10\). The sums of squared deviations remain \(5\) and \(500\), so the wrongly calculated value is
The change from \(1\) to \(-0.200\) is not a feature of the students; it is a consequence of pairing each study time with the wrong score. Before calculating, check that every row still represents one observational unit and that sorting one variable has not left the other behind.
Reordering whole rows does not change \(r\), because both values in every pair move together. The danger is reordering just one variable or otherwise breaking the row-wise matches. A useful data check is to retain an identifier, such as a student code, while merging or sorting records, then verify several pairs against the original source.
Error 2: Calling \(r\) a Percentage or Dropping Its Sign
The value of \(r\) is a unitless coefficient, not a percentage. A value such as \(0.80\) is not, by itself, a statement that 80% of anything is present, explained, or accurate. Also, the sign is part of the result. Reporting only the absolute value \(|r|\) removes whether the linear association is positive or negative.
This follows directly from the formula. The denominator is positive when both variables vary, so the sign of \(r\) is the sign of the sum of cross-products in the numerator. Replacing \(r\) by \(|r|\) can preserve a measure of magnitude while erasing the direction of the linear pattern. The sign indicates whether the variables tend to move together or in opposite directions; it does not establish that one causes the other.
Worked Example: Keeping the Sign in a Report
Imagine four hypothetical observations of weekly practice sessions, \(x\), and completion time on a task, \(y\), in minutes.
| Observation | Practice sessions, \(x\) | Completion time, \(y\) |
|---|---|---|
| 1 | 1 | 8 |
| 2 | 2 | 6 |
| 3 | 3 | 4 |
| 4 | 4 | 2 |
The means are \(\bar{x}=2.5\) sessions and \(\bar{y}=5\) minutes. The cross-products of deviations are \(-4.5,-0.5,-0.5,-4.5\), which sum to \(-10\). The sums of squared deviations are \(5\) for practice sessions and \(20\) for completion time. Thus,
The correct report is that these four hypothetical observations show a perfect negative linear association: more practice sessions correspond to shorter completion times in this data set. Saying “the correlation is 100%” is incorrect, because \(r\) is not a percentage. Saying “the correlation is \(1\)” is also incorrect, because it reverses the sign and describes a positive association. Even this perfect sample pattern does not, on its own, prove that additional practice caused the shorter times.
Error 3: Reporting a Number Without Its Variables or Pattern
A bare statement such as “the correlation is \(-0.84\)” leaves out essential context. Readers need to know which two variables were measured and, in ordinary language, what direction and kind of pattern the coefficient summarizes. Include measurement units when naming the variables. A precise report should also avoid making the coefficient sound like a causal effect.
Worked Example: Turning a Coefficient Into a Contextual Report
Consider four hypothetical people, with weekly hours of physical activity and resting pulse rate measured in beats per minute.
| Person | Activity, \(x\) (hours per week) | Resting pulse, \(y\) (beats per minute) |
|---|---|---|
| A | 1 | 77 |
| B | 2 | 75 |
| C | 3 | 76 |
| D | 4 | 72 |
The means are \(\bar{x}=2.5\) hours per week and \(\bar{y}=75\) beats per minute. The \(x\)-deviations are \(-1.5,-0.5,0.5,1.5\); the \(y\)-deviations are \(2,0,1,-3\). The cross-products sum to \((-1.5)(2)+(-0.5)(0)+(0.5)(1)+(1.5)(-3)=-7\). The squared-deviation sums are \(5\) for activity and \(14\) for resting pulse. Therefore,
An informative report is: “Among these four hypothetical people, weekly physical activity hours and resting pulse rate had a negative linear association (\(r=-0.837\)); higher activity hours tended to occur with lower resting pulse rates.” The report identifies both variables, includes the sign, and describes the pattern in context. It does not say activity lowered pulse rate, which would claim a causal effect not established by this correlation.
A Short Check Before Reporting
A few deliberate checks prevent many avoidable errors. Confirm that both variables are quantitative and that their values are paired on the same observational units. Verify that the calculation uses every intended pair and that the reported value retains its sign. Then translate the sign and magnitude into a contextual description of the linear pattern, rather than treating the coefficient as a percentage or a causal result.
Confirm that each \(x_i\) and \(y_i\) come from the same person, object, or occasion, and that records were not independently sorted.
Use the paired values, retain the numerator’s sign, and confirm that both variables vary so the denominator is nonzero.
Name both variables and their units, report \(r\) with its sign, describe the direction and linear pattern, and do not claim causation from correlation alone.
Common Mistakes
- Using mismatched observations. A correlation can be numerically valid for the entered columns but meaningless for the intended question if the rows no longer match.
- Writing \(r\) as a percentage. \(r=0.70\) is a correlation coefficient of \(0.70\), not “70% correlation.”
- Reporting only the magnitude. “The correlation is \(0.84\)” does not describe a result of \(r=-0.84\); the minus sign conveys the negative direction.
- Leaving out context. State which variables are associated and describe how they vary together, with units where appropriate.
- Turning association into cause and effect. A correlation describes a pattern in paired data; it does not by itself show that changing one variable changes the other.
Check Your Understanding
Use these questions to test your understanding of calculation and reporting errors.
- A researcher sorts a data set by blood pressure but does not move the corresponding age values with each person. What has gone wrong, and why can the resulting \(r\) mislead?
- In your own words, what information is lost when a reported correlation of \(-0.62\) is written as \(0.62\)?
- Explain why \(r=0.75\) should not be reported as “a 75% correlation.”
- Write a contextual report for \(r=-0.48\) between commute distance in kilometers and walking time in minutes. Avoid implying that the correlation proves causation.
- If both variables’ rows are reordered together, does the correlation change? Explain why the pairing matters.