All guides › Verification and method comparison
Does high correlation mean two laboratory methods agree?
High correlation tells you that results move together. Agreement depends on how far apart the paired results are and whether those differences are acceptable for their intended use. Two methods can have a correlation of 1.00 and still give systematically different results.
Worked example at a glance
1.00
Pearson correlation
+20%
Method B relative to A
+2 to +20 units
Absolute paired differences
Start with paired results from the same specimens
Suppose a laboratory is considering whether it could use method B in place of method A. Each specimen is measured with both methods, so each point on the first plot represents two results from the same specimen. The horizontal axis shows A and the vertical axis shows B. The dashed equality line shows where the two results would be identical. [1] [2]

The ten fictional pairs lie exactly on a straight line. As method A increases, method B increases in step. Pearson’s correlation coefficient is 1.00. Look again at the dashed line. Every point lies above it, because every result from B is 20% higher than its paired result from A. The line the points follow and the equality line describe different relationships.
A scatterplot is a useful first view. It shows the range of results included in the comparison and whether the pairs sit above, below or around equality. It can be hard to judge the size of the difference from that view alone, especially when the whole measuring range fits on one set of axes. The second plot makes the difference the quantity you read directly.
Plot the difference for each pair
For each specimen, subtract A from B and plot the difference against the mean of the two results. The pair mean places the difference along the measuring range while treating the two methods symmetrically. The dashed line is zero. Points above it mean B gave a higher result for those specimens. [1]

| A | B | B − A |
|---|---|---|
| 10 | 12 | +2 |
| 50 | 60 | +10 |
| 100 | 120 | +20 |
At the low end, 12 minus 10 gives +2 units. At the high end, 120 minus 100 gives +20 units. The percentage difference relative to A is the same for every pair, but the absolute difference grows as the concentration increases. The slope on the difference plot makes that visible. If you report a percentage, name the denominator. In this example it is A.
How do I judge whether the difference matters?
Start with the proposed use. A laboratory considering replacement needs to know where the results influence interpretation and what difference would be acceptable there. A 2-unit difference and a 20-unit difference have different implications if the requirement is expressed in absolute units. Near a decision threshold, even a small difference can change which side of the threshold a result falls on. The comparison needs specimens in the part of the range that matters for the decision. [2]
Real pairs also vary around the overall pattern. Check how widely the differences scatter, whether that spread changes across the range, and whether individual pairs show differences that a summary would hide. An average difference gives one number for the whole range. In this example it would conceal how the difference grows from the low end to the high end. [1]
The ten pairs were constructed to make the proportional pattern easy to see. In real data, two methods could have a small average difference because positive and negative differences cancel, while individual pairs still differ appreciably. Reading both the pattern and the spread helps you see which results a single summary would miss. [1]
Use the language of comparison in your conclusion. Here, B reads higher relative to A. Establishing which is closer to a reference requires separate evidence about that reference and its suitability. The plots show the size, direction and pattern of the differences between these two methods.
How you might summarise this comparison
Example wordingAcross these fictional pairs, method B gives results 20% higher than method A. The absolute difference increases from 2 to 20 units across the range shown. To judge whether the methods could be used interchangeably, compare those differences with the requirements for the intended use at the relevant concentrations.
Member Learning
Plan a study and prepare the data
Planning a laboratory study and preparing your data takes you from a clear study question through data preparation to a summary you can explain.
Sources and related reading
- Bland and Altman · Statistical methods for assessing agreement (1986) · Why correlation and agreement answer different questions.
- CLSI EP09 · Measurement Procedure Comparison and Bias Estimation Using Patient Samples · Guidance on planning and interpreting paired patient-sample comparisons.
Read about method verification · Read about analytical performance specifications
By Stephen MacDonald · Suggest a correction · Back to Find a guide