Is concordance the same as correlation?

It is a pairwise concept used in some association measures; correlation is a broader family of summaries.

Does a tie count as concordant?

Not under the strict definition used here; particular statistics specify how ties contribute.

Is every concordance index calculated from every possible pair?

No. Some applications, including censored-outcome analysis, restrict which pairs are comparable.

AI Glossary / Statistics and probability / Intermediate

Concordant pairs

Definition

Concordant pairs are pairs of observations whose ordering agrees on two variables. If one observation has a higher value than the other on the first variable and also a higher value on the second, the pair is concordant. The same applies when it is lower on both. Concordance concerns relative order, rather than equal values, equal differences, or a causal connection between the variables.

What it means

Each observation contains two measurements, such as a predicted priority and an independently assessed outcome. Select two observations and compare their ordering on both measurements. Agreement makes the pair concordant; opposite ordering makes it discordant. A tie on either variable requires separate treatment.
Counting these pairs helps summarize ordinal association. This is useful when ranks are meaningful but the numerical distance between categories is not. For example, a difference between satisfaction ratings of one and two need not represent the same experiential change as a difference between four and five.

How it works

For observations i and j with values (x_i, y_i) and (x_j, y_j), inspect the product (x_i - x_j)(y_i - y_j). A positive product identifies a concordant pair; a negative product identifies a discordant pair. A zero product indicates a tie in at least one variable.
With n complete observations, there are n(n - 1)/2 distinct unordered pairs. Compare each pair once. The resulting pair indicators share observations, so they should not be treated as independent experimental units when calculating uncertainty.
Without ties, Kendall's tau can be calculated as (C - D)/(C + D), where C and D count concordant and discordant pairs. Tie-adjusted versions use different denominators. State the measure and tie convention rather than assuming all rank statistics treat equality identically.
Check the direction of both scales before interpreting agreement. If a larger score means higher priority but a smaller numerical rank means better performance, the orientations differ. Reversing one scale changes concordance to discordance for untied pairs; reversing both preserves agreement. Missing values, repeated observations, and restricted comparison sets also need explicit handling.

Three practical examples

A quality team compares two assessors' ordering of the same products and counts pairs they place in the same order.
An AI team checks whether pairs of cases ordered by predicted urgency are ordered similarly by a reference assessment.
A customer research analyst studies ordinal association between service ratings and willingness to recommend, keeping tied responses separate.

Think of it this way

Imagine two judges arranging the same items from lowest to highest. Choose any two items. If both judges put the same one ahead, that pair is concordant. They can agree on the order even if their numerical scores differ substantially. Pairwise agreement is therefore about ordering, not identical scoring.

Worked example

Consider an original example with four observations: A=(1,2), B=(2,1), C=(3,4), and D=(4,3). There are 4 times 3/2 = 6 unordered pairs and no ties.
A versus B is discordant: the first coordinate rises from 1 to 2 while the second falls from 2 to 1. A versus C and A versus D are concordant because both coordinates rise. B versus C and B versus D are also concordant. C versus D is discordant because the first coordinate rises while the second falls.
Thus C=4 and D=2. The concordant proportion is 4/6, approximately 0.667. Kendall's tau is (4 - 2)/6 = 1/3, approximately 0.333. These are different summaries: with no ties, tau equals twice the concordant proportion minus one.
Now replace D's second coordinate with 4. C and D become tied on that variable, so this pair is neither strictly concordant nor strictly discordant. A tie-aware statistic needs its appropriate denominator. The example illustrates counting; four observations do not justify a strong population conclusion.

Illustrative case study

In a fictional company, Northline Support tests whether an AI prioritization score agrees with a human reference ordering of support requests. The team initially compares numerical score differences, but the score and reference rating use different scales.
An analyst proposes pairwise ordering checks. The team defines which direction means more urgent, documents how tied ratings are handled, and examines whether certain request categories account for most disagreements. It also notes that the human reference can itself contain uncertainty.
Concordance helps reveal ranking behavior, while separate checks assess missed urgent cases and score calibration. This fictional case illustrates a complementary evaluation tool; it does not claim that strong rank agreement alone makes the system suitable for deployment.

Common misconceptions and mistakes

A concordant pair does not require matching scores or equal gaps. A high concordant proportion does not imply calibrated probabilities. Do not count ties as agreement without stating that convention. Do not confuse the number of pairs with the number of independent observations. A monotonic association can be strong even when the relationship is nonlinear. Concordance alone cannot identify why two variables move together or establish that changing one will change the other.

Business use

Businesses use pairwise concordance to compare rankings, study ordered customer responses, and assess whether a prioritization system preserves useful ordering. It is especially relevant when decisions depend on which item comes first. Report the comparison population, scale directions, treatment of ties, and uncertainty. For operational use, combine ranking evidence with the costs of specific errors; a single reversal involving a critical case can matter more than many harmless agreements.

Check your understanding

Are (2,7) and (5,9) concordant?

Answer: Yes, both coordinates increase.

Are (2,7) and (5,4) concordant?

Answer: No, their orderings disagree.

Why is the worked example's tau not 0.667?

Answer: Tau subtracts the discordant contribution, giving 0.333.

Why must an analyst check scale direction?

Answer: A low numerical rank may indicate a high standing, reversing the apparent relationship.

Related terms

Discordant pairs
Kendall's tau
Calibration
tied ranks
ordinal association
rank correlation
Spearman correlation
concordance index
pairwise comparison

Learn next

Learn Kendall's tau and its treatment of ties, then compare rank association with Pearson correlation and probability calibration. Practice counting all pairs in a small dataset before interpreting software output. When analyzing prediction systems, study how comparison rules and outcome availability determine which pairs can legitimately be evaluated.

Sources

David Howell, Statistical Methods for Psychology glossary, Concordant pairs: https://www.uvm.edu/~statdhtx/methods8/Glossary/Glossary.html SciPy official documentation, kendalltau, including concordant, discordant, and tied-pair definitions: https://docs.scipy.org/doc/scipy/reference/generated/scipy.stats.kendalltau.html

Previous: Conceptualization (information science) Next: Concrete word