What Is Discriminant Validity? | Definition & Examples
Discriminant validity (or divergent validity) captures whether a test designed to measure a specific construct yields results different from tests designed to measure theoretically unrelated constructs.
Discriminant validity is evaluated alongside convergent validity, which assesses whether a test produces results similar to tests that measure related constructs. Together, convergent and discriminant validity provide complementary evidence of construct validity—whether a test measures the construct it’s supposed to.
To ensure that your test is sensitive to this trait and not others, you compare infants’ scores on your test to their scores on a fine motor skills test. You do not expect to observe a relationship between vocabulary recognition and fine motor skills.
A low correlation between infant scores on these tests supports the discriminant validity of your protocol.
What is discriminant validity?
Discriminant validity (less commonly called divergent validity) refers to a test’s ability to produce results that differ from other tests of theoretically distinct constructs. Constructs, which are common in psychology research, are phenomena that cannot be directly measured, such as intelligence or depression.
When designing a new test or measure, it’s important to ensure that you’re measuring the construct you intend to and not something unrelated. This can be determined by comparing results from your test to a test that measures a different construct. If your test is able to discriminate between these constructs, the results from the two tests will be uncorrelated, providing evidence of discriminant validity.
Discriminant validity is one form of evidence for construct validity, an overall measure of whether a test is measuring what it’s supposed to. Discriminant validity is generally measured alongside convergent validity, which is whether a test yields results similar to tests of related constructs. Other types of validity that provide evidence of construct validity include face validity, content validity, and criterion validity.
The researcher computes the correlation between scores on both tests. She is surprised to observe a strong correlation between the two, indicating that her test does not have discriminant validity. She must redesign her test to better isolate impulse control.
Convergent and discriminant validity
Convergent validity and discriminant validity both provide evidence of construct validity, but each provides distinct information about the validity of a measurement.
- Convergent validity assesses whether a test yields results that match other tests of theoretically related constructs. If two tests are measuring similar things, their results should be strongly correlated.
- Discriminant validity ensures that a test is sensitive to the construct it’s been designed to measure without inadvertently measuring an unrelated construct. In other words, a test’s results should be uncorrelated to those of a test that measures something else.
Because convergent and discriminant validity provide complementary evidence, they are both required to demonstrate construct validity. Convergent validity is usually measured first, as it is important to ensure that your test is measuring the construct it’s supposed to before verifying that it can discriminate between unrelated constructs.
When developing a new measure, the tests you use to assess convergent and discriminant validity depend on what you hope to capture with your measure. It’s important to thoroughly research the construct you’re studying and carefully consider how your test results relate to your research question.
The following example illustrates how the same test could be used to assess convergent or discriminant validity, depending on the researcher’s intention.
1. Imagine the psychologist is simply concerned about whether the survey accurately measures anxiety symptoms.
To test for convergent validity, they could compare their results to an existing depression test. Because anxiety and depression have similar symptoms, the psychologist would expect to see a high correlation between both tests. Discriminant validity could be assessed by comparing the anxiety survey to a test of a condition with fewer overlapping symptoms, such as obsessive-compulsive disorder.
The psychologist would expect to observe a higher correlation between their survey and a depression survey relative to a test of obsessive-compulsive disorder, demonstrating convergent and discriminant validity, respectively. They could conclude that their survey is likely measuring symptoms of anxiety.
2. Now imagine the psychologist is instead trying to discern whether their patients are experiencing anxiety or depression.
In this case, they might use an existing anxiety scale to assess convergent validity, which should yield highly similar results. As they would like to differentiate anxiety from depression, they could use a test of depression to determine discriminant validity.
The correlation between the psychologist’s survey and the existing anxiety measure should be very high (demonstrating convergent validity). The correlation between the new survey and the depression survey should be relatively lower (demonstrating discriminant validity).
As the example above illustrates, the relationship between convergent and divergent validity is important—a test’s results should be more similar (i.e., more strongly correlated) to those of tests measuring corresponding constructs than to tests measuring unrelated constructs.
Discriminant validity example
Consider the following example of how discriminant validity might be used in psychology research.
To assess the construct validity of your game, you first assess convergent validity by comparing participants’ scores to their results on a problem-solving task. Because both measures relate to cognitive flexibility, you expect to observe a strong correlation between the two.
Once you’ve demonstrated convergent validity, you measure discriminant validity. You are concerned that your game might simply be measuring how quickly participants respond to external stimuli, so you compare your results to scores on a test of processing speed.
You find a strong correlation between your game scores and the results of the problem-solving task, but a relatively low correlation between your game and the processing speed test. These outcomes demonstrate, respectively, the convergent and discriminant validity of your test, providing evidence of construct validity.
How to measure discriminant validity
Discriminant validity measures the extent to which a target test provides results that differ from tests that measure unrelated constructs. Discriminant validity can be determined using the following approach.
Step 1: Determine convergent validity
Before assessing the discriminant validity of your target test, it is important to measure its convergent validity. If your test does not measure the thing it’s supposed to, an assessment of discriminant validity is not informative.
Step 2: Select a measure for comparison
You must carefully choose a test to compare your results to, and this choice will depend on what you hope to achieve with your target test. The comparison test should measure an unrelated but meaningful construct.
Step 3: Administer all tests to a sample
Once you’ve selected a discriminant measure to compare your own test to, both tests should be administered to a representative sample.
Step 4: Compare the results of both tests
Discriminant validity can be assessed by computing the correlation between scores on the target and comparison tests. Pearson’s correlation is often used to determine the relationship between two measures.
Pearson’s correlation coefficient, r, falls between -1 and +1. There are several potential outcomes of a correlation analysis:
- Strong, positive correlation: If people who score high on one measure score high on the second measure (or low on both measures), r will be close to +1. The results are said to be positively correlated.
- Strong, negative correlation: If people who score high on one measure score low on the second measure and vice versa, r will be close to -1. The results are negatively correlated.
- Weak or nonexistent correlation: If there is no relationship between scores on two tests, r will be close to 0. The results are uncorrelated.
Step 5: Interpret your findings
The thresholds for “strong” or “weak” correlations vary across different fields. However, when assessing discriminant validity, it’s generally more important to compare patterns between convergent and discriminant measures: you should expect to see a stronger correlation between your target test and a similar test and a weaker correlation between your target and an unrelated test.
Strong negative correlations can in fact provide evidence of convergent validity. For example, a test measuring stress should be negatively correlated with a test measuring relaxation. Because stress and relaxation are strongly related, this example could be interpreted as evidence of convergent validity.
Discriminant validity is instead generally evidenced by weak or nonexistent correlation between two measures. If two constructs are unrelated, there should be no relationship (positive or negative) between them.
The distinction between convergent and discriminant validity is complex and highly contextual. It is therefore crucial to carefully consider the expected relationship between different constructs in your field and prespecify how you expect measures of convergent and discriminant validity to relate to one another.
Frequently asked questions about discriminant validity
- Why are convergent and discriminant validity often evaluated together?
-
Convergent validity and discriminant validity (or divergent validity) are both forms of construct validity. They are both used to determine whether a test is measuring the thing it’s supposed to.
However, each form of validity tells you something slightly different about a test:
- Convergent validity indicates whether the results of a test correspond to other measures of a similar construct. In theory, there should be a high correlation between two tests that measure the same thing.
- Discriminant validity instead measures whether a test is similar to measures of a different construct. There should be a low correlation between two tests that measure different things.
If a test is measuring what it is supposed to, it should correspond to other tests that measure the same thing while differing from tests that measure other things. To assess these two qualities, you must determine both convergent and discriminant validity.
- Are discriminant and divergent validity the same thing?
-
In short, yes! The terms discriminant validity and divergent validity are often used synonymously to refer to whether a test yields different results than other tests that measure unrelated concepts. However, “discriminant validity” is the more commonly used and accepted term.
- What is a construct?
-
A construct is a phenomenon that cannot be directly measured, such as intelligence, anxiety, or happiness. Researchers must instead approximate constructs using related, measurable variables.
The process of defining how a construct will be measured is called operationalization. Constructs are common in psychology and other social sciences.
To evaluate how well a construct measures what it’s supposed to, researchers determine construct validity. Face validity, content validity, criterion validity, convergent validity, and discriminant validity all provide evidence of construct validity.
