When we create a scale, questionnaire, or survey, it’s not enough to just ask questions. We also have to make sure the tool measures what we think it does. That’s where concepts like convergent validity and divergent (or discriminant) validity come in.
These two forms of validity are part of a larger process known as construct validation, basically, a way to show that your tool reflects the concept you’re aiming to measure. Let’s walk through what they are, why they matter, and how you can test them.
What Is Convergent Validity?
Convergent validity is about making sure that your tool relates well to other tools that measure similar ideas. The goal here is to show alignment between different ways of capturing the same concept.
Let’s take an example. Imagine you’re developing a new scale to measure pain.
Based on what we already know from theory and clinical practice, we’d expect people who report more pain to also show more signs of depression. So, if your new pain scale is valid, it should correlate reasonably well with an existing depression scale.
But how strong should that correlation be?
In general, we expect a moderate to strong correlation, usually somewhere between 0.40 and 0.80. This range tells us that the two scales are related but not identical. If the correlation is too low (say 0.10), it suggests that your pain scale might be picking up something completely different. If it’s too high (like 0.90), it could mean your new tool is essentially a duplicate of the depression scale, which raises questions about whether we need it at all.
So, convergent validity helps answer, Does this tool behave like it should when we compare it to similar tools?
What Is Divergent (Discriminant) Validity?
If convergent validity is about showing relationships that should exist, divergent validity is about confirming relationships that shouldn’t exist.
This means that your scale should not be highly related to something it’s not supposed to measure. For example, we don’t expect pain levels to be closely related to intelligence. If someone scores high on a pain scale, that shouldn’t tell us much (if anything) about how smart they are.
When you test for divergent validity, you’re usually looking for a very weak or no correlation, close to zero. This tells us that the tool is specific to its intended domain and isn’t picking up on unrelated traits.
In simple terms, divergent validity is about showing that your tool doesn’t drift into areas it wasn’t designed to measure.
How to test both? The role of item-level Discriminant Validity
Sometimes we want to go a step further and zoom in on individual items within a scale. This is especially useful when a tool is made up of several domains, or groups of related questions that each measure a sub-part of a broader concept.
Here’s where item-level discriminant validity comes in.
The idea is to make sure each item (question) in your scale correlates more strongly with the total score of its own domain than with scores from other domains. But we need to be careful, when we calculate this, we exclude the item from the domain total to avoid artificially inflating the result. This is called a corrected item-to-total correlation.
For a well-functioning item, we generally want a corrected correlation of at least 0.40 with its own domain total. If the item correlates more strongly with another domain, that’s a red flag. It might be in the wrong category, or maybe it doesn’t fit the structure of the questionnaire as well as we thought.
So, when you’re validating a multi-domain tool, you can run this analysis to make sure each item belongs where it’s supposed to. Think of it like making sure each player is on the right team.
A real example: The SEAR Questionnaire
Let’s make this concrete with a real-world example.
The Self-Esteem and Relationship (SEAR) questionnaire is a 14-item tool used to evaluate psychosocial factors in men with erectile dysfunction (ED). It’s split into two main domains:
- Sexual Relationship Satisfaction (8 items)
- Confidence (6 items)
Researchers tested both convergent and divergent validity when they developed SEAR. Here’s what they did, and what they found.
Divergent Validity in action
The Sexual Relationship Satisfaction domain was expected to have low correlations with tools measuring general health and well-being like the Psychological General Well-Being Index (PGWBI) and the Short Form-36 (SF-36). That made sense, the SEAR items are focused on sexual relationships, not general physical health. The results confirmed that the correlations were weak, just as the theory predicted.
The same was true for the Confidence domain. It showed low correlations with physical health sections of the SF-36, like Physical Functioning and Bodily Pain. Again, this supported the claim that the Confidence items are not capturing physical health, they’re tapping into something more emotional or interpersonal.
Convergent Validity in action
On the flip side, the Confidence domain did show moderate correlations with mental health measures. These included the Mental Component Summary of the SF-36, and several subdomains from the PGWBI such as Anxiety, Depressed Mood, Positive Well-Being, and Self-Control. This confirmed that the Confidence items are indeed related to aspects of emotional and psychological well-being, just as expected.
Item-level checks
They also ran corrected item-to-total correlation checks and found that each item fit better with its own domain than with others. This helped confirm the internal structure of the questionnaire that the two domains were distinct and each item lived in the right place.
Wrapping Up: Why This Matters
If you’re developing or testing a scale, whether it’s for pain, anxiety, quality of life, or even something like job satisfaction, you need to go beyond intuition. Just because a question feels like it fits doesn’t mean it actually works the way you think it does.
That’s where convergent and divergent validity help.
- Convergent validity shows your tool relates to what it should.
- Divergent validity shows it doesn’t relate to what it shouldn’t.
- Item-level checks make sure each piece fits into the right part of the puzzle.
You can test all of these using straightforward statistical techniques like correlation analysis. It’s not about chasing perfection, it’s about making sure your tool is accurate, focused, and trustworthy. And when your tool is used in clinical or research settings, that level of trust really matters.
By following this process, you build stronger, more credible instruments, the kind that truly reflect the reality you’re trying to measure. Also, here is the detailed steps how to develop a PRO instrument.