Discriminant Validity
You may face a situation when it is impossible to measure the construct validity of your test because it does not correlate with other tests. The problem is that the tests may measure different constructs. It is especially important for the constructs that are not observable, like abstract concepts, attitudes, or behaviors.
So, you need to support your hypothesis that your new tests and other tests deal with different constructs and they are not highly related to each other. You have to assess discriminant validity here, meaning the extent to which your test is not related to other tests. If this validity shows some relationships, you cannot be sure that these tests are not measuring the same construct. So discriminant validity is a reliable indicator of the difference level between constructs.
Usually, you evaluate discriminant validity combined with convergent validity. It is about the similarity levels between constructs. The other name for discriminant validity in some fields of study is divergent validity.
Suppose you want to research the level of creativity in web design students. If you want to detect discriminant validity, you need another test for an unrelated construct, for example, for technological awareness. So, you develop a test to measure creativity and the other questionnaire to measure technological awareness. These tests should be different to see whether your discriminant validity is appropriate.
The two constructs here are expected to be unrelated, so the scores should not demonstrate significant relationships. If you see a correlation between the answers to these two tests, you may be dealing with the same construct in both questionnaires, so the discriminant validity of your research project is poor.
The Meaning of Discriminant Validity
Construct validity involves two subtypes - discriminant and convergent validity. Both of them are important to show how well a test measures something it was meant to measure.
Different constructs cannot theoretically be related to each other. Discriminant validity should show the lack of such a relationship.
You have two tests to measure introversion and depression. Their scores cannot ideally correlate. It means that those respondents who demonstrate high scores in introversion, do not have to demonstrate the same high scores in depression. If the results show different scores, the two tests have high discriminant validity.
You need to consider discriminant validity thoroughly because it shows how accurately your test targets the construct needed for your research without the involvement of any other separate constructs not related to your research design. The results may also indicate the accuracy of operationalization. It is about your skill to make measurable variables or observations from entirely abstract concepts.
It does not matter much which way you have chosen. If you get a low, zero, or negative correlation between the two test results, you can ensure that you are measuring two different constructs. It is the best evidence of high discriminant validity.
Comparison and Contrast of Discriminant and Convergent Validity
You need both discriminant and convergent validity to define the construct validity of your tool or research project. Though, you should know how different they are to make a distinction. Discriminant validity is supposed to ensure that the two tests that were not expected to be related are truly unrelated. Convergent validity, on the contrary, deals with the tests that were expected to be related and demonstrates that they are actually related.
In short, discriminant validity deals with differences, while convergent validity highlights similarities. You need to obtain evidence of the construct validity, so you have to check:
- ✔️ how well your test correlates with other tests that measure similar constructs;
- ✔️ how much your test does not correlate with the tests that measure different constructs.
As a researcher, you consider both discriminant and convergent validity to ensure evidence of high construct validity. Evaluation of discriminant validity before the assessment of convergent validity always brings better results.
You can check discriminant validity by picking one of the two variants:
- ✔️ choose two entirely opposing constructs (e.g., extroverts vs. introverts);
- ✔️ choose two absolutely unrelated concepts (e.g., aggression vs. color distinction).
You want to obtain the discriminant validity of a new scale for workaholism. If you choose the option of picking an opposite construct, think about laziness. All psychological findings argue that these constructs used for researching individual character traits are entirely opposite. Therefore, a rating scale that is meant to measure workaholism has to be negatively correlated with the test scores that measure laziness. It means that people who have displayed high scores in workaholism will show lower scores in laziness and vice versa.
Or you can opt for another approach - choose a construct that never relates to workaholism. For instance, you cannot expect that a habit of smoking cigarettes can relate to workaholism. Most research results show that workaholism does not have anything in common with a smoking habit. So, the test for workaholism can demonstrate only zero or minimal correlation with the test scores for a smoking habit.
It does not matter much which way you have chosen. If you get a low, zero, or negative correlation between the two test results, you can ensure that you are measuring two different constructs. It is the best evidence of high discriminant validity.
Measuring Discriminant Validity
You need to ensure that there is no correlation between unrelated constructs to get evidence that your measures have high discriminant validity.
The first term you need to know here is a correlation coefficient or Pearson’s r. Its value ranges between +1 and -1, so it is a reliable indicator of the possible relationships between variables.
The correlation coefficient values have the following interpretations:
- r = 1: the correlation is absolutely positive
- r = 0: the correlation is neutral (no correlation at all)
- r = -1: the correlation is absolutely negative
You can use specific online statistical software to calculate Pearson’s r. It can be Excel, SPSS, or R. All calculations are done automatically to save you time.
High correlations between scales or test items can result in issues with the correct determination of discriminant validity. So, the best values that will allow you to make a conclusion about the correlation and high discriminant validity are those that start at r = 0.85. Nevertheless, consider the most accurate conclusions. For instance, you may have noticed that a majority of studies in your field provide correlation coefficients of about 0.8-0.9, so if you receive 0.54, it should be considered low in the given context.
The essential things to take into account are that you will always get weaker correlations between unrelated concepts than between related but opposing constructs.
You want to research bipolar personality disorder and have created a test to measure it. Now, you need to evaluate the discriminant validity of your questionnaire or survey. So, you may decide to compare your bipolar disorder test results with those of another, pretty unrelated construct test.
Some researchers argue that people who have the characteristics of bipolar disorder tend to exhibit poor sexual activity. You cannot observe the direct relationship between bipolar disorder and low sexual activity, so it is a good choice of two unrelated constructs.
You pick out a sample of 65 respondents and ask them to answer two different questionnaires. Then, you calculate the correlation coefficients for both scales on bipolar disorder and sexual performance. You find out that bipolar disorder correlates as r = 0.3 with poor sexual activity. Such correlation can be regarded as negligible, and the discriminant validity of your tests will be pretty high.
Nevertheless, your next step should be establishing the convergent validity before you come to any conclusions about the overall construct validity. So, you need to prove that there is some positive correlation between the bipolar disorder scale and the scales for related constructs, like anxiety or apathy.
Final Thoughts
Now that you know what discriminant validity is and how to use it for establishing complete construct validity, you have to remember how it works, how to differentiate it from convergent validity, and how to measure it. Consider discriminant validity every time you need to provide evidence for your construct under research overall validity. If the validity of your study is high, other people can use it for their further academic work, or you can continue your study to achieve excellent academic results.