Validity of an Instrument (part 2)
In the previous post, I talked about three types of validity: face validity, criterion-related, and content validity. And now for the rest of the story…
A fourth type of validity is construct validity. This type is much more often used to validate instruments that measure abstract concepts, such as we tend to use in the social sciences. Suppose you write an instrument measuring self-esteem. Your task is to relate the construct “self-esteem” to a similar construct in order to establish construct-validity. Let’s say people with higher self-esteem tend to have more friends. You could give a group of people your self-esteem instrument, and ask them to also report how many friends they have. If those with higher self-esteem also have the most friends, then you have begun to validate your instrument. Construct validity cannot be obtained by showing only one relationship between two constructs. You should also link your construct (e.g. self-esteem) to many other constructs over a period of time and with different samples in order to establish construct validity. As you can see, construct validity is an on-going process.
The last two types of validity are actually more specific types of construct validity, those of convergent validity and discriminant validity, with the main difference being directionality. Convergent validity can be established by correlating scores on two similar instruments that measure the same construct. For example, in establishing that a new instrument really measures self-esteem, you would give a sample of people the new instrument as well as another, established, instrument that also measures self-esteem (or at least some aspect of it). If the correlation between them is decent strong and in the expected direction, evidence for convergent validity is established. When establishing discriminant validity, on the other hand, you want to show that you aren’t measuring the same variable as one measured by another instrument and just calling it by a different name. For example, if your new instrument is meant to measure self-esteem (one’s evaluation of his/her self-worth) you’d want to be able to show that your questions are conceptually different from questions that measure self-confidence (one’s evaluation of his/her abilities). To do this, you would give a sample of people your self-esteem instrument and an instrument that measures self-confidence. These constructs are bound to be related to each other, but as long as the correlation between them is not strong, evidence for discriminant validity can be shown. The lower the correlation between the two instruments, the stronger is your argument for discriminant validity of your new instrument.
When considering whether to write your own instrument for your research, you’ll want to also think about how you will establish reliability and validity of that instrument. This is important because without these markers of instrument quality, the quality of your research results will also come into question.