In experimental research, we often operationalize constructs using nominal variables when they are actually continuous variables. We do this because it can be easier to examine the impact of different levels of the independent variable on the dependent variable when these levels are discrete. In the table below, we provide some examples where we may choose to use discrete levels where continuous measurements are more reflective of the behaviour of a given construct.
| Construct | Discrete levels | Continuous measurement |
| Age | Under 21 21 - 30 31 - 40 41 - 50 51 - 64 65 + |
Age is a continuous measurement from birth (i.e., 1 second old) to death (e.g., 80 years, 23 days, 5 hours, 2 minutes, 17 seconds). It is a continuous variable that can be measured with accuracy. |
Background music Loudness of music |
Low Medium High |
Sound pressure level (SPL) with a range from 0 dB to 120 dB (i.e., audible range of the human ear). |
| Teacher ability Level of experience |
1 year 10 years 20 years |
Experience may be better evaluated in years since there is a large difference between 1 and 10 years. Also, it may be debatable whether 20 years of experience is twice as much experience as 10 years (in real terms). |
When we operationalize constructs using nominal variables when they are actually continuous variables, we cannot be sure that the same results would have been obtained if more precise measurements were taken that better reflect the behaviour of a given construct.
It is possible that the treatment can sensitise the participant to the construct that is being measured in such a way that the factorial structure of the construct changes, at least in terms of the way that the participant views such a construct. This is a threat to construct validity known as treatment-sensitive factorial structure (Heppner et al., 2008).
For example, we may view a construct as one-dimensional (e.g., employee stress/burnout or addiction). However, after having received the treatment, especially if such a treatment is an educational program of sorts that leads the participant to understand the construct better than they did originally (e.g., they may be able to distinguish between different aspects of employee stress/burnout, such as loss of desire, difficulty getting up in the morning, attitude towards co-workers, etc.). This can make a one-dimensional construct into a multi-dimensional construct, despite the fact that participants only have one way of responding (i.e., a total score for stress/burnout). This can happen when mono-measures are used.
There is a lot of ambiguity, not only in the way that constructs can be operationally defined, but also how different constructs relate to one another (e.g., how the construct, anger, relates to the construct, depression). What are the boundaries between these different constructs? Where does one construct start and the other end?
When different construct overlap, the results that we generate when measuring these construct can become confounded. We discuss the impact of extraneous and confounding variables in more detail in the article: Extraneous and confounding variables. In the article, Convergent and divergent validity, we discuss one of the ways of checking whether constructs have been confounded.
You can learn more about the different types/tests of validity that help to establish construct validity in the following articles: Content validity, Convergent and divergent validity, Criterion validity (concurrent and predictive validity) and Reliability in research.
Campbell, D. T., & Fiske, D. W. (1959). Convergent and discriminant validation by the multitrait-multimethod matrix. Psychological Bulletin, 56: 81-105.
Cronbach, L. J., & Meehl, P. E. (1955). Construct validity in psychological tests. Psychological Bulletin, 52: 281-302.
Goodwin, C. J. (2009). Research in Psychology: Methods and Design. John Wiley and Sons.
Messick, S. (1980). Test validity and the ethics of assessment. American Psychologist, 35: 1012-1027.