If I had to prioritize one in structural equation modeling, I would consider discriminant validity slightly more important, although both are essential for establishing ...
If I had to prioritize one in structural equation modeling, I would consider discriminant validity slightly more important, although both are essential for establishing construct validity.
From a theoretical perspective, convergent validity tells us whether the items that are supposed to measure the same construct actually converge well together. For example, in my thesis on students’ perceptions of AI-assisted academic writing, items such as AI helps generate ideas, AI improves vocabulary, and AI supports grammar correction should all load strongly on the same latent construct. This is usually supported by high factor loadings and acceptable AVE values.
However, I would prioritize discriminant validity because it ensures that one construct is truly distinct from another. For instance, in my research, perceived usefulness and ease of use may be closely related, but they should not be measuring the same thing. If discriminant validity is weak, the constructs may overlap conceptually, making it difficult to interpret the SEM model meaningfully. This is especially problematic when testing hypotheses between latent variables because the relationships may become inflated or misleading.
A practical example from the literature is when constructs such as motivation and engagement are highly correlated. Without discriminant validity, it becomes unclear whether the model is truly testing separate theoretical variables or simply renaming the same construct.
One major consequence of weak validity is that the theoretical conclusions of the thesis may become questionable. Weak convergent validity means the items do not adequately represent the construct, while weak discriminant validity means the constructs are not sufficiently distinct. In my opinion, the latter is more dangerous because it directly affects hypothesis testing and interpretation of causal paths in SEM. In fact, many scholars emphasize that both forms should be evaluated together because construct validity cannot be established with only one.