Convergent vs. Discriminant Validity: Which Is More Important?

I would prioritize convergent validity. For example, in a motivation scale, items measuring the same construct should load strongly together. If they don’t, the construct itself is not valid. Even with good discriminant validity, the model would still be weak. Theoretically, we must first ensure each construct is measured consistently before comparing it with others. Weak convergent validity can lead to unreliable results and an invalid model overall.