A depression questionnaire has excellent internal consistency and excellent test-retest reliability. A critic argues it may still be measuring the wrong thing. Why do the reliability statistics fail to answer that objection?

C. Because both compare the measure to itself, not to the construct.

Both statistics are internal to the instrument, the items agree with each other, and the scale agrees with itself a fortnight later. Twenty items about sleep would be beautifully consistent, stable, and correlated with depression, without measuring it. Construct validity needs outside evidence: convergence, discrimination, and movement when treatment works.