As written about earlier, Gipps and Murphy (1994) discuss validity and bias (or lack of it). They cite Cronbach and Messick’s notions of a unitary model of validity based on the construct. This construct validity is regarded, say Gipps and Murphy, as one of the three dimensions – the others being content and criterion. They argue that no content or criteria can ever be free from bias, and hence these are less dominant aspects when looking for validity.
On the other hand the Futurelab report I wrote about yesterday takes this overarching construct validity as being composed of four aspects.
- content validity: are items fully representative of the topic being measured?
- convergent validity: given the domain definition, are constructs which should be related to each other actually observed to be related to each other?
- discriminant validity: given the domain definition, are constructs which should not be related to each other actually observed to be unrelated?
- concurrent validity: does the test correlate highly with other tests which supposedly measure the same things?
(ibid, appendix A, page 24)
The definition in the report is that construct validity = the extent to which a test measures what it claims to measure. This is not at odds with Gipps and Murphy.