Content validity is the qualitative measure of whether the subjects covered by the test represent the full range of subjects that the test is designed to cover. Content validity is a prime indicator of whether the test is useful for determining the knowledge held by the test taker on any given subject. A good score on a test with content validity is an indicator that the test taker knows the subject well. A good score on a test without content validity does not necessarily prove that the subject is well understood.
Content validity is often tested by a panel of experts in the subject or test analysts. Each question is rated essential, useful or irrelevant to the subject at hand, and is either kept on the test or removed from the test according to the average rankings put forward by the analysts. This is related to face validity, which asks test takers for a personal opinion on the subject matter covered by the test. It is different however, because of the use of expert analysts who rate the same questions on the same scale and those ratings are then compared.
A test can be scored with high content validity, sometimes just called content validity, or low content validity. It either offers a highly valid method of determining knowledge on the subject, or the test is written poorly and it does not cover all aspects of the subject. In the case of low content validity, the test should be reconstructed in order to test the full range of subjects as intended. The new test should then be evaluated for content validity.
Content validity should be tested before the test is ever given to test takers. If there is a question about the possible validity of a test after the test has been given, it can still be analyzed by a minimum of two independent panelists and its validity determined. In this case, however, if the test is found to have low validity the test must be rewritten, re-analyzed and then taken again by the test takers. It is much simpler to have the test analyzed before offering the test than to analyze after the scores of the test takers are not as predicted.