Definition

• The degree to which a test measures what it is supposed to measure

Types of Validity

• Content - The validity of specific items of a test (includes face validity)
• Construct - the degree to which the test relates to the underlying theoretical construct of the test
• Criterion related - (the most measurable type) - The ability of a test to accurately predict performance on a real life criterion

Criterion Related Validity

• Concurrent - attempts to give an indication of current state of affairs (e.g., diagnostic and IQ tests)
• Predictive - test attempts to predict future behavior (eg., SATs, suicide scales).

Criterion Contamination

• Occurs when knowledge of a test score influences the criterion score
• This factor can artificially inflate validity

The Validity Coefficient

• R is the validity coefficient. When this is squared (R2) you get an indication of the variance of test scores explained by the test
• Example: If R = .8, then the squared validity coefficient would be .64 - This means 64% of the variance can be explained by the test

Classical Decision Theory

• Relates to how validity affects the tests we use to make selection decisions
• Example: How SAT scores used for college admission decisions
• Base rate is the number of successes occurring naturally in the population. (In the previous example it is 14 out of 21 for a success rate of selection = .67. The test increases the base rate to 8 out of 9 selected for a base rate of .89
• Selection Ratio - that proportion of the applicant or sample pool that is accepted.
• Both a high cut-off point and a low selection ratio increase the success rate (this is why Ph.D. programs have a large pool of applicants for just a few positions)

Other Types of Validity

• Convergent validity relates to how well a test correlates with other measures of the same construct
• Discriminant validity - a test should not correlate highly with tests unrelated to the construct at hand

The Relationship Between Reliability and Validity

• A test must be reliable to be valid (but not vice versa) However, a reliable test does not need to be valid.
• Maximum validity:
  Rmax = R11 R22 (R11 and R22 are the reliability coefficients of the two variables)