Validity
Psychology of Testing & Measurements
Lecture, Chapter 5

What is validity?
Validity is the degree to which a test measures what it is intended to measure.
Validity is the evidence for inferences made about a test score.
    Content-related evidence
    Criterion-related evidence
    Construct-related evidence
Validity should be viewed systemically.

Face Validity
The appearance that a test measures what it is intended to measure.

Involves an overview of a measurement to determine that a measurement appears valid “at face value.

Content-Related Evidence for Validity
The content of a test adequately represents the body of knowledge or the attribute for which the test has been written. Has the test been constructed properly?
    Construct underrepresentation
    Construct-irrelevant variance

Criterion-Related Evidence for Validity
Evidence that a test correlates with a specific (well-defined) criterion.
    Predictive validity evidence
    Concurrent validity evidence

Construct-Related Evidence for Validity
Constructs are not clearly defined nor do they have established criterion against which validity can be measured.
Construct validity evidence is assembled through a series of activities involved in showing a relationship between the test and other measurements. Over a series of related studies, the meaning of a test is established, piece by piece.
    Convergent evidence
    Discriminant evidence or divergent validation
Most important type of validity!

Validity Coefficients
Correlation coefficient between a test and a criterion.
.30 to .40 are acceptable as “high” (note that this is much lower than acceptable reliability coefficients).
The squared validity coefficient indicates the percentage of variation in the criterion that is attributable to the test score.

Relationship between Reliability and Validity
To establish validity, a test must first be reliable. Once reliability is established, however, validity is NOT guaranteed.
Maximum validity coefficient between 2 variables is equal to the square root of the product of their reliabilities.
Because validity coefficients are not expected to be high, modest validity between true scores on two traits can be missed if the tests for the traits are not highly reliable.