Reading focus
Syllabus: How Do We Know What a Psychological Test Measures?
Follow how the nomological network connects observations, theoretical expectations, and revisions to an interpretation.
Why you’re reading it. Move from transparent measurement choices to an argument about what a score means.
A question to consider. Which predicted relationships would help distinguish persistence from prior attainment, and what would a failed prediction require you to reconsider?
Why this reading
Cronbach and Meehl ask how evidence can justify interpreting a test in terms of a psychological construct when no definitive criterion is available. Their account of a nomological network connects measurement to theoretical relationships. Read the original to see why a convenient score, high internal consistency, or one predictive correlation cannot carry every interpretive claim.
Before you read
Be comfortable with the distinction between an observed score and a psychological construct, with basic correlations, and with the difference between reliability and validity, which are not interchangeable.
The paper was written as part of the committee work that produced the 1954 Technical Recommendations for psychological tests. Its distinction between predictive, concurrent, content, and construct validity was a proposal, not yet a convention.
Read for
- Why the authors argue that internal consistency alone cannot establish what a test measures, and what they think it can establish.
- The concept of a nomological network: how measurement claims are embedded within broader theoretical networks, and what happens to the construct when a predicted relation fails.
- The "bootstraps" argument: how a construct can be validated by a test that is itself imperfect, without circularity.
- An assumption later work challenges: that validity is a property of interpretations and inferences rather than of the test itself.
Consider afterwards
- Suppose a new resilience questionnaire has highly correlated items and strongly predicts well-being. What additional evidence would you want before accepting the interpretation that the questionnaire measures resilience?
- Can a construct be scientifically useful before researchers agree on exactly what it is? What does the paper's answer commit you to?
- When a predicted relation in the network fails, the authors say we may revise the theory, the test, or the inference. How would you decide which?
Editorial context
The paper's account of theoretical networks belongs to its historical setting, but the problem of justifying interpretations of scores remains central. The 2014 Standards for Educational and Psychological Testing, chapter 1, organize validation around evidence for specified score interpretations and proposed uses. That emphasis should not be reduced to a slogan that a test is simply valid or invalid. Different interpretations and uses require their own supporting arguments.