Questions: Convergent and Discriminant Validity: Multitrait Analysis

Question 1

A researcher develops a new measure of 'resilience' and finds it correlates r = .78 with established resilience scales, but also correlates r = .74 with measures of optimism, r = .71 with extraversion, and r = .69 with self-efficacy. What validity problem does this pattern reveal?

• A. Poor convergent validity — the new measure does not correlate strongly enough with established resilience scales
• B. Poor discriminant validity — the measure fails to distinguish resilience from conceptually distinct constructs
• C. Poor face validity — the items do not look like they measure resilience
• D. Poor test-retest reliability — correlations with other measures will change over time

Answer: (B)

The pattern shows strong convergent validity (high correlation with similar constructs) but poor discriminant validity. If resilience is supposed to be a distinct construct from optimism and extraversion, those correlations should be substantially lower than the convergent ones. When a measure correlates nearly equally with theoretically distinct constructs, it may be measuring something broad like general positive affect or social desirability — or the construct boundaries are poorly defined. Convergent evidence alone is not sufficient; you need the pattern of correlations to be selectively high for related measures and lower for unrelated ones.

Question 2

What is the primary purpose of the multitrait-multimethod (MTMM) matrix in establishing construct validity?

• A. To test whether a measure has high internal consistency across its items
• B. To simultaneously assess convergent and discriminant validity across multiple constructs measured by multiple methods, separating construct variance from method variance
• C. To identify which factor structure best describes the items in a psychological scale
• D. To compare the predictive validity of two different measures against the same external criterion

Answer: (B)

The MTMM matrix, developed by Campbell and Fiske, measures multiple constructs using multiple methods (e.g., self-report, observer rating, physiological measure). This design allows you to check whether high correlations reflect shared construct (good convergent validity) or merely shared method (two self-report measures inflating each other through method variance). Discriminant validity is supported when measures of different constructs using the same method correlate less than measures of the same construct using different methods. Internal consistency, factor analysis, and predictive validity are separate forms of validity evidence.

Question 3

A new anxiety scale that correlates r = .82 with an established anxiety measure has demonstrated sufficient construct validity to use in research.

• T. True
• F. False

Answer: False

High convergent validity is necessary but not sufficient for construct validity. Demonstrating that the new scale correlates with an existing anxiety measure shows it measures something similar — but without discriminant validity evidence, we cannot rule out that it is simply measuring general distress, negative affect, or neuroticism. A measure that correlates equally with anxiety, depression, and worry scales has strong convergent but poor discriminant validity, undermining the claim that it specifically captures anxiety as a distinct construct. Both forms of evidence are required.

Question 4

Method variance can inflate convergent validity correlations between two self-report measures even when the measures are designed to assess distinct psychological constructs.

• T. True
• F. False

Answer: True

This is a core insight motivating the MTMM framework. Two self-report questionnaires share systematic response tendencies — acquiescence bias, social desirability, positive self-presentation — that inflate their intercorrelations regardless of construct content. If a 'resilience' scale and a 'well-being' scale are both self-report, they will correlate partly because of what they share as method, not purely because the constructs overlap. This is why using multiple methods (behavioral observation, physiological measures, informant report) is important: convergent validity across methods provides stronger evidence than convergence within a single method.

Question 5

Why is it problematic if a psychological measure correlates just as highly with theoretically unrelated constructs as it does with the construct it is supposed to measure?

Model answer: If a measure correlates equally with related and unrelated constructs, it fails to demonstrate discriminant validity — meaning it cannot distinguish between distinct psychological entities. This suggests the measure is capturing something broad (e.g., general distress, response bias) rather than the specific construct it claims to assess. Without discriminant evidence, the construct itself is empirically undefined: if 'resilience' predicts everything equally well, it is not a specific, distinct psychological reality but merely a label for a diffuse pattern.

This question targets the logical structure of construct validation: a construct is meaningful only insofar as it is distinguishable from other constructs. Convergent validity shows that a construct is real; discriminant validity shows that it is distinct. A measure that converges with everything proves that something is being measured but not what. The scientific value of a construct depends on its discriminant boundary being as real and empirically supported as its convergent core.