What Is Convergent Validity?
Convergent validity is the extent to which a measure correlates positively with other measures that are theoretically expected to measure the same or a closely related construct. If you've built a new customer satisfaction scale, convergent validity asks: does it correlate with existing satisfaction measures, with repurchase intent, and with other indicators that should move in the same direction as satisfaction? High convergent validity means your measure agrees with other measures of the same construct, it's capturing the right thing. Along with discriminant validity, convergent validity is a key component of construct validity, providing evidence that your instrument measures what it claims to measure rather than something else.
Why Convergent Validity Matters in Research
A measure that doesn't correlate with other measures of the same construct is either measuring something different or measuring nothing at all. Convergent validity protects against the risk of building research programs, tracking dashboards, or business decisions on an instrument that looks good on the surface but doesn't actually capture the concept it claims to. It's especially important when introducing new or proprietary measurement tools that haven't been validated in published research.
How Convergent Validity Works
Assessing convergent validity requires collecting data with your measure alongside at least one other measure of the same or a related construct, then examining the strength of their association.
Correlation-Based Assessment
The simplest approach is to administer your new measure and an established measure of the same construct to the same sample, then calculate the Pearson correlation between them. For convergent validity, you'd expect a moderate to strong positive correlation, typically r = 0.50 or higher, though the threshold depends on how closely related the constructs are. Two measures of customer satisfaction should correlate highly (r > 0.60). A satisfaction measure and a loyalty measure might correlate moderately (r = 0.40-0.60) since they're related but distinct.
Average Variance Extracted (AVE)
For multi-item scales analyzed with confirmatory factor analysis, the average variance extracted (AVE) provides a more rigorous test. AVE represents the average amount of variance in the indicators that's explained by the latent construct. An AVE of 0.50 or higher means the construct explains more variance in its items than error does, the conventional threshold for adequate convergent validity. AVE below 0.50 suggests the items don't share enough common variance to claim they're measuring the same thing.
Multitrait-Multimethod Matrix (MTMM)
The gold standard for assessing both convergent and discriminant validity simultaneously is the MTMM matrix, proposed by Campbell and Fiske in 1959. You measure multiple traits using multiple methods, creating a correlation matrix that shows convergent coefficients (same trait, different methods, these should be high) and discriminant coefficients (different traits, same method, these should be lower). While MTMM is thorough, it's resource-intensive because it requires measuring multiple constructs with multiple methods in the same study.
Factor Loadings
In confirmatory factor analysis, convergent validity is also evidenced by the factor loadings of individual items on their intended factor. Standardized loadings of 0.50 or higher (ideally 0.70+) indicate that items converge on the construct they're supposed to measure. Items with low loadings aren't contributing meaningfully to the construct's measurement.
When to Use Convergent Validity Assessment
- Validating new measurement instruments before deploying them in ongoing research programs or tracking studies
- Adapting existing scales to new languages, cultures, or populations, convergent validity in the original context doesn't guarantee it holds after adaptation
- Evaluating vendor or syndicated measurement tools that your organization is considering adopting
- Scale shortening: when reducing a long scale to fewer items, checking that the short version still converges with the full version and related measures
- Publishing research where reviewers expect evidence that your measures capture the intended constructs
Common Mistakes to Avoid
- Using only one comparison measure: convergent validity is more convincing with multiple lines of evidence; correlating your scale with just one other scale leaves room for doubt about whether both scales share a method artifact rather than genuine construct overlap
- Accepting low correlations as evidence of convergent validity because the constructs are "related but distinct", there's a difference between moderate convergent validity (appropriate for related-but-different constructs) and weak convergent validity that suggests your measure isn't working; be honest about where the correlation falls relative to what theory predicts
- Ignoring measurement reliability when interpreting convergent correlations: the maximum possible correlation between two measures is constrained by their reliabilities; low correlations may reflect low reliability rather than poor convergent validity; correct for attenuation before drawing conclusions
How Quali-Fi Supports Convergent Validity
Quali-Fi's survey platform makes it straightforward to include multiple measurement instruments in a single study, your new scale alongside established measures, NPS, behavioral intent items, or other validators. The platform's real-time cross-tabulation and correlation tools let you monitor convergent relationships during data collection, and SPSS/CSV exports support confirmatory factor analysis and AVE calculations in your preferred statistical package.
Frequently Asked Questions
What correlation counts as "good" convergent validity?
There's no universal threshold because the expected correlation depends on how closely related the constructs are. Same construct measured two ways: r > 0.60 is expected. Closely related constructs: r = 0.40-0.60 is reasonable. The key is that the convergent correlations should be meaningfully higher than the discriminant correlations (correlations with unrelated constructs).
How is convergent validity different from reliability?
Reliability assesses whether a measure produces consistent results across items, occasions, or raters. Convergent validity assesses whether the measure agrees with other measures of the same construct. A measure can be highly reliable (internally consistent) but lack convergent validity if its items consistently measure something other than what's intended.
Can convergent validity be tested with a small sample?
Correlation estimates are unstable with small samples. For basic correlation-based convergent validity, you need at least 100 respondents for reasonable confidence in the estimate. For CFA-based approaches (AVE, factor loadings), 200+ is recommended. With very small samples, report confidence intervals around your convergent validity coefficients to acknowledge the uncertainty.
Related Topics
- Discriminant Validity
- Construct Validity
- Criterion Validity
- Reliability in Research
- Likert Scale
- Face Validity
Validating a new measurement instrument? See how Quali-Fi's multi-scale survey tools support validation research with built-in analytics and flexible exports.