Example Of Discriminant Validity | Simple Scale Check

Discriminant validity tells you whether two constructs that should differ in theory also differ in the data. When you design a survey scale or test, you want each construct to capture its own concept instead of mirroring another scale by accident.

Without sound discriminant validity, a model of latent variables turns fuzzy. Paths in a structural equation model may look strong, yet the underlying constructs blur together, so you cannot tell which one really explains an outcome.

What Discriminant Validity Means

Discriminant validity sits inside construct validity. It checks whether a construct is distinct from other constructs that sit in the same conceptual space and are measured in the same project.

Many researchers first hear the term beside convergent validity. Convergent validity asks whether indicators that should relate to a construct actually hang together. Discriminant validity asks a different question: whether constructs that should not overlap too much stay separate.

Discriminant Validity Versus Other Validity Types

The table below places discriminant validity beside other common validity checks so you can see where it fits inside a full measurement plan.

Validity Type	Main Question	Typical Evidence
Discriminant	Do related constructs stay distinct from each other?	Correlations, Fornell–Larcker, HTMT ratio
Convergent	Do indicators of one construct move together?	Average variance extracted, factor loadings
Content	Do items cover the full domain of the construct?	Expert review of items and definitions
Criterion	Does the scale relate to an external benchmark?	Correlation with a gold standard or outcome
Face	Does the scale look reasonable to respondents?	Pilot feedback and simple item checks
Internal Consistency	Do items within a scale hang together?	Cronbach’s alpha, composite reliability
Test–Retest	Is the scale stable across time?	Correlation between repeated administrations
Predictive	Does the scale predict later outcomes?	Regression or structural paths to later scores

From this overview, you can see that discriminant validity deals with the pattern of relations between constructs rather than the strength of relations inside a single scale.

Discriminant Validity Example In Practice

Consider a study that measures job satisfaction and job stress among employees. The theory says that these two constructs relate to each other but are not identical. Higher stress often comes with lower satisfaction, yet a worker can feel stressed and satisfied at the same time.

Researchers create two multi item Likert scales. One scale measures job satisfaction through items on enjoyment, pride, and willingness to stay. The second scale measures job stress through items on pressure, overload, and conflict at work.

Step 1: Define The Constructs Clearly

The team writes short, precise definitions for job satisfaction and job stress. Each definition lists what the construct includes and what it excludes. This step shapes the pool of items and sets up a clear expectation that satisfaction and stress should show moderate negative correlation, not near perfect overlap.

Step 2: Write Items For Each Scale

Items for job satisfaction might say, “I enjoy my daily tasks” or “I would recommend this workplace to a friend.” Items for job stress might say, “My workload feels heavy” or “Deadlines at work feel hard to meet.” Each item clearly belongs with only one construct.

Step 3: Collect Data And Inspect Correlations

After collecting responses from several hundred workers, researchers compute scale scores. They then review the correlation matrix. Items within the same scale show high positive correlations, while satisfaction items show moderate negative correlations with stress items.

At the scale level, the correlation between job satisfaction and job stress might sit near −0.50. This value shows that the constructs relate in a meaningful way but are far from being simple reflections of each other.

How This Scenario Shows Clear Discriminant Validity

This job satisfaction and job stress project gives a concrete example of discriminant validity. The constructs relate but do not collapse into one. Respondents answer satisfaction items differently from stress items, and the pattern of correlations reflects the theory.

In text, a researcher might write, “Evidence of discriminant validity appears in the moderate negative correlation between job satisfaction and job stress, which stays well below values that would indicate redundancy.”

How To Check Discriminant Validity With Numbers

Evidence for discriminant validity gains strength when you back the story with formal tests. Three common approaches rely on correlations, average variance extracted, and the heterotrait–monotrait ratio.

Simple Correlation Thresholds

Many applied studies start with a simple rule of thumb. If the correlation between two constructs is below about 0.85, researchers treat discriminant validity as acceptable. When correlations rise above that level, the constructs may overlap too much.

This practical rule should never stand alone. Correlations depend on sampling error, measurement error, and the true relation between constructs. A low correlation might come from poor items, and a high correlation might reflect a strong yet still conceptually clear link.

Fornell–Larcker Criterion

The Fornell–Larcker criterion uses the average variance extracted for each construct. For a given pair of constructs, the square root of each AVE should be greater than the correlation between the two constructs.

To apply this rule, you first compute AVE for each construct in a confirmatory factor model. Then you write a table in which diagonal cells carry the square root of AVE and off diagonal cells carry correlations between constructs. When each diagonal value exceeds the correlations in its row and column, discriminant validity looks stronger. A detailed walk through appears in many structural equation modeling manuals and in resources such as the Fornell–Larcker overview on Wall Street Mojo, which describes the logic and steps behind this criterion.

HTMT Ratio For Discriminant Validity

More recent work recommends the heterotrait–monotrait ratio of correlations, often shortened to HTMT. HTMT compares the average correlation between indicators of different constructs with the average correlation between indicators of the same construct.

Henseler, Ringle, and Sarstedt proposed HTMT in 2015 and showed that it detects lack of discriminant validity more reliably than the Fornell–Larcker criterion. In practice, many authors treat HTMT values below about 0.85 as evidence for discriminant validity. Values near or above 0.90 often raise concern that two constructs do not differ enough.

Software packages make HTMT straightforward to compute. One resource is the SmartPLS documentation on discriminant validity assessment, which explains the calculations and output. The open access article on HTMT by Henseler, Ringle, and Sarstedt gives the original simulation based argument.

Reading Discriminant Validity Evidence

Technical rules matter, yet they always sit beside theory. When you interpret discriminant validity checks, think about both numerical thresholds and the story of your constructs.

Indicator	Pattern That Fits Distinct Constructs	Pattern That Signals Overlap
Simple Correlations	Moderate values, often below 0.85 in absolute terms	Values close to 0.90 or above
Fornell–Larcker	Square root of AVE exceeds construct correlations	Square root of AVE lower than some correlations
HTMT Ratio	HTMT below about 0.85, or at least below 0.90	HTMT at or above 0.90 for a construct pair
Cross Loadings	Each item loads highest on its intended construct	Several items load higher on another construct
Theory	Definitions clearly distinguish the constructs	Definitions re use many of the same phrases
Model Fit	Good fit with separate constructs retained	Fit hardly changes when constructs merge
Item Wording	Items target distinct aspects of the topic	Items for different constructs sound almost identical

When several indicators in the left column point in the right direction, confidence in discriminant validity grows. When many entries in the right column show up at once, the constructs in question may need to be redefined or combined.

Example Of Discriminant Validity In A Report

Suppose the job satisfaction and job stress study reaches the reporting stage. The researcher now needs to present this concept in clear academic prose.

A typical paragraph in the methods or results section might say that a confirmatory factor analysis supported separate factors for satisfaction and stress. The author would then mention that the correlation between the two latent variables stayed below 0.85, the square roots of AVE exceeded shared variances, and HTMT values fell below 0.90.

Notice how this set of results ties theory, measurement, and data together. The theory predicted related but distinct constructs. The measurement model placed items in two scales that match those definitions. The statistical checks then confirmed that satisfaction and stress remained separate in the data.

Common Pitfalls When Assessing Discriminant Validity

Because discriminant validity depends on several moving parts, researchers fall into predictable traps. Recognizing these issues early can save time and prevent misleading claims.

Using Only One Numerical Rule

One trap involves relying on a single threshold, such as a correlation below 0.85, while ignoring other checks. A model can pass a simple correlation test yet fail HTMT or Fornell–Larcker checks.

A stronger approach gathers multiple lines of evidence. When correlations, AVE comparisons, and HTMT all point to the same conclusion, readers can trust the result much more than when only one index is reported.

Ignoring Theory And Item Content

Another trap flows from a narrow focus on numbers. Even if HTMT sits below 0.85, item wording might still be nearly identical across scales. In that case, respondents may not perceive a real distinction between constructs.

Careful reading of items, alongside definitions, protects against this problem. If items for different constructs feel interchangeable, the concept map should be revised and the item pool adjusted before heavy modeling work begins.

Overreacting To Slight Threshold Violations

A final trap appears when a construct pair slightly misses a threshold, such as an HTMT value of 0.86. Some readers treat any value above 0.85 as automatic failure, yet thresholds arise from convention, not hard natural limits.

In such cases, you should weigh the size of the violation, sampling error, and theory. If constructs are almost identical in meaning and HTMT sits near 0.90, a change in the measurement model may be wise. If constructs differ clearly in content and HTMT only slightly exceeds a rule of thumb, a transparent justification can be reasonable.

Quick Checklist For Discriminant Validity

The following checklist condenses the main ideas so you can apply them in your next measurement project or thesis.

Before Data Collection

• Write clear definitions that separate each construct from related constructs.
• Design items so that each item fits only one construct and avoids overlap.
• Ask content experts to review items for redundancy across scales.

During Analysis

• Inspect the correlation matrix for very high correlations between constructs.
• Compute AVE and apply the Fornell–Larcker criterion.
• Compute HTMT values and compare them with recommended thresholds.
• Review cross loadings to check that items load highest on their intended constructs.

When Reporting Results

• Describe at least one numerical check and one item based check.
• Connect discriminant validity evidence back to your construct definitions.
• Present a brief example of discriminant validity, such as the relation between job satisfaction and job stress, to make the idea tangible for readers.

Handled with care, an example of discriminant validity does more than tick a methodological box. It reassures readers that each construct in a model carries its own meaning, so any conclusions drawn from that model rest on clear and distinct building blocks.