#SmartPLS4 Series 7.2 - Concept and Assessment of Construct Reliability

Construct reliability in reflective measurement models hinges on two checks: whether indicators reliably reflect their latent construct, and whether the construct’s indicators hang together consistently. The first step is indicator reliability, which is assessed by how much of each indicator’s variance is explained by its construct. In practice, this means squaring the indicator loadings (the loading is the correlation between indicator and construct) to obtain explained variance, or communality. Loadings above 0.708 are the benchmark because they imply the construct explains more than 50% of the indicator’s variance—an acceptable level of indicator reliability.

When an indicator’s loading falls below the threshold, removing it is not automatic. Researchers are advised to examine how deleting the indicator affects both reliability and validity. In social science work—especially with newly developed scales—loadings between 0.4 and 0.708 may be candidates for removal only if the deletion improves internal consistency reliability or convergent validity. Content validity adds another constraint: if an item’s wording or meaning is important to the construct’s coverage, low-loading indicators may need to stay even if they underperform statistically. A hard rule is also given for extremely weak indicators: items with loadings below 0.4 should be eliminated from the measurement model.

After indicator reliability comes internal consistency reliability, which evaluates whether the set of indicators consistently measures the construct. In SmartPLS, the primary metric is composite reliability (often written as CR or ρc). Higher values indicate stronger internal consistency, with values above 0.70 typically considered reliable. The guidance is more nuanced: 0.60–0.70 is acceptable for exploratory research; 0.70–0.90 is satisfactory to good; and values above 0.90—especially above 0.95—can be problematic because they suggest redundancy among indicators. That redundancy can inflate correlations among indicator error terms, a phenomenon linked to “straightlining” response patterns.

Cronbach’s alpha (α) is another internal consistency measure, using similar thresholds to composite reliability. However, alpha assumes tau-equivalence: that all indicator loadings are equal in the population. When that assumption fails, alpha tends to produce lower reliability estimates than composite reliability. Composite reliability is described as more liberal, Cronbach’s alpha more conservative, with the construct’s “true” reliability falling between them. SmartPLS also reports rho_A (ρA), a reliability coefficient positioned as an intermediate compromise between alpha and composite reliability.

In the workflow described, SmartPLS4 output is accessed via the report under “Quality criteria → Construct reliability and validity → Overview.” The example results show rho_A sitting between Cronbach’s alpha and composite reliability, supporting the conclusion that the construct is reliable. The overall takeaway is procedural: verify indicator loadings first (with communality logic and content-validity safeguards), then confirm internal consistency using CR, alpha, and rho_A while watching for signs of redundancy.