Statistics glossary
Clear definitions of the terms in consumer and social-science research — what each one means, when it matters, and how to report it.
HTMT (Heterotrait-Monotrait Ratio)
HTMT is the heterotrait-monotrait ratio of correlations, a discriminant validity statistic. Values below .85 (strict) or .90 (liberal) support it.
Open average variance extractedAverage Variance Extracted (AVE)
Average variance extracted (AVE) is the mean squared loading of a construct's items, used for convergent validity. The .50 cutoff and how to report it.
Open composite reliabilityComposite Reliability (CR)
Composite reliability (CR) measures internal consistency from factor loadings. Definition, the .70 cutoff, and why it beats Cronbach's alpha.
Open cronbach alphaCronbach's Alpha
Cronbach's alpha measures internal consistency reliability. Definition, the .70 convention, its assumptions and limits, and how to report it.
Open mcdonald omegaMcDonald's Omega
McDonald's omega (ω) is a reliability coefficient from factor loadings that does not assume equal loadings. Cutoffs (.70, .80), omega vs alpha, reporting.
Open cfiCFI (Comparative Fit Index)
CFI compares your model's fit against a baseline model. Definition, the .95 cutoff from Hu & Bentler, and how to report it in CFA and SEM.
Open tliTLI (Tucker-Lewis Index)
TLI is an incremental fit index that penalizes model complexity. Definition, the .95 cutoff, and how it relates to CFI in CFA and SEM.
Open rmseaRMSEA (Root Mean Square Error of Approximation)
RMSEA (root mean square error of approximation) is an absolute fit index for CFA and SEM. Cutoffs .06 good, .08 acceptable, the 90% CI, how to report it.
Open srmrSRMR (Standardized Root Mean Square Residual)
SRMR is an absolute fit index based on residual correlations. Definition, the .08 cutoff from Hu & Bentler, and how to report it in CFA and SEM.
Open discriminant validityDiscriminant Validity
Discriminant validity means a construct is empirically distinct from others. Definition, how HTMT tests it, and how to report it.
Open convergent validityConvergent Validity
Convergent validity means a construct's indicators agree with each other. Definition, how AVE and loadings test it, and how to report it.
Open common method biasCommon Method Bias (CMB)
Common method bias is variance caused by the measurement method, not the constructs. Definition, why Harman's test is weak, and better remedies.
Open measurement invarianceMeasurement Invariance
Measurement invariance tests whether a scale measures the same way across groups. Definition, the configural/metric/scalar steps, and how to report it.
Open latent variableLatent Variable
A latent variable is a construct you cannot measure directly, inferred from observed indicators. Definition, role in SEM, and how to report it.
Open factor loadingFactor Loading
A factor loading is the strength of the link between an indicator and its latent factor. Definition, the .70 guideline, and how to report it.
Open mediationMediation
Mediation explains how X affects Y through a mediator M. Definition, the indirect effect with bootstrap CIs, and why it is not causal proof.
Open moderationModeration
Moderation is when the effect of X on Y depends on a third variable. Definition, the interaction term, and how to report it.
Open reflective vs formativeReflective vs Formative Indicators
Reflective indicators are caused by the construct; formative ones cause it. Definition, why it changes your validity tests, and how to choose.
OpenRun your whole study in one place
Design it, collect it, and get publication-grade analysis with a draft APA paragraph you can edit — free to start.
Start free