Questions: Measurement Invariance and Equivalence Across Groups

Question 1

A researcher compares depression scores between US and Japanese samples and finds US participants score significantly higher. The researcher established metric but not scalar invariance. What is the correct interpretation?

• A. The mean difference is valid because metric invariance ensures the scale works the same way in both cultures
• B. The mean comparison is problematic: without scalar invariance, item intercepts differ between groups, and the observed mean difference may reflect a measurement artifact rather than a true difference in depression
• C. The researcher should report the difference but note it is only approximate
• D. Metric invariance is always sufficient for observed mean comparisons; scalar invariance is only needed for latent mean comparisons

Answer: (B)

Metric invariance (equal factor loadings) means items respond to the underlying trait with the same sensitivity in both groups — the 'unit size' of the yardstick is equal. But scalar invariance (equal intercepts) ensures the 'zero point' is the same. Without equal intercepts, the scale's baseline differs between groups, meaning a given item score means something different (in absolute terms) across groups. Mean differences could be entirely due to these differing baselines rather than true differences in depression. Observed mean comparisons require scalar invariance.

Question 2

A researcher fits a configural CFA model and then constrains all factor loadings to be equal across groups. This sequence of model comparisons tests for:

• A. Configural invariance — whether the same factor structure holds in both groups
• B. Metric invariance — whether items respond to the latent factor with equal sensitivity across groups
• C. Scalar invariance — whether item intercepts are equal across groups
• D. Strict invariance — whether residual variances are equal across groups

Answer: (B)

Adding the constraint that factor loadings are equal across groups (while keeping the configural structure free) tests metric invariance. The factor loading (slope) captures how much item scores change per unit increase in the latent trait — equal loadings mean the unit of measurement is the same across groups. Configural invariance only requires the same pattern of loadings (which items load where). Scalar invariance goes further by also constraining intercepts. Strict invariance additionally constrains residual variances.

Question 3

Configural invariance requires only that the same general factor structure — which items load on which factors — holds across groups, without requiring any parameter values to be equal.

• T. True
• F. False

Answer: True

Correct. Configural invariance is the least restrictive level. It only requires that both groups show the same pattern of factor loadings (e.g., all four items load on a single depression factor in both groups), without requiring the loading magnitudes or intercept values to be equal. Configural invariance establishes that both groups are measuring something conceptually analogous. All higher levels of invariance (metric, scalar, strict) build on configural invariance by adding increasingly restrictive parameter constraints.

Question 4

Establishing metric invariance across two cultural groups is sufficient evidence to justify comparing their latent mean scores on a psychological construct.

• T. True
• F. False

Answer: False

Metric invariance (equal factor loadings) establishes that the scale has equal unit size in both groups, but it does not ensure equal origin points (intercepts). Latent mean comparison requires scalar invariance — equal intercepts AND equal loadings. Without equal intercepts, a group difference in observed item means could reflect different baseline item endorsements rather than a true difference in the latent trait. Scalar invariance is the empirical prerequisite for meaningful latent mean comparison. Metric invariance alone only supports comparing relationships among variables (e.g., correlations, regression coefficients).

Question 5

A cross-cultural study achieves metric but not scalar invariance. Modification indices reveal that two out of five item intercepts are non-invariant. Can the study still produce valid group comparisons, and if so, under what conditions?

Model answer: Yes — through partial scalar invariance. If at least two items have equal intercepts across groups, researchers can anchor latent mean comparisons on the invariant items while freeing the non-invariant intercepts. The comparison is valid but must acknowledge that the non-invariant items may reflect genuine cultural differences in item interpretation, not just measurement noise.

Partial invariance is common in applied cross-cultural research and is often scientifically defensible — a non-invariant intercept may indicate that a particular item captures a nuance of the construct that is culturally specific. The key requirements are: (1) at least two invariant items to identify the latent mean difference, (2) the non-invariant items should be freed in the model, and (3) the researcher should discuss whether the non-invariance reflects substantive cultural differences or measurement problems. Claiming full invariance when only partial invariance holds is a more serious error than acknowledging and reporting the partial structure.