Advanced analyses: MGA, IPMA and NCA

A solid structural model answers your hypotheses. Advanced analyses add depth that reviewers and practitioners value: do the relationships hold for all groups of respondents, which constructs should managers improve first, and which conditions are necessary for the outcome? Each runs in its own free tool, with built-in example data you can explore before using your own.

Multi-group analysis (MGA)

When a categorical variable such as gender, country, channel or experience might change the relationships, MGA estimates the model separately in each group and tests whether the path coefficients differ. This is categorical moderation, and it lets every path differ at once.

MGA has two stages. First, measurement invariance of composites (MICOM) checks that the constructs mean the same thing in both groups: configural invariance (same model and settings, guaranteed by the tool), compositional invariance (the composite formed with group 1’s weights correlates almost perfectly with the composite formed with group 2’s weights, tested with permutations), and equal composite means and variances. Steps 1 and 2 establish partial invariance, enough to compare paths; all three establish full invariance.

In the example (420 customers, half online and half in store), MICOM establishes full invariance. The trust → satisfaction path is much stronger online (β = 0.57) than in store (β = 0.22); the difference of 0.34 is significant in all four tests (permutation, Henseler’s PLS-MGA, parametric and Welch-Satterthwaite; p < .001). Satisfaction → loyalty is somewhat weaker online (0.35 versus 0.49), but the difference is not significant (permutation p = .089). Note the logic: “significant in one group but not the other” does not show a difference; only the test of the difference does.

Importance-performance map analysis (IPMA)

Path coefficients tell you what matters; IPMA adds how well each driver is doing. For a target construct, it plots each predecessor’s importance (its total effect on the target) against its performance (its average score rescaled to 0–100). Constructs with high importance and low performance are the top priorities.

In the example, trust and satisfaction drive loyalty. Satisfaction has an importance of 0.478 (a one-point increase in satisfaction performance raises loyalty by 0.48 points on the 0–100 scale) and a performance of 52.4, below average: the top priority. The indicator-level table goes further and shows which satisfaction items to work on. For your course model, the target would be use behaviour or intention, and the map would tell an app developer whether to invest in usefulness, ease of use or trust.

cIPMA and necessary condition analysis (NCA)

Regression-based methods like PLS-SEM follow “more is better” logic: more of X goes with more of Y on average. Necessary condition analysis (Dul, 2016) asks a different question: is a certain level of X required for a given level of Y, so that its absence cannot be compensated by other factors? In the NCA example, R&D investment (d = .29) and employee skills (d = .24) are necessary conditions for high innovation performance, while leadership support is not (d = .04, p = .520). The bottleneck table shows, for each desired level of the outcome, the minimum level of each condition. The combined IPMA (Hauff et al., 2024) adds necessity to the importance-performance map, so a construct can be both a driver and a must-have.

Other extensions in the Studio

• fsQCA: the Configurations tab calibrates construct scores into fuzzy sets and finds combinations of conditions that lead to the outcome, a configurational complement to PLS-SEM.
• Latent variable scores: export them from the Export tab to use the constructs in other analyses, such as cluster analysis to find segments.
• Higher-order constructs: model a broad construct with several dimensions using the repeated indicators or two-stage approach.

When to use which