IPMA in SmartPLS 4: From the Map to Defensible Action Priorities

1. Start with the decision and the target

Suppose service quality and trust both relate to loyalty. Path coefficients describe relationships but do not show how customers currently rate those attributes. When quality is already strong and trust is weak, action priorities may differ from the ordering of coefficients. Importance-performance map analysis addresses this additional question (Ringle & Sarstedt, 2016).

Choose the outcome to improve before interpreting the map. Satisfaction, continued use and loyalty represent different targets. Each target has its own importance values; a ranking for satisfaction cannot simply be transferred to loyalty (Hair et al., 2018, Chapter 3).

This guide uses SmartPLS 4 for the workflow. The complete textbook chapter explains the calculations using an earlier software version; current settings are checked against SmartPLS documentation. Numerical worked examples below are constructed for teaching and are not client findings.

2. Read the two axes before the quadrants

Importance is the unstandardized total effect of a predecessor on the target, adding its direct effect and all indirect paths in the linear model. Do not substitute a standardized beta from the Path coefficients table. Interpretation depends on a compatible scaling of the constructs involved (Hair et al., 2018, Chapter 3).

Performance is the mean rescaled construct score on a 0–100 scale. It is not the percentage of satisfied respondents, a purchase probability or explained variance. A score of 60 locates the construct within its specified range. Construct scores depend on indicator weights and are not automatically arithmetic item means (Hair et al., 2018, Chapter 3).

The mean lines in the supplied illustration are orientation aids for the displayed constructs, not universal quality thresholds. Adding a construct can move a mean line and change a point's quadrant without changing its coordinates. Retain the numerical table and the set of included predecessors (Hair et al., 2018, Chapter 3).

3. Prepare the data and check the requirements

Keep the data, specified model, item wording, codebook and model assessment report together. IPMA inherits the meaning and limitations of that model. It cannot repair invalid measurement, miscoded variables or unjustified structural paths (Hair et al., 2018, Chapter 3).

Rescaling requires numerical distances with a defensible interpretation. Industry codes 1, 2 and 3 do not represent increasing performance. Explain the treatment of Likert spacing and verify that higher scores represent more of the intended attribute. Distinguish a positively valued attribute from a burden such as risk or inconvenience (Hair et al., 2018, Chapter 3).

Inspect outer weights as well as loadings. Negative weights can reflect coding, specification or collinearity and compromise a monotonic interpretation of the score. Do not take absolute values, clip scores or delete items merely to obtain an attractive map. Investigate the item and model, document changes and repeat the assessment (Hair et al., 2018, Chapter 3).

• Specify the target and decision first.
• Match theoretical item ranges to the codebook.
• Distinguish favorable attributes from adverse attributes such as cost or risk.
• Check reverse wording, negative weights and unresolved model problems.
• Record data, model and software versions.

4. Worked example: performance and total importance

Constructed example: three items on a theoretical 1–5 scale have means of 4.2, 3.4 and 3.8. Applying the rescaling formula in Hair et al. (2018, Chapter 3) gives item performance values of 80, 60 and 70. Keep the theoretical lower bound of 1 even if the sample's observed minimum is 2.

Suppose the weights have been converted to the appropriate units and normalized to sum to one: .20, .50 and .30. Construct performance is .20 × 80 + .50 × 60 + .30 × 70 = 67. Loadings cannot replace these weights, and equal weights cannot be assumed (Hair et al., 2018, Chapter 3).

Suppose X has a direct effect of .20 on Y and one indirect path through M with coefficients .40 and .50, all unstandardized on the compatible IPMA scale. Total importance is .20 + .40 × .50 = .40. Include every specified indirect path rather than selecting only statistically significant ones (Hair et al., 2018, Chapter 3).

P_item = 100 × (mean_item − min_theoretical) / (max_theoretical − min_theoretical)
P_X = 0.20 × 80 + 0.50 × 60 + 0.30 × 70 = 67
Importance(X → Y) = 0.20 + 0.40 × 0.50 = 0.40
Model-implied change in Y for +5 performance points in X = 5 × 0.40 = 2

5. SmartPLS 4 workflow and outputs to retain

SmartPLS offers All predecessors and Direct predecessors. This controls which constructs appear on the map; it does not switch importance from total effects to direct effects. Record the scope when comparing maps (SmartPLS, n.d.).

6. Negative importance and the limits of indicator priorities

E in the supplied illustration has negative importance. Establish whether it measures a desirable attribute or a burden. If it measures risk, reducing rather than increasing it may be the relevant direction. An unexpected negative effect for quality requires checks of coding, collinearity, offsetting direct and indirect paths and uncertainty. A position on the left does not by itself establish irrelevance (Hair et al., 2018, Chapter 3).

Indicator importance combines a rescaled weight with the construct's unstandardized total effect. In the constructed example, the second item has importance .50 × .40 = .20. This describes the model's scoring relationship. In a reflective scale, changing an item's wording or inducing higher responses is not equivalent to improving the underlying service (Hair et al., 2018, Chapter 3).

7. Make priorities conditional on uncertainty and feasibility

Do not rank nearby estimates rigidly. Inspect resampling evidence for the correct parameter and scale. A standardized beta interval is not an interval for an unstandardized total effect. Check what the installed software actually outputs; if a corresponding interval is unavailable, report that limitation rather than inventing one.

Statistics and resource decisions are separate steps. High importance and low performance identify a candidate only when measurement is defensible, the relationship is sufficiently stable and the attribute can be changed. Costs, timing, feasible gains and risks are additional managerial judgments, not quantities automatically supplied by IPMA.

Assess sensitivity using substantively justified alternatives such as another target or defensible model specification. Keep the same numerical table when comparing maps. Deliberately incorrect scale bounds are coding errors, not legitimate robustness specifications.

8. Reporting template and reproducibility checklist

Replace the brackets below with actual findings. The numerical teaching example and supplied illustration are not evidence from your own study. Identify the predecessor scope, coding direction and unstandardized total effects explicitly (Hair et al., 2018; Ringle & Sarstedt, 2016).

• Identify the target, predecessor scope and importance scale.
• Retain theoretical bounds, coding and missing-data decisions.
• Keep the numerical table alongside the map.
• Explain the reference lines without treating them as universal cutoffs.
• Discuss negative effects and the limits of item-level actions.
• Report uncertainty evidence or its absence.
• Retain the model, data version, settings and decision history.