# 23. PLS-POS

## Purpose and applicability

PLS-POS searches for prediction-oriented segments whose memberships improve a PLS objective. This tutorial asks whether a stable two-segment structure can be reproduced in the organizational-identification model.

Use PLS-POS as an exploratory heterogeneity procedure when theory permits unobserved segments and the sample can support repeated multi-start estimation. Do not treat the lowest objective value as sufficient: convergence, segment size, optimum reproduction, and stability requirements must also pass.

## Practice study

- Dataset: [organizational-identification-model-comparison.csv](../samples/organizational-identification-model-comparison.csv)
- Prepared project: [organizational-identification-multimod.qpls](../projects/organizational-identification-multimod.qpls)
- Observations: 305 complete cases
- Data type: deterministic synthetic instructional data
- Candidate: PLS-POS, K = 2, 20 deterministic starts
- Research question: Does the fitted model support a reproducible two-segment solution?

The model contains Organizational Prestige (8 reflective indicators), Organizational Identification (6), Affective Commitment (Joy) (4), and Affective Commitment (Love) (3). Its paths are Organizational Prestige → Organizational Identification and Organizational Identification → both commitment dimensions.

## Preparation

1. Open the prepared project, then open **Data** and confirm 305 observations.

   ![Organizational-identification data prepared for PLS-POS](../screenshots/pls-pos/01-data-ready.png)

2. Return to **Model** and confirm the saved four-construct diagram. Its coordinates are already arranged and fitted; do not press **Arrange**.

   ![Saved arranged model for PLS-POS](../screenshots/pls-pos/02-arranged-model.png)

3. Choose **Validate** and confirm no blocker. Segmentation requires a valid ordinary PLS model before any candidate is meaningful.

## Calculation

4. Choose **Calculate → PLS-POS** in the shared MultiMod workspace.
5. Select K = 2 and retain 20 deterministic starts. Review the minimum segment-share and optimum-reproduction requirements in the setup rather than changing them to force acceptance.

   ![PLS-POS K = 2 multi-start setup](../screenshots/pls-pos/03-calculation-setup.png)

6. Leave **Open Results when finished** selected and choose **Start calculation** once. Wait for all starts and candidate diagnostics to finish.

## Results and interpretation

7. Open **Candidate segmentation diagnostics**. Check algorithm, K, completion state, converged starts, stable starts, and objective before reading information criteria.

   ![PLS-POS candidate diagnostics](../screenshots/pls-pos/04-results-primary.png)

8. Open **Candidate blockers**. In the qualified teaching run, 10 of 20 starts converged, but only one reproduced the optimum while two were required. The candidate is therefore correctly marked unstable and is not locked.

   ![PLS-POS stability blocker](../screenshots/pls-pos/05-results-secondary.png)

An unstable candidate is a valid scientific outcome, not an application failure. Do not interpret segment-specific paths, memberships, or substantive segment labels when the stability gate withholds the locked solution.

## Reporting and reproducibility

Report the algorithm, K, number of starts, seed, convergence count, reproduced-optimum count, stability threshold, objective/criteria retained, and the decision not to lock the solution. Export candidate diagnostics and blocker evidence; preserve the saved project so the exact result and provenance can be reopened.

Common mistakes include selecting K from one criterion alone, reducing the stability threshold after seeing the result, assigning substantive names to unstable segments, and confusing observed groups used by MGA with latent segments sought by PLS-POS.

## Limitations and related methods

This 305-case deterministic study is retained precisely because its K = 2 candidate does not pass the stability gate; it demonstrates how QuickPLS prevents overinterpretation. A different research dataset may support a locked solution, but stability must be established rather than assumed. For a mixture-model alternative, see [FIMIX-PLS](24-fimix-pls.md). For known groups, use [MICOM](19-micom.md) and [PLS-MGA](20-pls-mga.md).

Last verified with QuickPLS 2.62.8.