# 18. Prediction-Oriented Model Selection

## Purpose

Which of three organizational-identification models should be retained when equation-level predictive evidence is the primary criterion?

## Practice study

- File: [organizational-identification-model-comparison.csv](../samples/organizational-identification-model-comparison.csv)
- Prepared project: [organizational-identification-poms.qpls](../projects/organizational-identification-poms.qpls)
- Observations: 305
- Data type: deterministic synthetic instructional data
- Complexity: a realistic multi-construct model with multi-item measurement blocks

**Prerequisites**

- Use the prepared project because it retains three compatible candidate models, their ordinary PLS results, and the qualified POMS result

## Exact procedure

1. Start QuickPLS and choose **Open Project**. Select [organizational-identification-poms.qpls](../projects/organizational-identification-poms.qpls). It contains three saved candidates built from the same data and reflective measurement blocks.

2. Open **Data** and confirm the linked **organizational-identification-model-comparison.csv** dataset contains 305 observations and 22 variables. The CSV is also supplied separately for inspection.

   ![Prediction-Oriented Model Selection imported data](../screenshots/prediction-oriented-model-selection/01-data-ready.png)

3. Confirm the reflective measurement blocks: Organizational Prestige (org_pre1–org_pre8), Organizational Identification (org_ident1–org_ident6), Affective Commitment (Joy) (ac_joy1–ac_joy4), and Affective Commitment (Love) (ac_love1–ac_love3).

   The baseline estimates Prestige → Identification and Identification → Joy and Love. Candidate 2 adds Prestige → Joy; Candidate 3 instead adds Prestige → Love.

4. The prepared project stores an arranged layout for all three models. Review the active baseline diagram; do not press **Arrange** or edit a candidate before the tutorial.

   ![Prediction-Oriented Model Selection prepared model or variable roles](../screenshots/prediction-oriented-model-selection/02-arranged-model.png)

5. Confirm all three candidate names are available. Do not save the unchanged prepared project.

6. Use **Calculate → Prediction-Oriented Model Selection**.

   ![Prediction-Oriented Model Selection calculation setup](../screenshots/prediction-oriented-model-selection/03-calculation-setup.png)

7. Complete the method-specific settings:

- Select all three compatible saved models.
- Retain the common Organizational Identification, Affective Commitment (Joy), and Affective Commitment (Love) outcomes.

8. Read **Readiness**, leave **Open Results when finished** selected, and choose **Start calculation** once. Wait for **Completed**; do not close the project while the result is being saved.

9. In **Results**, confirm **Prediction-Oriented Model Selection** is selected and inspect these outputs in order:

1. **Selection overview** for the overall decision.
2. **Outcome decisions** for the winner of each endogenous equation.
3. **Candidate BIC values** for the underlying evidence.
4. **Candidate ledger** and **Run details** for exact provenance.

   ![Prediction-Oriented Model Selection primary result](../screenshots/prediction-oriented-model-selection/04-results-primary.png)

   ![Prediction-Oriented Model Selection secondary result](../screenshots/prediction-oriented-model-selection/05-results-secondary.png)

10. Use **Copy** or **Export** for the required table. The installed calculation, canonical result, formatted copy, and project persistence passed. **Save Report** returned operating-system error 3 in the isolated disposable qualification path, so report persistence is not claimed for this fixture; do not treat a failed report dialog as a calculation failure.

## Reading guidance

- The data are deterministic synthetic teaching data. They are suitable for reproducing the workflow, not for substantive publication claims.
- “Mixed evidence” is meaningful here: different candidates win different endogenous equations, so QuickPLS does not force a single overall winner.