English · Русский · Українська

In 90 seconds

The validation programme turns the four-level framework into a sequence of answerable studies. It begins with clear constructs and reliable observations, then tests whether the proposed typological models add useful prediction. The programme does not produce a numerical pair verdict as an early deliverable.

Programme and gates

  1. Specify domain, population, outcome, and harm constraints.
  2. Define constructs and observable indicators independently of type labels.
  3. Establish content validity, inter-rater agreement, test-retest reliability, and measurement invariance where appropriate.
  4. Test convergent and discriminant validity against rival constructs.
  5. Test individual process hypotheses on held-out data.
  6. Test pair mechanisms and longitudinal outcomes.
  7. Replicate across settings before evaluating decision support.

Failure at a gate leads to revision or rejection, not automatic progression.

Shared example

For relocation, first define outcomes such as decision clarity, burden balance, factual accuracy, freely given consent, and later regret. Only then ask whether proposed strategic, operational, or tactical indicators predict distinct outcomes beyond resources, role, and prior experience.

Data and provenance

Use repeated tasks, self-report, partner report, independent coding, relevant records, and longitudinal outcomes where consent permits. Preserve instrument version, language, timing, missingness, exclusions, and the distinction between raw source and project interpretation.

Hypothesis families

  • H1: four-level annotation improves clarity and coder agreement.
  • H2: proposed indicators form distinguishable constructs.
  • H3: typology-informed models add prediction beyond simpler contextual models.
  • H4: pair-level mechanisms predict named outcomes across time.
  • H5: conversation guidance helps without increasing coercion or false confidence.

Rival models

Compare against demographics, Big Five or other established measures when relevant, relationship duration, skill, role, culture, resources, attachment, health, and a no-typology baseline. A complex model must earn its added complexity.

Stop rules and non-claims

Do not release ranking, matching, or automated go/no-go decisions from exploratory data. Stop or redesign when measures are unreliable, invariance fails, harms concentrate in a subgroup, consent is compromised, or prediction does not replicate.

Questions for every study

  • What precise outcome matters, to whom, and over what period?
  • What simpler explanation is the typology competing with?
  • What result would reject the hypothesis?
  • How will null, contradictory, and adverse findings be preserved?

Research outputs

Publish construct definitions, preregistrations, instruments, codebooks, anonymized analysis where lawful, calibration and validation reports, limitations, and negative findings. Keep normative and theological review distinct from psychometric review.

Next reading

Read Compatibility Measurement Roadmap and Compatibility Measurement Methods.