spectralbranding/exp-cross-domain-primacy

---
license: cc-by-4.0
task_categories:
- text-generation
language:
- en
tags:
- brand-perception
- spectral-brand-theory
- serial-position-effect
- primacy-bias
- moral-foundations-theory
- political-attitudes
- cross-domain
pretty_name: "Experiment F2: Cross-Domain Primacy (Brand vs Political Attitudes)"
size_categories:
- 1K<n<10K
configs:
- config_name: default
data_files:
- split: train
path: data.jsonl
---

# Experiment F2: Cross-Domain Primacy (Brand vs Political Attitudes)

## Summary

2,400 LLM API calls testing whether serial position primacy generalizes from brand perception to political attitude measurement. Uses two parallel 8-dimension frameworks: Spectral Brand Theory (SBT) for brands and Moral Foundations Theory (MFT) for policy evaluation. Part of the R15 study on dimensional collapse in AI-mediated brand perception (Zharnikov, 2026v).

- **Design**: 2 domains (brand, political) x 2 formats (JSON, Likert) x 8 orderings x 5 stimuli x 5 models x 3 reps
- **Models**: Claude Haiku 4.5, GPT-4o-mini, Gemini 2.5 Flash, DeepSeek V3, Grok 4.1 Fast
- **Brand stimuli**: Hermes, IKEA, Patagonia, Erewhon, Tesla
- **Policy stimuli**: Universal basic income, Immigration reform, Carbon emissions cap, Defense budget reallocation, Social media content moderation
- **Total cost**: $0.86

## Key Findings

1. **Brand/JSON primacy confirmed**: d = +.193, p < .0001 (n = 598).
2. **Political primacy absent**: d = -.020, p = .619 (n = 597).
3. **Domain interaction significant**: t = 2.772, p = .006.
4. **Likert eliminates primacy in both domains**: Brand Likert d = +.012, Political Likert d = +.006.
5. **Primacy is domain-specific**: Not a universal LLM positional bias but a brand-perception artifact.

## Political Dimensions (Moral Foundations Theory)

1. Care/Harm — Compassion, protection from harm, empathy
2. Fairness/Cheating — Justice, equal treatment, reciprocity
3. Loyalty/Betrayal — Group loyalty, patriotism, solidarity
4. Authority/Subversion — Respect for authority, social order
5. Sanctity/Degradation — Purity, sacredness, moral disgust
6. Liberty/Oppression — Individual freedom, autonomy
7. Efficiency/Waste — Resource optimization, practical outcomes
8. Tradition/Progress — Preservation of customs vs innovation

## Dataset Structure

Each line in `data.jsonl` is one API call with these fields:

| Field | Description |
|-------|-------------|
| `timestamp` | ISO 8601 UTC timestamp |
| `experiment` | `f2_cross_domain_primacy` |
| `model` | Model key |
| `model_id` | Specific model identifier |
| `domain` | `brand` or `political` |
| `stimulus` | Brand name or policy scenario key |
| `response_format` | `json` or `likert` |
| `ordering_index` | Latin-square rotation index (0-7) |
| `dimension_order` | Ordered list of 8 dimensions as presented |
| `repetition` | Repetition number (1-3) |
| `prompt_hash` | SHA-256 hash of prompt text |
| `prompt_text` | Full prompt sent to the model |
| `raw_response` | Raw model response text |
| `parsed_weights` | Parsed dimension weights dict |
| `position_weights` | Weights mapped to serial positions 1-8 |
| `elapsed_ms` | Response latency in milliseconds |
| `cost_usd` | Estimated cost per call |
| `error` | Error message if call failed |

## Citation

```bibtex
@article{zharnikov2026v,
title={Dimensional Collapse in AI-Mediated Brand Perception: Large Language Models as Metameric Observers},
author={Zharnikov, Dmitry},
year={2026},
doi={10.5281/zenodo.19422427}
}
```

## License

CC-BY-4.0