AI, Statistics, and Decision-Making

AI, Statistics, and the Future of Decision-Making

From our GPT-4 study to the evolution toward augmented, auditable analytics systems

Research Findings

What Our Study Found

Confirming SPSS Results With GPT-4 and o3-mini Models

In our published study, we tested whether LLMs could replicate SPSS outputs across a range of statistical tests.

Evolution

Beyond the Study

What's Changed Since Then

When we published, models worked in isolation. Now, augmentation strategies have transformed best practice.

RAG

Retrieval-Augmented Generation

Retrieves authoritative sources before answering
Reduces hallucinations by grounding in verified references
Would have prevented degree-of-freedom errors

CAG

Cache-Augmented Generation

Reuses validated outputs for consistency
Eliminates drift in repeated analyses
Addresses o3-mini’s coefficient inflation

RAG reduces hallucination risk, while CAG enforces repeatability.

Integration

The Power of Integration

Natural Language Meets Complex Software

Model Context Protocol (MCP)

Bridges natural language and technical precision

Users ask in English

Natural, conversational queries

Execute Real Code

R, Python, or SAS execution

Audit-Friendly

Reproducible, verified outputs

This is the future of analytics: human-friendly interaction with machine-verified results.

The next frontier isn’t just smarter models

it’s smarter humans empowered by them.

Adopt boldly. Govern wisely. Build systems that amplify human potential beyond its previous limits.

About This Research

This analysis builds on our published study examining GPT-4’s statistical capabilities compared to SPSS. As AI continues to evolve, the conversation has shifted from “can models do this?” to

“how do we build trustworthy, auditable systems that amplify human expertise?”

Blake Oxford

Systems Architect & Workflow Strategist

AI-driven decision systems, healthcare analytics, enterprise automation