Evaluating the methodological admissibility of generative AI tools for linear regression in graduate-level research

Abstract

With the growing use of generative artificial intelligence (AI) in academia, a key methodological question concerns the statistical correctness of AI-assisted quantitative analysis. This study empirically evaluates the use of generative AI tools for linear regression in graduate-level research. The authors used a methodological approach in which estimates from four AI systems (ChatGPT 4.0, DeepSeek v3.2, Gemini 3 Pro, and Grok 4.1) were compared with estimates obtained using Microsoft Excel (Windows 10). The analysis was performed on five time series using a fixed prompt structure. Comparability was assessed using thresholds for regression coefficients, the coefficient of determination (R²), and predicted results for 2030. The results show that under controlled conditions and within the ordinary least squares (OLS) method, the AI tools generate statistical results with varying degrees of accuracy. However, deviations in coefficients and predictions highlight the need for systematic validation. The study concludes that AI tools can serve as auxiliary methodological support, provided transparency, reproducibility, and threshold-based verification are ensured in graduate research practice.