Traditional hypothesis testing methods have been significantly reshaped in both theory and practice by ongoing developments in computing technologies. Built on classical statistical theory, traditional hypothesis testing methods like Z-tests and t-tests, are tailored to structured and small-sample data, but their assumptions limit broader applicability. Yet these methods face considerable challenges in adapting to high-dimensional, unstructured, and large-scale datasets, particularly in terms of flexibility and computational efficiency. This paper examines the historical development of computer-assisted hypothesis testing methods, offering a comparative analysis of traditional and modern techniques with respect to their theoretical foundations, practical applications, and domain-specific performance in complex data environments. By analyzing representative literature and key methodologies, it traces the shift from classical statistical methods to modern computational techniques, thus highlighting their applications in high-dimensional data analysis, causal inference, and automated model validation. The results indicate that computation-driven hypothesis testing approaches greatly enhance the ability to handle complex data structures and extend the scope of statistical inference. However, these methods involve trade-offs that emphasize the necessity of selecting approaches suited to the data characteristics and research objectives.
Siyu Liu (Tue,) studied this question.