Foundations of Agnostic Statistics

A rigorous yet accessible approach to statistical inference that doesn’t rely on distributional assumptions. This book is particularly valuable for data scientists working with real-world data that rarely meets textbook assumptions.

The authors present a framework for statistical analysis that’s robust to model misspecification - a common issue in practice that’s often glossed over in traditional statistics texts. The emphasis on design-based inference and randomization-based methods provides tools that work even when we can’t specify a data-generating process.

The mathematical rigor is balanced with practical intuition, though readers should be comfortable with undergraduate-level probability and statistics. The connection to causal inference is particularly well done.

Recommended for practitioners who want a deeper understanding of when and why their methods work, beyond the “assume normality” approach of introductory texts.