Hypothesis tests are crucial to research in social science, medicine, and other fields. I will discuss a family of highly practical hypothesis tests derived from rank-based test statistics.

In traditional statistics, these tests are used to avoid distributional assumptions (e.g., assuming a normal distribution), but we find them to be highly favorable to differential privacy. We find that in all cases our new nonparametric tests achieve large improvements in statistical power over prior results. In describing this research, I will focus on larger principles that can be applied to a wide variety of research on practical private statistics.