2 problems

The type-I error rate conjecture. The type-I error rate under rerandomization should not exceed that under complete randomization:

Let $\lambda \geq 0$ and $a_{\lambda} \geq 0$ be prespecified constants. Suppose $(\bar{\mathbf{x}}_T-\bar{\mathbf{x}}_C)\mid\mathbf{x}\sim\mathcal{N}(0,\mathbf{\Sigma})$, with…