On Stopping Times of Power-one Sequential Tests: Tight Lower and Upper Bounds
By: Shubhada Agrawal, Aaditya Ramdas
Potential Business Impact:
Finds best time to stop experiments faster.
We prove two lower bounds for stopping times of sequential tests between general composite nulls and alternatives. The first lower bound is for the setting where the type-1 error level $\alpha$ approaches zero, and equals $\log(1/\alpha)$ divided by a certain infimum KL divergence, termed $\operatorname{KL_{inf}}$. The second lower bound applies to the setting where $\alpha$ is fixed and $\operatorname{KL_{inf}}$ approaches 0 (meaning that the null and alternative sets are not separated) and equals $c \operatorname{KL_{inf}}^{-1} \log \log \operatorname{KL_{inf}}^{-1}$ for a universal constant $c > 0$. We also provide a sufficient condition for matching the upper bounds and show that this condition is met in several special cases. Given past work, these upper and lower bounds are unsurprising in their form; our main contribution is the generality in which they hold, for example, not requiring reference measures or compactness of the classes.
Similar Papers
Computable Bounds for Strong Approximations with Applications
Statistics Theory
Makes math rules work even with unknown numbers.
Asymptotic optimality theory of confidence intervals of the mean
Statistics Theory
Find true average more accurately with less data.
Ranking and Invariants for Lower-Bound Inference in Quantitative Verification of Probabilistic Programs
Logic in Computer Science
Finds how long computer programs will run.