Toward Sufficient Statistical Power in Algorithmic Bias Assessment: A Test for ABROCA