Accounting for Defect Characteristics in Empirical Studies of Software Testing