what would be absolutely necessary to ensure that a study with two between-subjects levels of a single independent variable has high internal validity?