Improving likelihood-based inference in control rate regression