Are Visual Explanations Useful? A Case Study in Model-in-the-Loop Prediction