Improving research and policy interactions requires a better understanding of what works in different contexts