Optimal Estimation when Researcher and Social Preferences are Misaligned