Improving Sampling Practices on Mechanical Turk: Acquiring Naive, High Quality, and More Representative Participants

We examine how recruitment practices on MTurk affect data quality and nonnaïveté. Targeting workers with high approval ratings and HIT completion history excludes naïve workers. Less active workers are more naïve to standard measures, and provide quality data. We make recommendations for designing MTurk HITs maximizing data quality and naïveté.