Using Crowdsourcing Platforms in Study Designs during the COVID-19 Pandemic: Lessons from a Mechanical Turk Survey Experiment on Mask Wearing
Noelle Chesley Phd MPA1, Helen Meier2, Sarah Laurent3
1Sociology, 2University of Michigan, 3University of Wisconsin - Milwaukee

The strengths and weaknesses of crowdsourced samples for COVID-19 research and how they compare to estimates generated from research using random samples is understudied. We compared a crowdsourced MTurk survey (N=3,860) focused on mask wearing during the pandemic to results from a similar study but with a random sample conducted in the same time frame by the Pew Research Center (N=13,200) to understand the where MTurk samples replicate patterns in population-based studies and where they fall short. Our MTurk sample replicated several key demographic bivariate patterns in the Pew study, but deviated from the US population on racial/ethnic and education distributions. MTurk samples may be useful to investigate demographic patterns in health behaviors when the sample approximates the population on that characteristics. This, in combination with their speed and low cost, crowdsourced samples from MTurk are a reasonable alternative to random samples for surveys where timeliness and resources are limited.