Key points are not available for this paper at this time.
A common practice in HCI research is to conduct a survey to understand the generalizability of findings from smaller-scale qualitative research. These surveys are typically deployed to convenience samples, on low-cost platforms such as Amazon's Mechanical Turk or Survey Monkey, or to more expensive market research panels offered by a variety of premium firms. Costs can vary widely, from hundreds of dollars to tens of thousands of dollars depending on the platform used. We set out to understand the accuracy of ten different survey platforms/panels compared to ground truth data for a total of 6,007 respondents on 80 different aspects of demographic and behavioral questions. We found several panels that performed significantly better than others on certain topics, while different panels provided longer and more relevant open-ended responses. Based on this data, we highlight the benefits and pitfalls of using a variety of survey distribution options in terms of the quality, efficiency, and representative nature of the respondents and the types of responses that can be obtained.
Bentley et al. (2020) studied this question.