Identifying and preventing fraudulent responses in online public health surveys: Lessons learned during the COVID-19 pandemic

Web-based survey data collection has become increasingly popular, and limitations on in- person data collection during the COVID-19 pandemic have fueled this growth. However, the anonymity of the online environment increases the risk of fraudulent responses provided by bots or those who complete surveys to receive incentives, a major risk to data integrity. As part of a study of COVID-19 and the return to in-person school, we implemented a web- based survey of parents in Maryland between December 2021 and July 2022. Recruitment relied, in part, on social media advertisements. Despite implementing many existing best practices, we found the survey challenged by sophisticated fraudsters. In response, we iteratively improved survey security. In this paper, we describe efforts to identify and prevent fraudulent online survey responses. Informed by this experience, we provide specific, actionable recommendations for identifying and preventing online survey fraud in future research. Some strategies can be deployed within the data collection platform such as careful crafting of survey links, Internet Protocol address logging to identify duplicate responses, and comparison of client-side and server-side time stamps to identify responses that may have been completed by respondents outside of the survey's target geography. Other strategies can be implemented during the survey design phase. These approaches include the use of a 2-stage design in which respondents must be eligible on a preliminary screener before receiving a personalized link. Other design-based strategies include within-survey and cross-survey validation questions, the addition of "speed bump" questions to thwart careless or computerized responders, and the use of optional open-ended survey questions to identify fraudsters. We describe best practices for ongoing monitoring and post-completion survey data review and verification, including algorithms to expedite some aspects of data review and quality assurance. Such strategies are increasingly critical to safeguarding survey-based public health research.