The literature on survey data fabrication is fairly thin, given the serious threat it poses to data quality. Recent contributions have focused on detecting interviewer fabrication, with an emphasis on statistical detection methods as a way to efficiently target reinterviews. We believe this focus to be too narrow. The paper looks at the problem of fabrication in a different way, exploring new data that shows the problem goes beyond interviewer curbstoning. A surprising amount of apparent fabrication is easily detected through comparatively rudimentary methods such as analysis of duplicate data. We then examine the motivations behind survey data fabrication and explore the utility of fraud investigation frameworks in detecting survey data fabrication. We finish with a brief discussion of the importance of additional research in this area and suggest questions worth exploring further. This paper is a synthesis of presentations given by the authors at an event sponsored by the Washington Statistical Society.

Writing in 1945, Leo Crespi warned "The prevalence of so-called 'cheating' by interviewers in the process of obtaining public opinion and market research data has become an increasingly grave concern to responsible opinion researchers" [10]. He could have said essentially the same today, though the cause for concern is more complex than in the past.The forms of survey data fabrication we are facing today are more diverse, and in some cases more insidious than individual interviewers sitting on the proverbial curbstone and making up survey responses. Recent papers and presentations have described apparent cheating by field supervisors [7], wholesale fabrication of datasets by researchers [4], creation of final survey results without any apparent data collection [3,17], and the apparent mass duplication of all or part of survey responses most likely using a computer [9,16]. With the proliferation of types of fabrication, Crespi's charge from 70 years ago remains pertinent.