Ulric B. and Evelyn L. Bray Seminar in Political Economy

Many of the most high-profile surveys conducted today are done via face-to-face interviewing. Needless to say, this mode of interview is expensive. And although data gathered via in-person interviews of area probability samples are described as the ``gold standard'' in terms of data quality, there have been no properly executed randomized tests for the effects of mode of interview on U.S. adult populations. In this paper, I analyze data from an experiment conducted at the MGM Grand in Las Vegas, Nevada on an adult population. I randomly assigned people to a mode of interview after they agreed to take a survey, resulting in 500 interviewer-administered surveys and 500 self-completed surveys done on a computer. Results show dramatic differences due to mode of interview, particularly on questions about attitudes and information, even when topics were not political. The data reveal more constraint and higher levels of knowledge for respondents who completed the instrument themselves relative to those who completed the survey with an interviewer. The low-constraint, non-ideological (Converse 1964) foundation upon which sixty years of public opinion work was built may have been an artifact of survey mode.