Respondents’ Understanding of Confidentiality Language

privacy, data sharing. The aim of this paper1 is to examine the way in which respondents understand and reason about confidentiality statements. These statements may have increasing importance in the current survey context. They are often a major way in which surveys attempt to communicate about confidentiality policy, with the intent of improving, or at least stabilizing, response rates. Four factors are identified here in respondent processing: 1. the effects of misleading associations; 2. the stimulation of doubt; 3. the effects of prior knowledge and experience; and 4.the search for evidence to support belief in confidentiality. Background The effect of privacy beliefs and attitudes has been studied in connection with the decennial census. Evidence suggests that privacy and confidentiality concerns have an effect on respondent behaviors. Thus, in 1990 and 2000, trust in the Census Bureau’s assurance of confidentiality was a predictor census mail response in 1990 (Fay, Bates, and Moore, 1991; Singer et al. 2001). Martin (2001) found that respondents who had concerns about privacy were less likely to fill out a Census 2000 form completely and mail it back. Similarly, in Census 2000, (S inger et al. 200 1), found that four factors were reliably predicted nonresponse: high privacy concerns, negative views on the Census Bureau’s confidentiality practices, disapproval of data sharing, and a lack of willingness to provide SSN. These findings suggest that item nonresponse may be connected with privacy and confidentiality concerns, as well as unit nonresponse. The effects of presenting confidentiality statements to respondents have also been studied experimentally. Singer (1978) has found that assurances of absolute confidentiality significantly decreased item nonresponse rates and resulted in better quality data for sensitive questions. However, responses to such messages are complex. Research suggests that confidentiality assurances may be counterproductive in some cases, appearing to raise rather than lower suspicion ( Singer, Hippler, and Schwarz,1992.) The sensitivity of particular questions which follow the statement may also affect the effectiveness of such statements (Fre