Analyzing potential mode effects in the National Crime Victimization Survey
The National Crime Victimization Survey (NCVS) is a nationally-representative survey of the non-institutionalized U.S. population aged 12 and older and utilizes a 7-wave rotating panel design. Prior to 2006, first wave interviews were not used in crime victimization estimation - these interviews were used only to provide a temporal landmark for survey participants to control telescoping, ensuring that the second- through seventh-wave interviews had bounded reference periods. Beginning in 2006, the Bureau of Justice Statistics (BJS) began using first-wave interview data to produce estimates as a way to cut costs without sacrificing estimate precision. This change is significant in many ways, not the least of which is its potential for the introduction of multiple sources of bias. This paper addresses how the authors approached analysis of mode effect bias - one of the potential sources of error in the NCVS resulting from this change. Since first-wave interviews are primarily conducted in-person and second- through seventh-wave interviews are primarily conducted over the phone, the potential for mode effect bias is high. Quantifying this potential source of error is important for understanding how incorporation of first-wave data affects victimization estimates in the NCVS. After accounting for respondents' level of exposure to the NCVS, the difference in victimization rates between in-person and telephone groups is shown to be non-significant, suggesting the apparent mode effect is actually a symptom of respondent fatigue.