Among responsible survey organizations, understanding survey data quality is a pervasive and continuing concern. Data consumers—clients—have a right to demand the highest possible quality for given resources, and to know as much as possible about quality limitations which might affect their uses of the data. For those who conduct surveys, information about quality is essential for making optimal decisions about survey design. One key design decision faced by all household surveys is how much effort to expend to obtain data about each sample person from that person himself/herself. Stringent self-response rules can add substantially to survey costs and lengthen data collection periods, but are also generally assumed to yield higher qua1ity data.

This paper examines response error differences between self- and proxy respondents in a major and relatively new Census Bureau household survey—the Survey of Income and Program Participation (SIPP). Standard SIPP procedures call for self-response interviews with all adult members of samp1ed households if possible, but allow proxy interviews for those not present (or unable or unwilling to be interviewed) at the time of the interviewer's visit. As a result of these respondent rules, a substantial proportion of SIPP interviews are conducted by proxy. The 1arge number of proxy interviews, coupled with concerns about the quality of SIPP data, have led to questions about the contribution of proxies to SIPP data quality problems. If proxies were found to contribute disproportionately to SIPP's quality problems, the survey's designers and sponsors would have to consider imposing more stringent self-response rules.