Nonresponse among cohort members causes bias in survey estimates when the outcomes of respondents and nonrespondents are shown to be different. A bias analysis was conducted on the 2001 BPS:96/01 survey results to determine if any variables were significantly biased due to nonresponse. Considerable information was known from the 1996 and 1998 surveys for nonrespondents to the 2001 interviews, and nonresponse bias could be estimated using variables with this known information. Weight adjustments were applied to the BPS:96/01 sample to reduce any bias found due to unit nonresponse. After the weight adjustments, some variables were found to reflect zero bias, and for the remaining variables the bias did not differ significantly from zero. This analysis was performed on variables found on the frame where the true value is known for both respondents and nonrespondents. For other variables collected in the survey, where data is available only for respondents, it is not known whether the weight adjustments completely eliminate bias.

Item Response Bias

All the variables used in this report and defined in appendix A had item response rates above 85 percent. Therefore, a bias analysis for individual survey items was not necessary.