Recent methodological discussions in survey research have addressed the relationship between low response rate and the potential for
response bias. Sampling frames supplemented with data on all sample elements can be used to assess the existence and severity of response
bias and make analytic adjustments as appropriate. Hierarchical sampling frames afford the opportunity to quantify individual and contextual
characteristics that affect response likelihood. A sampling frame of randomly selected faculty members nested within departments and
institutions was constructed from publicly available data and used to assess response likelihood from each level of the hierarchy.Anonymous
surveys were mailed to each of 10 faculty members in each of 10 departments at 50 universities. Survey items asked about questionable
research practices of respondents and their colleagues, and were therefore sensitive in nature. Returned numbered postcards confirmed
survey completion so that response status but not survey responses could be linked to each sample element. The sampling frame included
characteristics of universities (public/private, AAU/Carnegie membership, region, NIH rank), departments (fi eld, size), and faculty (rank,
sex). Multilevel logistic regression predicted confi rmed complete status from faculty, department and university characteristics. Postcards
were returned from 50 universities, 476 departments and 1402 faculty. Faculty less likely to confi rm survey completion were those from
private and western universities; in anthropology, economics and some medical specialty departments; and those with unknown gender
or assistant, associate or unknown academic rank. Faculty in some allied health departments were more likely to confi rm completion.
Creating an enhanced hierarchical sampling frame from publicly available data was feasible. Supplemental information made a systematic
assessment of individual and contextual response biases possible so that predicted response likelihoods could be used to adjust the primary
analyses. A moderate response rate was observed but response bias was less severe than anticipated.