NON-RESPONSE BIAS ANALYSES OF THE SURVEY OF WORKPLACE VIOLENCE PREVENTION

This paper examines two methods for analyzing nonresponse bias using establishment indexed data sets from the 2005 Survey of Workplace Violence Prevention (WVP), the 2003 Survey of Occupational Injuries and Illnesses (SOII), and the BLS Longitudinal Database (LDB). First, comparisons were made using 2003 SOII and LDB frame data conditional on response status to the 2005 WVP survey. The data indicated WVP responders had higher SOII incidence rates than WVP nonrespondents, response rates were higher for larger establishments, and for industries with relatively higher risk for workplace violence. Our second method used a two-stage logarithmic regression procedure to investigate the correlation of efforts to contact respondents (mailings and non-response calls) to the likelihood of actually contacting the respondent and securing participation. Additional effort past the first non-response mailing and first round of non-response calls did not appear to improve the chances of contacting a given respondent. Conditional on making confirmed contact with the respondent, repeated efforts to obtain cooperation were negatively correlated with participation rates. Based on these findings, we conclude that WVP respondents tended to be establishments that were predisposed to cooperating, most likely had familiarity with occupational hazards, and tended to be those most likely to have programs and training. Therefore, the estimates produced by the Survey of Workplace Violence Prevention of program and training utilization should be regarded as conservatively high estimates despite their low levels. 1. The Survey of Workplace Violence Prevention