This paper summarizes past, current, and future work to control and measure nonsampling errors in the Survey of Industrial Research and Development, an annual survey of the U.S. Bureau of the Census. Errors are grouped into five categories: specification, coverage, response, nonresponse, and processing. Census and the survey's sponsor, the National Science Foundation, have recognized many of the sources of nonsampling errors in this survey, and have taken some appropriate steps to control and measure these errors. They are continuing this process by conducting an improvement project and other evaluation studies. The paper recommends a number of ways to further study, control, and measure nonsampling errors in the survey.
