This paper explores the use of post-stratification to compensate for the unit nonresponse in the 1987 U.S. Characteristics of Business Owners (CBO) Survey. As is often the case, post-stratification leads to estimators with desirable properties under both a quasi-random (response) and a parametric (superpopulation) model. Some care is necessary in setting up these simple models because the CBO is a survey both of firm characteristics and of the characteristics of firm owners. Variance estimation methods are proposed that measure parametric model variance and quasi-design mean squared error simultaneously even when finite population correction cannot be ignored.
