Survey research methods in evaluation and case-control studies

Survey research methods are widely used in two types of analytic studies: evaluation studies that measure the effects of interventions; and population-based case-control studies that investigate the effects of various risk factors on the presence of disease. This paper provides a broad overview of some design and analysis issues related to such studies, illustrated with examples. The lack of random assignment to treatment and control groups in many evaluation studies makes controlling for confounders critically important. Confounder control can be achieved by matching in the design and by various alternative methods in the analysis. One popular analytic method of controlling for confounders is propensity scoring, which bears a close resemblance to survey weighting. The use of population-based controls has become common in case-control studies. For reasons of cost, population-based controls are often identified by telephone surveys using random digit dialling (RDD) sampling methods. However, RDD surveys are now experiencing serious problems with response rates. A recent alternative approach is to select controls from frames such as driver license lists that contain valuable demographic information for use in matching. Methods of analysis developed in the survey sampling literature are applicable, at least to some degree, in the analyses of evaluation and population-based case-control studies. In particular, the effects of complex sample designs can be taken into account using survey sampling variance estimation methods. Several survey analysis software packages are available for carrying out the computations. Copyright (c) 2007 John Wiley & Sons, Ltd.