NSF Award
1259985
Much research in the social sciences and development of government policy relies on survey data, and the demand for survey data continues to grow. The need for more data has led to longer surveys, increasing the burden for survey respondents in terms of time and effort. Empirical evidence shows a positive correlation between survey length and survey nonresponse, which threatens the representativeness of the survey estimates. There also is evidence that measurement (reporting) error increases as respondents are asked to answer more questions in the survey. Collecting fewer variables may not satisfy a given study's objectives, however. This research project experimentally evaluates the ability to collect all desired data through a split questionnaire design in which respondents are asked only a subset of the questions. The project will use a multiple imputation method to complete the data in the sections that are not asked of particular respondents. The investigators' will extend current imputation methods to include semi-parametric and parametric models. The main hypothesis is that the split questionnaire design approach will yield estimates with less bias and even less total error compared to deploying the full questionnaire.<br/><br/>This project evaluates a method that essentially transfers part of the time and effort to complete the survey from the individual to the researcher. It also evaluates the ability to collect higher quality data as a result of this reduction in respondent burden. Finally, the study aims to extend the employed statistical methods to better preserve the properties of the data. The results will help to provide an alternative methodology for a wide array of surveys, improve split questionnaire design methodology itself, and provide information regarding the circumstances under which implementing such designs can be beneficial.
