Generalizing Multiple Imputation to Time Series Data, with Application to Survey Research and Evaluating Electoral Systems and Redistricting Plans

点击翻译按钮获取中文摘要

9223637 King Nearly everyone who has ever done quantitative analysis in political science has come across the problem of needing data where none exist. One might have found the right survey, but it is missing a critical question. A researcher may have data on U.S states, but not all the information is known for all states. Likewise, a researcher may have data on the cities but not the rest of the United States. The investigators study one important aspect of this problem: handling missing data in the analysis of a dataset organized as a time series of cross-sections; a problem for which no adequate solution presently exist. This problem arose as an important obstacle in two distinct areas of the substantive research undertaken by the investigators, one involving data comprising a series of presidential "trial heat" polls and, the other, a large quantity of legilslative electoral data. In searching for a solution, the investigators found that the problem is much more general and has arisen at least implicitly in statistics, political methodology, and numerous areas of substantive research in political science and elsewhere. The researchers plan to generalize the existing statistical method of "multiple imputation," simulating data mathematically, to deal with data such as these. ***