Inference from Matched Samples

Inference from Matched Samples - Inference From Matched...

Info iconThis preview shows pages 1–2. Sign up to view the full content.

View Full Document Right Arrow Icon

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon
This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: Inference From Matched Samples in the 2008 U.S. National Elections * Douglas Rivers † Delia Bailey ‡ Abstract The performance of matched sampling is assessed using data from the 2008 U.S. Presiden- tial election. The assumptions necessary for the validity of matched sampling, including ignorability, are described. With a matching ratio of about five, the matched sample re- produces the joint demographic distribution of the target population very closely. The sampling distribution of the associated state-level vote estimates is approximately standard normal with unit variance, suggesting little or no selection bias conditional upon a full set of demographic controls. The results are compared to RDD telephone and Internet samples, none of which are clearly better and some (such as the 2008 ANES Internet panel) are substantially worse. Key Words: sampling, matching, propensity score, surveys 1. Introduction 2008 was the year that Internet polling came of age. Most presidential campaigns conducted at least some polling on the Internet. The Economist , the Associated Press, and CBS News conducted Internet surveys throughout the campaign. Several large academic projects, including the American National Election Study (ANES), the National Annenberg Election Study (NAES), the Cooperative Congressional Election Study (CCES), and the Cooperative Campaign Analysis Project (CCAP) all collected data using the Internet. Unlike telephone polling, where similar sampling and weighting methods are used by most survey organizations, there is a wide discrepancy between how In- ternet surveys are conducted. Knowledge Networks (KN) uses a traditional sam- pling methodology—random digit dial (RDD)—and provides Internet access to se- lected respondents who do not currently have it. This is comfortably within the mainstream of survey sampling practice, as the sampling frame includes households without Internet access and uses probabilistic selection. The KN approach has been shown to produce similar results to RDD telephone surveys, but it also shares most of the problems of telephone interviewing (such as low response rates and relatively high costs) as well as the usual problems of access panels (attrition and geographical limitations). Matched samples allow low cost opt-in samples to be used for both descriptive population estimates and analytic studies. The purpose of this paper is to assess the effectiveness of matching and weighting to remove selection biases in polling applications. Because of low recruitment response rates and attrition, it is very difficult to operate an access panel with anything approximating “known” probabilities of se- lection. Even very expensive RDD panels (such as that built by KN for ANES in * The views expressed here are those of the authors, not of any organizations they are currently or formerly associated with....
View Full Document

This note was uploaded on 12/26/2011 for the course ECON 245a taught by Professor Staff during the Fall '08 term at UCSB.

Page1 / 13

Inference from Matched Samples - Inference From Matched...

This preview shows document pages 1 - 2. Sign up to view the full document.

View Full Document Right Arrow Icon
Ask a homework question - tutors are online