|
Activity Number:
|
459
|
|
Type:
|
Topic Contributed
|
|
Date/Time:
|
Wednesday, August 5, 2009 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
Scientific and Public Affairs Advisory Committee
|
| Abstract - #305132 |
|
Title:
|
Statistical Auditing Techniques and Small Errors in Four Races in California
|
|
Author(s):
|
Luke W. Miratrix*+ and Philip B. Stark
|
|
Companies:
|
University of California, Berkeley and University of California, Berkeley
|
|
Address:
|
1513 Woolsey St, Apt C, Berkeley, CA, 94703,
|
|
Keywords:
|
election auditing ; risk-limiting audit ; error tolerance ; nonparametric
|
|
Abstract:
|
We report results gained and lessons learned from audits of four contests in California in 2008. The audits were risk-limiting in that they were guaranteed to have at least a 75% chance of leading to a full manual count if a manual count would show a different outcome, but were also designed to allow certification if only small errors were observed. To our knowledge, they are to date the only risk-limiting election audits that have been performed. Two overall methods were used to accommodate different logistical issues and to minimize the expected amount of manual counting. Both methods pose the audit as a nonparametric sequential test of the hypothesis that a manual count would change the outcome. A main difference between the two is the sampling scheme. One uses simple random sampling, the other, sampling with probability proportional to a rigorous bound on the error in each precinct.
|
