|
Activity Number:
|
255
|
|
Type:
|
Topic Contributed
|
|
Date/Time:
|
Tuesday, August 4, 2009 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Section on Government Statistics
|
| Abstract - #303249 |
|
Title:
|
Use of Record Linkage Techniques to Improve Data Capture Rates
|
|
Author(s):
|
Mario Hendricks*+ and David Bonatz and Andrew Graves
|
|
Companies:
|
Lockheed Martin and Lockheed Martin and Lockheed Martin
|
|
Address:
|
7833 Walker Drive, Suite 200, Greenbelt, MD, 20770,
|
|
Keywords:
|
record linking ; intelligent character recognition ; census
|
|
Abstract:
|
Making significant reductions in the labor costs for capturing survey data from paper instruments requires high rates of Intelligent Character Recognition (ICR). ICR recognition rates are limited by the ability to assert with confidence that a response is accurately recognized. It has been observed that as many as 50% of the low confidence recognitions are actually correct. This paper explores the use of record linking techniques with commercially available data sources to improve recognition confidence levels. Tests show that reductions of up to 40% of low confidence response for certain response types can be achieved. The search is illustrated with test data from the U.S. 2010 Census.
|
