Presenter: Tracy Wessler June 5, 2007

1. U.S. Census Bureau Decennial Response Integration System (DRIS) Presenter: Tracy Wessler June 5, 2007 The Use of High Speed Data Processingto Capture Census Data

2. Often Overlooked Critical Success Questions • Is the Form design compatible with the data capture system design? • How does the system prevent loss of data? • Is the optical mark recognition capable of picking the intended answer correctly in complex situations? • Is it appreciated that use of Optical Character Recognition effectively requires a significant investment in tuning and testing? • Does the system have adequate quality assurance and quality control.

3. Form Design Considerations • Chief cause of automation failures • Respondent confusion sets stage for data capture errors • Respondent Friendly vs. machine friendly design. • Significantly impact data capture accuracy • Forms are considered as inputs to the data capture system, including considerations of variability contributed by respondents.

4. Forms Design Success Factorsfor Census • Established representative team of knowledge people in areas of content, layout, printing and mailing considerations, and data capture considerations. • Acquisition required vendors to demonstrate a thorough understanding of the complexities, interactions, and tradeoffs. • Technologically superior systems are capable of processing forms optimized for the respondent vs. forms optimized for the computer.

5. Preventing Data Loss • How does the system control inventory? • Bar Code tracking • Detecting Double Feeds during scanning • Forms Check Out Process Established • Data acknowledgement (receipt for delivered data)

6. OMR Considerations • Defined as capture of data from multiple choice boxes • Placed emphasis on what we call Optical Answer Recognition. Census wants to know the answer the respondent was trying to communicate, and not just which boxes contained some sort of a mark. • Optical Answer Recognition is a specialized form of OMR – Many OMR products do not do Optical Answer Recognition

7. Optical Character Recognition Considerations • Beware of exaggerated vendor claims for data accuracy • Census believes to obtain both a high percentage of work captured by OCR (80% or higher accept rate) and a high accuracy rate (99% or higher measured at the field level) requires a significant upfront development investment

8. OCR Considerations Continued • For Census, return on investment is significant for most forms due to extremely large census volumes. • For example, prior to Census 2000 and the use of OCR, it was not possible to capture full names on all forms. Name capture was critical to resolve a large number of duplicates experienced.

9. Quality Data • Census experience is that many commercial applications lack adequate quality assurance and quality control. • Significant focus by Census - Quality is engineered into all process • Testing is not enough to ensure quality data – the most rigorous testing cannot completely simulate the live Census environment. • Data quality is measured during live processing so that errors can be detected and corrected