We have a variety of software systems that are used in the 2010 Census. Three of them are central to the conduct of the census—a personnel and payroll system, an operation control system, and an optical scanning system to process the completed form.

Two of these systems are doing just fine – the payroll system and the optical scanning system.

The third is a shaky system that caused us serious concerns when the nonresponse follow-up stage began on May 1. The system performs the functions of assigning work to crew leaders and enumerators, checking in their completed work, and preparing it for shipment to the three optical scanning processing centers. In the first week of nonresponse follow-up, the system was not reliable; work was assigned to enumerators through a manual backup to make sure that the no harm was done to completing the field work. Questionnaires were completed by enumerators in the field at about the rate we expected, but since the control system was not supporting checking-in the work as desired, completed questionnaires backed up in the census offices, waiting to be checked in to the database before being shipped for scanning.

For example, by Monday of this week we expected to have completed about 15% of the nonresponse follow-up workload and to have checked in about 10%. We completed about the expected amount but checked in only 5% or so.

We reorganized the management of the team overseeing the software somewhat, bringing in some outside experts on databases and the operating system. Developers visited census offices to see users actually using the system. Changes were made. Changes made about May 7 or so had dramatic positive effects on the performance of the system. So, the past few days have been better, but not perfect.

Several times in my career, I have experienced first-uses of complicated survey data collection systems. The first use is rarely a pretty affair, mainly because of the difficulty of designing testing regimens reflecting all the combinations of steps that occur in real production with thousands of diverse users. The challenges with the 2010
Census operational control system brings back those memories to me.

For the 2010 Census we are by no means out of the woods with the operation control system software. There will be other problems, if my past experience is any guide. Our goal is to make gradual improvements in performance to complete the work. Failing that, we have contingency backups in place for key functions, and we’re blessed with great staff in the local census offices who are committed to working around day-to-day problems. I can pledge that all of us at headquarters are focused on making this tool as strong as possible for our colleagues in the field.

Please submit any questions pertaining to this post to ask.census.gov