The 2010 Census Address Canvassing operation was implemented to improve the completeness and accuracy of the initial census address list. The U.S. Census Bureau then used the updated addresses to mail or deliver census questionnaires. To conduct Address Canvassing, t he universe was divided into collection blocks, which were then grouped into Assignment Areas for operational convenience.

The original listers used hand-held computers to conduct the verification of the address list and census maps. The hand-held computers offered several advantages over paper listing operations, one being the ability to collect Global Positioning System coordinates for living quarters.

The Address Canvassing Dependent Quality Control check was designed to detect significant errors and provide an evaluation of each work assignment. Every Assignment Area was checked in the Dependent Quality Control . The quality control lister’s task during the Dependent Quality Control check was to compare what was found on the ground to what was on the address list.

For each Assignment Area, a quality control lister was assigned to check a sample of addresses, beginning with an address randomly selected by computer software. The quality control lister would begin with the randomly selected address and work from the ground to the address list, ensuring what was found on the ground was reflected on the address list. The sample size was based on the number of addresses in the Assignment Area, and the software notified the quality control lister when enough addresses had been checked. It is important to understand that the quality control sample was designed to be representative of the entire Assignment Area. If the quality control sample contained errors, then it was assumed that there was about the same rate of errors in the remainder of the Assignment Area. Consequently, if the quality control sample failed, then the entire Assignment Area failed and was recanvassed.

These Dependent Quality Control check procedures are used to identify, deter, and correct errors and data falsification; however the quality control lister can only identify poor performance on the part of the original lister if the sampled group of structures coincides with an area where the original lister made a faulty update to the address list. In past listing operations, the quality control focused on the validation of listing data to identify procedural errors and/or data falsification, however Global Positioning System data can be used as a tool to detect and deter procedural violations, which may or may not lead to poor data.

Global Positioning System technology was first incorporated into the Address Canvassing operation with the introduction of hand-held computers during the 2006 Census Test. As listers manually recorded map spots for structures, the Global Positioning System technology simultaneously collected longitude and latitude coordinates for nearly all structures visited by field listers. It should be noted that the Global Positioning System coordinates were only available where/when the Global Positioning System signal could be received. In addition, Address Canvassing procedures specified that listers were to only record map spots for one unit in multi-unit structures due to limitations in the software capabilities. Software on the hand-held computer stored the Global Positioning System coordinates of the physical location as well as the manual map spot identified by the lister. Thus, the two sets of coordinates provided a simple, but previously unexplored, method to check the manual map spot coordinates recorded by listers in the field.

A review of data using hand-held computers during both the 2006 Census Test and 2008 Census Dress Rehearsal has shown instances where the Global Positioning System coordinates did not correspond well with the coordinates of the manual map spots recorded by listers. In many cases, a series of these discrepancies were clustered together in a small area, which suggests that listers were recording manual map spots for structures without actually canvassing the entire area and visiting every structure. Current quality control methods do not compare the Global Positioning System and manual map spot coordinates and have a limited probability of identifying these types of procedural violations. Both the distance between a manual map spot and Global Positioning System map spot (referred to as a “strand length”) and clustering of Global Positioning System map spots in a relatively small area (referred to as a “curbstoning cluster”) could potentially be used to identify procedural violations that could negatively impact the quality of Address Canvassing data. If clustered discrepancies between the two types of map spots were the result of listers who recorded map spots without canvassing an entire area, then that would represent a violation of canvassing procedures and/or data falsification. Thus, comparison of Global Positioning System and manual coordinates could represent an important new technique for the quality control of Address Canvassing field work.