Automating and computerizing data collection using computer-aided personal interviews (CAPI), computer-aided telephone interviews (CATI), and computerized self-administered questionnaires (CSAQ) have greatly facilitated survey data collection. However, computerization has also introduced new problems dealing with the use of the computer, the constraints of program logic, and the usability of the interface. In order to ensure reliable and valid data, the human-computer interface must be designed to match the flow of the interviewer process and the interviewee’s recall of information. New methods of interface design and powerful interface tools are available to aid in the generation of new interfaces for survey data collection.

Enabling users to benefit from government statistical databases requires much more than simply making them accessible. The staggering number of data tables and the complexity of terminology plus concepts makes effective use of statistical databases a challenge for novice as well as expert users.

The human/computer interface has become the focal point for both software development and applied research on the cognitive processes involved in users performing tasks on computers. On the one side, interface designers are interested in principles and guidelines that will help in specifying and implementing good user interfaces for everything from database entry and survey administration to accessing information on the World Wide Web (WWW). On the other side, researchers are interested in general principles informed by cognitive psychology as to how users search for information, navigate the interface, and make decisions based on the information found.

The proposed research by the HCIL at the University of Maryland will contribute to both facets of the interface issue: practical design guidelines and basic cognitive principles.

The proposed work has three prongs pertaining to three important issues facing the development of most human/computer interfaces: (1) the use of menus and the problem of navigation through databases and interfaces; (2) the use of maps and the problem of navigation through geographic databases; and (3) the use of data visualization tools and the problems of database query, search, and retrieval.

Considerable research has been done on the formatting, organization, and labeling of menus for interfaces (Norman, 1991). Menu design and navigational issues are important in the design of computer aided questionnaires (CAPI, CATI, and CSAQ). Automatic branching, methods of cycling by person or by factor, and techniques of edits filling in and correcting data are important areas of research.

As these interfaces migrate to the World Wide Web, many of findings and principles of menu navigation generalize. However, the WWW has some differences that have yet to be studied (e.g., the variety of navigational tools and paths, the mix of graphics and text menus, and inconsistent layouts). Many of these differences may have an effect on performance and user acceptance of ways in which the Census will be using the WWW for data dissemination and for self-administered surveys.

In order to study interface issues in survey data collection, we propose to develop a generic computer administered survey. This survey will allow us to explore navigational styles and to develop and test new widgets specifying ranges and other techniques of handling edits. It will also allow us to explore the issue of extreme guiding and focusing of survey flow versus a loose structure allowing the user to see overviews of the questionnaires and to alter the survey flow. In terms of human-computer interface issues, we will be exploring the technique of overview-zoom-details and interface apparency.

Previous research has found that splay menus providing a large number of well-organized options are superior to deeper hierarchical menus having many levels of short focused menus. Most of the studies, however, have looked at large sets of similar items that are clustered by similarity in a hierarchical manner. However, when levels are of different types such as selections among databases versus selections among search tools, this may not hold true. We propose to conduct a study in which there are two distinctly different layers of factors.

Matching navigational complexity to users, tasks, and databases continues to be of interest to both designers and cognitive psychologists. We propose to run a series of studies in which the interface is either overly rich in menus, generally matches the user’s needs, or is impoverished. The question is whether users can cope with the mismatch and in which direction it is best to err.

Current WWW browsers allow many different tools that support different user styles and strategies of navigation using forward/back, top, go to, and directly to bookmarked pages. We propose to conduct studies looking at the patterns of search based on different navigational tools provided to the users. It is hypothesized that these tools will not only affect user performance in locating information, but they will also affect the search strategies and decisions made by the users.

Much of the data to be disseminated by the Census is geographically indexed and data is aggregated different geographic levels (e.g., all U.S population, state level, county level, city level, postal code level). An increasing number of interfaces allow the user to browse using maps. Users locate a region, a state, a county, a city, or a neighborhood to find the data about that area. One way to access such geographically indexed data is to start from a map of the US, select a region and zoom in. Select a county and zoom in; select a city and zoom in; and so on. This approach sets a context and allows the searcher to see overviews first, then details on demand. However, zooming can also be difficult to use, time consuming, and depending on the interface can cause disorientation.

We plan to study what data types lend themselves to geographic searches via map interfaces. We plan to generate a list of pros and cons of map interfaces and a list of best practices for implementing data/map interfaces.

First, we plan to survey and evaluate a number of current data/map interfaces used by the Census and others. We will develop user mental models of maps that will help to guide interface design.

We propose to run a series of studies to determine what causes the negative consequences of zooming and to find ways to avoid them. For example, it is suggested in the literature that when switching from one screen to another, that "graceful" transitions are used such as fading or showing intermediate images.

A number of tools are used to navigate maps. These include selecting focal points, zooming, shifting up/down/left/right, etc. Despite their clarity of function to the designer, users can be easily surprised or confused by results. We will investigate the reasons why these functions work or don’t work from the user’s point of view. We will investigate the ease of use of various controls for map navigation.

As users navigate maps, the hope is that they not only find the information that they are looking for, but that also gain a deeper appreciation for the geographic context of the information. We will look that how much information is gained and how this additional information helps in understanding and decision making. Finally, we will investigate constraints on what can or should be shown when drilling down from high level aggregations to specific locations. What is the appropriate granularity of the level?

Locating a desired database or data table requires (1) that the users know what they are looking for, or (2) that the user interface provides sufficient clues to enable them to chose among the possibilities or (3) determine that what they are seeking is not available. Information may be organized geographically (by state, city, etc.), temporally (1998, 1997,

etc.), thematically (demographics, industry, housing, climate, etc.), or by other categorizations. Domain experts must provide the categorizations, but the user interface should support broad shallow hierarchies to enable rapid exploration. Hierarchies can be nicely represented by nested indented trees, by node-link diagrams, or more sophisticated information visualization methods such as treemaps or cone trees.

Once a table has been located, the terminology can be explained by explanatory notes or clicks that reveal context sensitive presentations. The tabular data can be presented in textual forms or by information visualizations. Textual forms enable users to import the data conveniently and rapidly into their own processing tools such as spreadsheets, statistical packages, or databases. Textual forms also facilitate use of assistive devices for visually impaired users and thereby promote universal service. Information visualizations (or audiolizations) help users to reveal patterns, show outliers, find clusters, or identify gaps. They are powerful tools to support discovery, decision making, and explanations.

Good user interface design and effective information visualization methods enable presentation of large amounts of information while allowing users to maintain their understanding. We propose to apply advanced user interface design principles to support universal usability and emerging information visualization methods in our prototypes. Alternative methods will be tested on typical users who will carry out benchmark tasks.

Studies will be conducted in which alternative visualizations using dynamic query and conventional query methods will be compared in terms of user performance and user satisfaction.

Knowledge extraction is generally facilitated using visualizations of the data rather than presenting either raw data or tables of data. Furthermore, it is hypothesized that search using dynamic queries will further help the user to understand both the structure of the data and inter-relationships between variables in the data. We will conduct experiments testing the knowledge of users about the database following different query methods.

Finally, we will look at controls for visualizations and the uses of metadata. We may also need to look at constraints on visualizations due to data definitions, assumptions, and statistical considerations. The question is how to inform the user of the assumptions and the caveats inherent in the data.

In order to maintain a close working relationship with the Census and an awareness of the types of problems and issues that need to be addressed, we propose the following mechanisms and deliverables:

Personnel in the HCIL will perform around 3 - 4 expert reviews of interfaces designated by the Census. These reviews will cover the functionality and the ease of use of the interface. Short reports on the evaluations will be given to the Census that outline the interface’s strengths and weaknesses and give recommendations for changes.

The HCIL will provide a number of training sessions either onsite or at the University of

Maryland for Census personnel. The topics of the training sessions may (a) usability testing methods and procedures, (b) techniques for interface evaluation, and (c) principles of good interface design.

The HCIL will assist the Census in the development of guidelines documents relevant to the interfaces under current development by the Census. These interfaces may pertain to the CAPI, the CATI, the CSAQ, and WWW pages. The guidelines documents will be primarily generated by the staff at the Census. The HCIL will help to facilitate this development by providing relevant literature, research, and reviews of the guidelines in progress.

The U.S. Bureau of the Census is charged with the task on not only collecting information about the population but also making that information accessible to the government and to the public. The human/computer interface and in particular the World Wide Web is the vehicle by which information will be disseminated. The research proposed here will help us to understand the principles behind information search, navigation, and retrieval and aid in the development of powerful yet easy to use interfaces.