Case Studies Index

2. Concerns

Against the background of widespread use of visualization in the natural sciences, there is perceived to be a relative lack of use of this approach in the social sciences and humanities, although outstanding examples of use can be shown. Interestingly, much of the current philosophical debate within social sciences would indicate that the approach of ViSc would be very welcome. Visualization involves close engagement with data rather than models of the data, while collection of the data necessarily involves abstraction of some kind that abstraction can be at many different levels.
Having identifed this potential use within Social Science, the JISC funded Advisory Group on Computer Graphics called a meeting at Burleigh Court Loughborough on 8/9^th May 1997 which attempted to:

• Survey current work in social science making use of graphical computing and visualisation techniques;
• To evaluate the potential of the available technology to enhance teaching and research in social science;
• To explore the pictorial; data requirements of the social sciences;
• To make recommendations to AGOCG on the infrastructure necessary to support these developments.

Arising from that meeting a series of actions were suggested. One of these was to do with education and awareness, and a clear need for some readily accessible case studies of modern graphical computing in use in social sciences was identified. It was decided to address this need by a call for proposals for two Reviews and a number of Case Studies of how visualization is used in the Social Sciencees. These have been jointly funded by AGOCG and ESRC. This volume presents the Case Studies arising from that initiative.
The Objectives of the case studies are to:

• introduce the unfamiliar aspects of vsualization from two directions, by posing the problem of making use of visualisation
• give a readily available source of technical information on the software and hardware systems used.

3. Why are social science data different?

Data in the physical sciences result usually from controlled experiments or as model outputs. Investigators has choice on frequency, sampling, mode of measurement and domain covered. A good example is an atmospheric GCM of the type used to investigate global warming (REF). The output is a regular grid of values of , say, temperature, and the visualisation problem is straightforward. A number of fundamental differences are evident in the Social Sciences:

• In most work in ViSc in the physical and natural sciences, it is important to realise that the domains that are sampled are almost always assumed to be continuous. Very commonly in social science applications, although the domain is continuous it may be that our measurements are an enumeration from a discrete region which is a subset of this space. This is typical of census data and provides the social sciences with a major analytical difficulty referred to as the modifiable areal unit problem (MAUP).
• Typically, ViSC software systems also insist on regularly distributed point sampling of these continuous domains. Such data are common in many applications in natural and physical science, for example as the outputs from simulation models or from devices of the type used in medical imaging. Their use also implies that the data models used in ViSc are very simple, at least by compared with social science information and the standards of, for example, Geographical Information Systems. Census data are, of course, nothing like this, though the use of interpolation and/or density estimation enables `surface models' to be created (Martin, 1989).
• In ViSc all independent variables have the same potential importance. Using a ViSc, the domain may well refer to an abstract space given by two other variables about which the investigator has little a priori information, so that location within the domain is not a paramount concern, although patterning within it most certainly is. A major problem in using ViSc techniques in applications where the domain is real geography is to find ways by which the spatial scientist can be provided with locational clues to say where he or she is in the real world.
• Fundamental to much social science information is its strong and important location in time and space component. This is in common with some previous work in ViSc and particuarly in animation, but much remain to be done, and the methods used to be evaluated for the social sciences.
• Much social sciences information is qualitative, being based on questionnaire survey. Here measurement is an attempt to record subjective opinions, using traditional Lickert scales (such as Agree, Neutral, Disagree), binary answers and free text. Analytical devices for these are varied, but research in ViSc has not addressed any aspect of how these should be visualised.
• Social information is usually multivariate and necessarily so. This is because variables are often surrogates for concepts that have no single unique or readily agreed operational definition. An example would be social deprivation or class.
• Just as social concepts are hard to define, so too are the entities or objects of study. This leads to results being highly conditional on aggregation used and scale of analysis. Ecological fallacy is a well known if not understood.
• For many years the mixed mode of social science data has given problems in statistical analysis. Within one analytical framework data may be in nominal, ordinal, interval and ratio. This also presents visualisation problems.