Space, place and health

Abstract

Space and place represent alternative ways of addressing the geographic world, the first based on geometry, coordinates and precise measurement, and the second based on place-names and their associations. While the spatial is a useful perspective for many forms of analysis of health data, only the platial perspective is able to inform the geographic aspects of human health behaviours, because perceptions are based on places. The two perspectives are contrasted, and examples are given of the kinds of reasoning from platial information that are routinely used by humans in solving tasks intuitively. The paper ends with a plea for more technical and theoretical work on place, to implement a vision of a platial technology that would in many ways complement geographic information systems.

Keywords:

• place
• health behaviour
• geographic information science

1. Introduction

Geographers have long distinguished between concepts of space and concepts of place. While the literature on place is rich, with many nuances, in this paper I discuss the concept in the specific context of geographic information science (GIScience). Thus, some of what follows may appear narrow to scholars of human geography, but this narrowing is necessary in order to discuss the potential of a technology oriented to place rather than space.

Humans perceive the world largely in terms of named places. A pair of latitude–longitude coordinates conveys little or no meaning to the average human, but humans acquire rich associations with places. Thus, the coordinates 43.08097N, 78.95198W would have virtually no meaning to someone searching for a house in western New York state, but Love Canal, the neighbourhood of Niagara Falls in which this point is located, would immediately suggest strong associations with major environmental contamination, and would deter most home buyers. People use place-names, not coordinates, to make associations, and it is those associations that tend to drive behaviours, many of them directly related to health. Understanding of health perceptions and behaviours, therefore, must be rooted in place-names.

Geographic information systems (GISs), on the other hand, express information about the world in coordinates. Conversion between coordinates and place-names is sometimes easy, when the named feature is small and point-like, or when an administrative entity such as a state has a legally sanctioned boundary. But the places of human discourse are often only vaguely defined and context dependent. Thus, ‘the restroom’ is highly ambiguous, since there are tens of millions of features with that name in the US, and is normally interpreted as meaning the nearest or most accessible restroom when embedded in a request for directions. Humans use a variety of heuristics to reason about place-names, whereas the coordinates of GIS are comparatively objective.

In short, people perceive, reason and thus behave with respect to geography in terms not of coordinates but of named places. Places are identified by name, are often vaguely defined and are context dependent, but are nevertheless the essential key to decoding human health behaviours. Spaces, on the other hand, are defined by the points, lines, areas and volumes of GIS, with their well-defined geometries. If the user interface of GIS is difficult to learn and use it may well be because it requires a reorientation of human thinking away from the places of everyday discourse to the somewhat alien representations of spatial databases. Distances and directions are readily computable in GIS, but only imprecisely known and reasoned about in the human mind. In what follows I contrast GIS as a spatial technology with the potential for a parallel, platial technology that would be more human centric and more readily suited to understanding health behaviours.

2. The ‘Names Layer’

GIS was originally conceived as a way of obtaining measurements from maps, as in Roger Tomlinson’s vision for the Canada Geographic Information System (Tomlinson 1998). Later, the concept of measurement was extended to the fully featured spatial analysis of today’s GIS, and additional functions, such as map editing and spatial data management, became important components. But place-names were considered largely irrelevant to measurement and spatial analysis, and when the National Spatial Data Infrastructure was implemented in the early 1990s no provision was made for place-names in the seven layers of the framework (www.fgdc.gov/framework), except as attributes of the layer of administrative boundaries. It was only later that the importance of place-names was recognized. Someone searching for data, in projects such as the Alexandria Digital Library (Goodchild 2004) that provided spatial search services on the Internet, was far more likely to define the object of search in terms of named places than in terms of coordinates. Additional momentum came from wayfinding services, such as that provided by MapQuest, since users would likely define the origin and destination of a trip using place-names.

For spatial search the most available source of place-names was the gazetteer, the list of officially recognized place-names that many countries had developed in order to standardize naming for purposes of administration and the addressing of mail. The traditional gazetteer had three elements in each record: the officially recognized name, a pair of coordinates and a classification of the type of the named feature (river, city, mountain, etc.). In the US, the Board on Geographic Names had borne this responsibility for over a century, and recently had collaborated in making the gazetteer available as a searchable index on the Internet (geonames.usgs.gov). For the Alexandria Digital Library, it was easy to incorporate a gazetteer in the interface so that searches based on place-names could be converted to coordinates.

For wayfinding, origins and destinations tend to be point-like features such as houses or businesses. Conversion of an address of a house or business to coordinates had long been enabled by geocoding, in which a location is interpolated along a link in the street network based on the addresses of the end points of the link. The necessary database to support this had been developed in support of the decennial census, and made freely available. The locations of named businesses could also be obtained from directories prepared by companies such as Dunn and Bradstreet, and generally known as point-of-interest or POI databases. By the late 1990s, then, it had become possible for users of wayfinding services to specify origins and destinations in a number of forms, as business names, street addresses or place-names. Today, such features are routinely used by the general public, and have been enhanced in various ways, for example by use of auto-complete techniques. I refer to such databases in what follows as place-name databases (PNDs) to reflect this integration of gazetteers and POI databases.

Unfortunately, PNDs remain limited to point geometry, and thus work best for small features. References to large features, such as states or long rivers, are nevertheless treated as references to points, using simple conventions. A long river is by convention treated in many PNDs as a point at its mouth, and a state is treated as a point at the geometric centre of the state. A request for a route from Colorado to Wyoming, for example, will likely result in a complex route from a point in the Rocky Mountains to a point in a Wyoming field, involving mountain roads that may be closed in winter. Hopefully, the next generation of wayfinding services will result in such cases in a dialogue to obtain more detail on the user’s origin and desired destination.

Crowdsourcing is increasingly important in the creation and maintenance of PNDs (Sui, Elwood, and Goodchild 2013). Opening or closure of a restaurant, for example, or the construction of a new street, is most effectively captured by a mechanism that allows citizens to volunteer information to the provider of the PND. Providers in turn must likely adopt semi-automated methods to deal with this flow of information in timely fashion (Goodchild and Li 2012).

To replace point geometry with something more appropriate to large features presents a major challenge, especially to do so in a cost-effective fashion. Many places do not have recognized boundaries, and the task would be even more complex if it had to deal with the effects of context by providing multiple context-specific geometries. Recently good progress has been made in using unconventional sources of data to obtain geometries. Montello et al. (2003) used interviews with human subjects to define the geometry of ‘downtown Santa Barbara’. Jones et al. (2008) have shown how a Web search can be used to define the geometry of a vernacular feature such as ‘The Cotswolds’. Li and I have used the Flickr database to find the location of features such as the Eiffel Tower, by searching for photographs tagged with that name and extracting the coordinates that were logged with the photograph when it was uploaded.

Related methods can be used to obtain the associations of places. Adams and McKenzie (2013) have analysed the content of travel blogs and similar sources, looking for the words that are frequently associated with specific place-names. For example, words associated with wine (‘bottle’, ‘glass’, ‘taste’, ‘vineyard’, ‘red’) also occur in travel blogs in association with place-names in wine-growing regions. This makes it possible to map associations – by mapping, for example, the wine-growing regions of Europe.

All of these methods work to bridge the gap between space and place, by making it possible to attach coordinates to place-names, even when the geometry of those place-names may be vague or poorly defined. They typically result not in the familiar polygons or polylines of vector GIS representations, but in fields of probability that might be captured using any of the standard methods for representing fields in GIS. The value of the field at any point lies between 0 and 1 and represents the probability that that point is contained in the feature.

3. Hierarchies of place

In a platial world, there is no need to insist on what is often described in GIScience as planar enforcement, the requirement that every point in the plane be contained within exactly one area. So while planar enforcement is true of administrative maps, such as the layer of states or the layer of counties, a point can lie within any number of named places, including zero. Named features can also form hierarchies of part–whole relationships, in which smaller places are parts of larger places. Thus, for example, the Eiffel Tower, Montparnasse and the Bois de Boulogne are all parts of the larger place Paris, and the place ‘Room 731’ is part of the Royal Park Hotel, which is part of Shatin, which in turn is part of the Hong Kong Special Administrative Region.

But as with planar enforcement, the lack of geometry in a platial world means that lower level places do not have to nest or be contained within higher level places. For example, most Americans would think of Disneyland Paris as ‘in Paris’, as implied by its business name, though it is not contained within any formally defined geometry of Paris. This is especially important for places that do not have well-defined geometries, and for which it would thus be impossible to determine containment.

Hierarchies are an essential element of a platial world, since they play a key role in associations and also in reasoning. The associations of Paris, or any other major city, are related to the associations of its component parts. We expect, for example, that some of the attributes of Paris will also be evident in its component neighbourhoods, while other associations of neighbourhoods will be unique to them. Thus, the Left Bank shares some attributes with those of all of Paris, but has other attributes, such as its density of cafes and restaurants, and its bohemian history, that are unique to it. Similarly, the associations of the various neighbourhoods of Paris combine to form the associations of the city as a whole. In what follows I use ‘include’ to refer to the component parts of a place, to avoid the geometric implications of ‘contain’.

Consider a simple reasoning task, such as the question ‘Which of Places B and C is nearer to Place A?’ In a spatial world, that is, in GIS, we would know the coordinates of all three places. If the places were point-like we could easily compute the distances between them, and answer the question. We could also invoke related GIS functions such as the buffer, to identify all places that were within a given distance of A. The ability of GIS to compute distance between points is of course a key element of its capabilities as a machine for spatial analysis, and is often extended to computing distance over the surface of the sphere or ellipsoid rather than over the plane, or computing distance through a network. If the identified places were large we would have to invoke some definition of distance between extended objects, and many options have been identified in the literature.

Humans, however, have no such abilities to compute distance in their essentially platial world, unless they are equipped to solve the task as a spatial problem in GIS. Yet, humans are faced with such tasks routinely, and must have developed some level of ability to solve them intuitively. In what follows I argue that we all use heuristics of various kinds to solve such tasks approximately.

Suppose, for example, that one is visiting Paris as a tourist, and wanting to see various sights, including the Eiffel Tower, the Musée d’Orsay and Notre Dame cathedral. After visiting the Eiffel Tower (A), one wonders which of the other two places to visit next. A smartphone service could of course resolve this issue by determining which of the Musée d’Orsay (B) and Notre Dame (C) is closer, but we assume here that no such technical support is available, and that one must rely on traditional unaided intuition. As a first approach one might apply the heuristic described above, that is, to define two places as near if they are part of the same higher level place. The difference in hierarchical levels provides a way of defining ‘near’ in this platial world. A higher level place that includes only A and not B or C will not help (the higher level place and its implied definition of ‘near’ is too small), and neither will a higher level object, such as Paris, that includes all of A, B and C (the higher level place and its implied definition of ‘near’ is too large). Of most interest, therefore, in solving the task is to find a higher level place that includes A, and either B or C but not both.

Thus, a straightforward solution to the task is to observe that both A and B are included in the 7th Arrondissement, but that C falls in the 4th. Thus, one would guess that B is nearer to A. There is obvious risk of being wrong, and it is easy to find counterexamples. Let A be Seattle, B be Los Angeles and C be Vancouver, BC. On the grounds that A and B are both in the United States and C is in Vancouver one might falsely conclude that B is nearer to A, though if one had chosen the less formal region of Cascadia as the higher level place one would have concluded correctly that C is nearer to A. But human reasoning is often imprecise, and in this case the heuristic gives an answer that is at least better than no answer.

Another commonly used heuristic reasons from what is often termed ‘route knowledge’, or knowledge of the routes that one needs to follow to get from frequently used origins to frequently used destinations. If one such route between A and C passes B, then one can assert with some confidence that B is nearer to A. Again it is possible to find counterexamples, when the known route from A to C is not the shortest. In the Paris example RER Line C passes through stations at A, B and C in that order, so one would reason that B is closer.

In summary, humans have developed intuitive heuristics for solving some of the kinds of problems that GIS now solves, but only imprecisely and in a platial rather than a spatial world. Thus, the platial world is rich, in its hierarchies, associations and support for the routine tasks that humans must perform. It is imprecise, and precision in problem-solving is an obvious argument for adoption of a spatial approach. But if one wishes to understand human perceptions, and their impacts on health behaviours, one has no alternative but to adopt a platial perspective.

4. Conclusions

The main focus of this paper has been on the dichotomy of space and place, and the importance of the latter in understanding human perceptions and behaviours, especially with respect to health. While GIS has proven very useful in many respects, in analysing patterns of disease and their correlates, in tracking human exposures to the environment, and in assessing spatial variations in health services and outcomes, its emphasis on space makes it very limited as a basis for understanding behaviours and perceptions. There is a significant need at this time for a better understanding of place, and how humans live, learn and reason in a platial world that is often unrelated to the spatial world of GIS. There is a need for a theory of place that is as powerful as the theories of space that have emerged from GIScience. There is also a need for a technology that can acquire, represent, store and share the rich information about place that humans possess. To date, crowdsourcing has proven extremely useful as a volunteer-based alternative to building and maintaining the familiar layers of the spatial data infrastructure, especially as it relates to built form. But the real value of crowdsourcing will be evident when it is able to tap the rich vein of place and platial associations that humans carry around with them.

Disclosure statement

No potential conflict of interest was reported by the author.