Geoscience Reference
In-Depth Information
This categorization reflects historical and disciplinary points of view, but in
practice, there is no compartmentalization between these families of methods. The
spatial analysis approaches combine, as a matter of fact, methods from one or an
other of these categories, and can give them different roles depending on the context
of the questioning. For example, a coefficient of spatial autocorrelation (indicator
falling within spatial statistics) can be utilized as a variable describing the intra-
urban configuration of a set of cities in a process of geographical data analysis.
Conversely, we will be able to mobilize the nearest neighbor method (spatial
statistics methods) to test the spatial dependency of a synthetic index characterizing
the social profile of schools computed from a multivariate analysis (method within
the scope of factorial analysis).
The analysis of geographical data has benefited in particular from the contributions
of descriptive analysis methods such as multivariate data analysis
1
[BEN 80, LEB 06]
and the exploratory analysis emanating from the current of “exploratory data analysis”
devised by Tukey [TUK 77]. Unlike the classical confirmatory approach in statistics,
which consists of testing a hypothesis that is made explicit
a priori
, the exploratory
approach proposes to use a set of methods and representations deriving from
descriptive statistics. This is in order to identify relationships and tendencies that will
allow us in a second step to lead to hypotheses about the state and the functioning of
the system that is of interest. These hypotheses will in turn lead to new and more
targeted explorations and, step by step, will allow new knowledge to be highlighted
about the phenomenon being studied. This approach supports the abductive approach
referred to in Chapter 2, which involves identifying
a priori
any surprising fact in
order to follow it up and explore it from different angles. In the field of geographical
data, this type of approach has been developed by synchronizing maps and graphs in
interactive environments [UNW 90, UNW 94, BAN 01]: the explorations are
conducted in a very visual manner from a selection-query through maps or graphs,
applying a specific analysis on the selection.
Today, with the increasing volumes of available data, these approaches are
integrated in the emerging field of “data mining and knowledge discovery” that
finds a geographical specification [WAC 02, MIL 09]. The new masses of data are
most often not derived from explicit collection protocols and are therefore imperfect
and not fit to be used in the often very restrictive conditions that the conventional
statistical models presuppose. Rather than testing
a priori
hypotheses, these
approaches propose sequences mobilizing varied analysis methods in order to reduce
and/or enhance the information and detect patterns of interest: automatically
1 Particularly in France where Benzecri's school has developed an original method, this
method was specifically adapted to qualitative data (correspondence analysis or factorial
analysis of correspondence), which has allowed very strong diffusion in the areas of the social
sciences (sociology, geography, psychology, etc.), facilitated by the development of free
software and training.
