The Dictionary of Human Geography

Suggested reading
Dean (1994); Elden (2001, chs. 4, 5).

general linear model (GLM) A family of
statistical procedures, used in the analysis of
two or more variables, based on the covaria-
tion among those variables – the degree to
which the pattern for one variable across a
set of observations is replicated in another.
Techniques based on this model are at the
core of muchspatial analysis, as well as
in the analytical procedures of comparable
disciplines (cf.spatial econometrics), and
operational algorithms are available in most
computer statistical software packages (cf.
software for quantitative analysis).
The core of GLM is the technique of
regression, which identifies relationships
among variables, one or more specified as
the independents (or causes, where causality
is implied in the modelling) and another as the
dependent (or effect): the associatedcorrel-
ation coefficient evaluates the regression’s
goodness-of-fit. Other commonly used tech-
niques includefactor analysisandprincipal
components analysis, which seek underlying
common patterns in the correlations among
groups of variables.
Data deployed in GLM techniques can be
at any one of the four different levels ofmeas-
urement – nominal, ordinal, interval and
ratio – and variables of each type can be used
in techniques incorporated within GLM, each
having particular technical issues that may
need resolution for it to be validly deployed.
(Some ratio variables have pre-defined upper
and lower values – such as percentages and
proportions – and have to be transformed in
order to meet the GLM requirements, as in
categorical data analysis,logit regres-
sion models and poisson regression
models: see alsocollinearity.) Spatial data
raise the particular problems ofspatial auto-
correlation.
Apart from regression using interval and/or
ratio data for both the independent and
dependent variables, commonly used GLM
techniques include the following:

 Analysis of variance(ANOVA), in which
the dependent variable is either interval or
ratio and the independent variables are
nominal or ordinal (although nominal vari-
ables can be incorporated within a regres-
sion framework using dummy variables
and continuous – interval and ratio –
variables can be placed in ANOVAs using
covariates).

 Binomial and multinomial regression, in

which the dependent variables are nominal

or ordinal (in binomial regression, there

are only two possible outcomes; in multi-

nomial there are more than two) and the

independents are also nominal/ordinal –

although continuous variables can also be

incorporated as independent variables.

 multi-level modelling, a form of regres-

sion (with either continuous or nominal/

ordinal variables), in which the observa-

tions are clustered into nominal categories.

 Factor and principal components analysis.

 Discriminant analysis, in which the depen-

dent variable is either nominal or ordinal

and the independent variables are factors/

components comprising groups of related

continuous variables (with the groupings

derived empirically rather than predeter-

mined).

Many techniques in spatial analysis (such as

geographical weighted regression: see also

local statistics) are based on the GLM. rj

Suggested reading

O’Brien (1992).

general systems theory (GST) An attempted

development of universal statements about the

common properties of superficially different

systems, initiated by Ludwig von Bertalanffy

(1901–72: see von Bertalanffy, 1968). GST

was introduced to geographers during itsquan-

titative revolutionas a framework that could

unite various strands of work, and used by

some to promote links betweenhuman geog-

raphy and physical geography (Haggett,

1965; Coffey, 1981): Chisholm (1967) dis-

missed it as an ‘irrelevant distraction’. The

search for isomorphisms across systems

focused on three ‘principles’:

 allometry– the growth rate of a subsystem

is proportional to that of the system as a

whole;

 hierarchical structuring(as incentral place

theory); and

 entropy.

Few substantial achievements resulted, how-

ever, apart from the early work onmacro-

geography and more recent analysis of

fractals. rj

genetic algorithm A search technique

deployed in computers to identify solutions

to largeoptimizationand other problems.

Gregory / The Dictionary of Human Geography 9781405132879_4_G Final Proof page 271 2.4.2009 6:30pm

GENETIC ALGORITHM