3.2 Issues surrounding multi-scale analyses
The effect of scale on statistical results was first
demonstrated by Gehlke and Biehl (1934, as cited in Dark &
Bram 2007; Dungan et al. 2002; Openshaw 1984) and Yule
and Kendall (1950 cited in Marceau, D. J. 1999; Marceau,
D.J. & Hay 1999). McCarthy (1956 cited in Marceau, D. J.
1999) showed that the statistical results valid at one spatial
scale may not be applicable at another scale. The problems of
scale differences can be described as the determination of
appropriate spatial scale to study a particular geographical
phenomenon, and the transferability of information between
two spatial scales (Marceau, D. J. 1999). The significant
effect of spatial aggregation of data was acknowledged by the
pioneer works of Blalock (1964) Clark and Avery (1976) and
Fotheringham (1991). The most common errors arising from
the use of multi-scale data are MAUP (Doll, C. N. H,,
Morley & Muller 2004; Marceau, D. J. 1999; Openshaw
1984) and ecological fallacy (Cao & Lam 1997; Doll, C. N.
H., Morley & Muller 2004; Robinson 1950).
The MAUP can affect the results in spatial studies using
aggregate data sources (Unwin 1996). The MAUP consists of
two components: the scale effect and the aggregation effect
(Doll, C. N. H., Morley & Muller 2004; Marceau, D. J. 1999;
Marceau, D.J. & Hay 1999). The scale effect is observed
when data from small regions are aggregated into larger
spatial units (Doll, C. N. H., Morley & Muller 2004; Wrigley
et al. 1996). Aggregation effect takes place due to the
combining of zone boundaries in a given scale of analysis
(Doll, C. N. H., Morley & Muller 2004).
The effects of scale and aggregation are usually manifested in
several ways in studies in spatial analyses depending on the
generalization of the datasets. The scale effect is
demonstrated through individualistic fallacy and ecological
fallacy, while the zoning or aggregation effect gives rise to
cross - level fallacy. Individualistic fallacy occurs when the
inferences from small or micro - levels are used to infer
results for macro regions. Ecological fallacy can be regarded
as the opposite of individualistic fallacy and is observed
when inferences about micro - regions are derived from
relationships at macro - regions (Cao & Lam 1997; Doll, C.
N. H., Morley & Muller 2004). Cross - level fallacies are
found in inferences derived for one sub - population from
another at the same spatial scale of analysis (Doll, C. N. H.,
Morley & Muller 2004).
There are many different approaches proposed in the
literature for managing issues of MAUP (Fotheringham
1989; Marceau, D. J. 1999; Marceau, D.J. & Hay 1999;
Openshaw 1984). For example, Openshaw (1984) proposed
the approach of optimal zoning system for spatial analyses.
An optimum scale was defined as *... the spatial sampling
unit corresponding to the scale and aggregation level
characteristic of the geographical entity of interest”
(Marceau, DJ. & Hay 1999, p. 6). An important
consideration of optimal scale approach was the absence of
unique optimal resolution. Another approach to manage
MAUP was the identification of basic entities. This approach
necessitated the study of an object of concern at a spatial
scale where it could be observed and measured
(Fotheringham 1989; Visvalingam 1991). The object was
aggregated in the entity based approach and therefore this
was one of the most effective ways to overcome MAUP
(Fotheringham 1989). Commonly used ones include
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XXXIX-B8, 2012
XXII ISPRS Congress, 25 August — 01 September 2012, Melbourne, Australia
abandonment of traditional statistical analyses and sensitivity
analyses.
The use of traditional statistics is limited in its application to
spatial data. Recent studies in remote sensing indicated the
use of spatial statistics such as geo-statistical tools and
autocorrelation indices in order to overcome the effect of
MAUP (Marceau, D.J. & Hay 1999).
In order to incur the least possible error from MAUP, this
study used the approach of optimal zoning system. In the
Indian census, the villages are aggregated to form taluks.
Therefore the taluks were considered to be the optimal
aggregation unit to propose the metrics for the villages. The
optimal models for the taluks (Roychowdhury et al. 2011b)
were used to predict the metrics for the villages. The results
from the predicted metrics were mapped for the districts of
Pune.
Table 1: Shortlisted census metrics to propose models for
surrogate census at the district and taluks
Number of households per | Total population density
square kilometre
Urban population density Female literates per square
kilometre
Total number of workers | Percentage of households
per square kilometre with car, jeep and van
Percentage of households | Percentage of households
with access to electricity as | with television
power source
Percentage of permanent | Per Capita District
houses Domestic Product
3.3 Application of the models to predict metrics at
villages
Number of female literates per square kilometre; percentage
of households with cars, jeeps and vans; percentage of
households with television; percentage of permanent census
houses and percentage of households using electricity as
power source were predicted and mapped at the village level.
The maps with the predicted metrics for the districts of Pune
are shown from figure 2.
4. DISCUSSION
In Pune, high values of number of female literates were
predicted for villages around the urban areas such as
Vadgaon Bk, Hadapsar, Khed, Kharadi, Kivale and Dehu in
the central part of the district and Jumner, Shirur, Baramati,
Kalamb and Bhor in other parts of the district. These areas
were predicted to have more than 200 female literates per
square kilometre. Most of the villages in the district have
approximately 20 to 80 female literates per square kilometre.
On an average two to five percent of the households in the
villages were predicted to have cars, jeeps and vans. Around
the urban centres there were five to ten percent of the
households predicted as having cars, jeeps and vans. Similar
trends were predicted for percentage of permanent census
houses. Urban areas were recorded to contain 70% to more
than 90 % of permanent houses. The villages in the district as
a whole showed to have 30 to 50 % of permanent houses.
More than 85% of the households demonstrated to have
Internatic
access to el
a whole exl
Due to the
the census,
from the pr
the best est
4.1 Comp
Census me
spatial scal
per square
figure 3. F
was obser\
household:
administrat
detailed di
of househo
in figure |
western, e
the lowest
square kil
around 70
demonstra!
household:
area of ap
1400 squa
to 500 ho
located ne
of Pune. $
moderate
showed a |]
Figure 3:
at (a) Dis
kilometre
The taluk
number o
shown in
the centr:
househol
kilometre
square ki
part whil
househol
from thes
areas infl
