The purpose of this research report is to utilise appropriate statistical (both non-spatial
and spatial) techniques to classify areas in the country into urban and rural. These
areas, as derived by means of each statistical method, are profiled and common
characteristics amongst them are summarised for classification and definition of urban
and rural areas. Population data for these areas were aggregated to determine the
overall urbanisation for the country.
The methodology utilised was that of supervised classification. Two sample data sets
of areas that are known with certainty to be urban or rural were derived and used
consistently throughout the study. The importance of utilising areas of known urban
and rural status was firstly to identify essential patterns or predominant characteristics
from areas that are known, and thereafter to apply similar characteristics to areas that
are not known or are ambiguous, in order to classify them as either urban or rural.
Sample 1 comprises all areas in the country with formal and informal urban
settlements, as well as formal rural areas, i.e. farms. Sample 2 is similar to sample 1,
but in addition it includes areas falling under the jurisdiction of traditional authorities,
known as tribal areas, which were classed as known rural. Non-spatial techniques,
namely linear logistic regression, classification trees and discriminant analysis, as well
as spatial techniques, namely straight-majority-rule and iterated conditional modes
(ICM), were researched, applied and analysed for both samples, for each province and
for South Africa as a whole, using the 2001 South African population census data.
Comparisons were made with the 1996 census information.
All three non-spatial statistical methods gave insight into those census variables and
their combinations that best describe the subject under research, i.e. urban and rural.
All three methods identified significant variables that clearly separate urban and rural
areas. The results of all three non-spatial statistical methods showed similarities within
each sample, but differences were noted between the two samples. All three nonspatial
statistical methods applied to sample 1 classified the majority of the tribal EAs
(Enumeration Areas) as urban, whilst the results from sample 2 are very similar to
those obtained from both censuses, since both censuses and sample 2 predefine tribal
settlements as rural.
| Identifer | oai:union.ndltd.org:netd.ac.za/oai:union.ndltd.org:wits/oai:wiredspace.wits.ac.za:10539/3897 |
| Date | 10 October 2007 |
| Creators | Laldaparsad, Sharthi |
| Source Sets | South African National ETD Portal |
| Language | English |
| Detected Language | English |
| Type | Thesis |
| Format | 7272269 bytes, application/pdf, application/pdf |
Page generated in 0.0045 seconds