Interesting Association Rules of Household Indicators of Poverty

1
Interesting Association Rules of Household
Indicators of Poverty

• By
• Nkumbuludzi Ndwapi

2
Contents

• Introduction
• Literature Review
• Methodology
• Analysis
• Conclusions Recommendations

3
Introduction

• One of the major problems troubling the continent
  of Africa is poverty.
• Poverty is the state of which one lacks the means
  to satisfy their basic needs.
• Defining poverty often leads to the question of
  how to determine when a household is to be
  considered poor. 

4
Introduction contd

• There are various methods in which poverty can
  be measured
• income poverty
• human poverty
• capabilities deprivation.
• Coleman and Cressay (1990) explained that
  measurement of income poverty are
• Absolute
• Relative

5
Introduction contd

• The above definitions define the multi complexity
  of poverty.
• Mining Association Rules can be used to unravel
  how certain aspects of human poverty are
  associated or related.

6
Statement of Problem

• Poverty alleviation has been an issue of major
  concern to the Government of Botswana.
• According to Buthali (1997) the government took
  the decision to focus on the productive mining
  sector.
• The redistribution of this revenue is based on
  understanding a poor household.

7
Statement of Problem contd

• To characterise a household as poor should not
  only be based on economic characteristics.
• According to Buthali (1997) the analysis of
  poverty in Botswana has been solely based on
  poverty baskets and poverty datum lines over the
  years.
• The Human Development programme (1997) reported
  that figures based on poverty lines are usually
  converted to dollars and this distorts the real
  levels of inflation.
• Purchasing Power Parity exchange rates that are
  used to turn the 1/day poverty line into
  national currencies are inappropriate.

8
Objectives

• To develop association measures that could be
  used to analyse the multi-factored nature of
  poverty
• To determine the most common (frequent) types of
  housing and living conditions based on the HIES
  2002/3 data set.
• To investigate interesting association measures
  that exist between different housing and living
  conditions which can not be determined using
  traditional statistics techniques
• To classify households using interesting rules as
  to whether they are poor or not.

9
Literature Review

• Income is limited as an indicator of poverty
  because it does not capture public goods, non
  market goods, rationing, and the problem of
  distorted or imperfect markets (Alkire and
  Leander, 2005)
• So income as the sole indictor of well being is
  inappropriate and should be supplemented by other
  attributes or variables for example housing,
  literacy, life expectancy, provision of public
  goods and so on.
• Alkire and Leander (2006) explain that
  multidimensionality was also advocated by the
  basic needs approach as Sens (1997) capability
  approach argued that wellbeing is
  multidimensional.

10
Literature Review contd

• Individuals and households that are able to meet
  their basic food needs but are unable to provide
  adequately for basic none food needs would still
  be classified as poor based on this criteria
  (Obuseng and Powder 2003).
• A research by Ngwame et al. (2002) revealed that,
  as with deprivation in terms of dwelling
  structure, lack of access to safe water in South
  Africa is common among the poor.
• Ngwame et al. (2002) it is explain that there is
  often a relationship between poor housing and
  poor sanitation facilities

11
Methodology

• The data
• Obtained from the HIES 2003
• Original variable s and their categories were as
  in table 1.
• General Categorization
• Categorization of Indicators of poverty
• Indicators of poverty as in table 2.

12
Mining Association Rules

• Mining Association Rules is a branch of data
  mining that seeks to understand the shopping
  behavior of supermarket customers
• which items tend to be bought together?
• Which items are bought as substitutes?
• Agrawal, Imiellinski and Swami (1993) are widely
  credited with introducing the method of
  Association Rules at the 1993 International
  Conference on Management of Data, held in
  Washington DC, USA.
• In the intervening 15 years, mining association
  rules has become one of the best studied
  problems of data mining.
• As with other data mining techniques, association
  rules (AR) have mainly been used in market
  research studies.
• MAR aims to answer questions such as how often
  does a shopping basket that contains meat also
  contain wine?
• MAR is also referred to as unsupervised learning.

13
Mining Association Rules contd

• Searching the database for interesting patterns
  is an unsupervised learning process.
• Suppose that X and Y represent 2 different sets
  of potential items in a shopping basket, e.g. X
  Meat, Maize meal and Y spinach
• Then an association rule between X and Y is a
  rule, r X ?Y.
• The rule r, is interpreted in this case as
  meaning that
• customers who buy Meat, Maize meal are likely to
  buy spinach with a certain probability.
• With unsupervised learning, one must then define
  conditions that make such a rule to be of
  interest.

14
Interesting Measures

• The rule r X?Y makes no prediction of Y in the
  entire database.
• In the original paper, Agrawal et al. (1993)
  introduced the support-confidence framework, and
  the apriori algorithm for mining the rules.
• Hahsler (2005) gives a comprehensive summary of
  commonly used and recent additions to measures of
  interestingness of mined association rules.
• Hahsler et al (2007) provide additional
  probability based measures of interestingness as
  well as an R-library, arules, for mining
  association rules, that utilizes two of the most
  popular algorithms - apriori and eclat

15
Interesting Measures contd

• The support of a rule, r X ?Y is the likelihood
  of finding a transaction containing all items in
  X and Y in the database. Estimated by
• proportion of transactions in which X Y are
  both present.
• The confidence of a rule r X ?Y is the
  likelihood of finding Y among all transactions
  that contain X.
• proportion of transactions containing X and Y
  among transactions containing X.
• This is a conditional probability Given that a
  transaction contains X, what is the likelihood
  that it also contains Y?
• Lift (initially called interest) of r X ?Y
  compares the likelihood of finding both X and Y
  in a transaction to the probability of their
  joint occurrence if they occurred independently
  i.e. if costumers bought the two itemsets
  independently.

16
Interesting Measures contd

• "chiSquare" (see Liu et al. 1999). The chi-square
  statistic to test for independence between the
  LHS (X) and RHS (Y) of the rule. The critical
  value of the chi-square distribution with 1
  degree of freedom (2x2 contingency table) at
  alpha0.05 is 3.84 higher chi-square values
  indicate that the LHS (X) and the RHS (Y) are not
  independent.
• "oddsRatio" (see Tan et al. 2004). The odds of
  finding X in transactions which contain Y divided
  by the odds of finding X in transactions which do
  not contain Y. Range 0...1... Inf ( 1 indicates
  that Y is not associated to X).

17
Interesting Measures contd

• Consider the following transactional matrix
• The various measures of interestingness can be
  computed as follows equations.

18
Analysis

• Descriptive statistics
• Mining Association Rules
• General Rules
• Rules of Poverty
• Cluster Analysis

19
Conclusions

• Considering the rules presented above mining
  association rules is technique that could be used
  to analyse household poverty and how certain
  household characteristics relate.
• For mining association rules to bring out the
  best rules surveys that are deliberately intended
  to capture the poor must be explored.
• Rules resulting from such surveys will be the
  best at defining and explain household poverty.

20
Further research and Limitations

• For further research ordinal regression model is
  proposed based on the clusters presented in this
  paper and the possibility of clustering the rules
  them selves is worth exploring.
• Considering the results of the paper the
  objectives of the research have been satisfied.
• The limitations of this paper is that it does
  not take into account different localities of
  household, indicators of poverty in the urban and
  rural areas are expected to be different.

21

• Thank You!