Module

1
Module 2 Q2 in the Analysis of Poverty
Correlates Characteristics
Training Seminar - Q-Squared (Q2) Combining
Qualitative and Quantitative Approaches in
Poverty Analysis
International Development Research Centre (IDRC)
(Canada) World Bank ASEM Trust Fund TF053933
Project

• Paul Shaffer
• Centre for International Studies
• University of Toronto, Canada

Centre for Analysis and Forecasting, Vietnamese
Academy of Social Sciences Hanoi, May 5-6, 2005
2
Relevance, Objectives and Topics Covered

• Relevance Why care about the Analysis of Poverty
  Characteristics/Correlates?
• All subsequent poverty analysis (dynamics,
  causes) and related analysis (impact assessment)
  presupposes some conception of well-being/poverty.
  Poverty identification is about determining
  which conceptions of well-being to use and the
  methodological implications of that choice.
• Objectives
• To provide an understanding of methodological
  features of different approaches to the analysis
  of poverty correlates/characteristics
• To illustrate these differences through Q2 case
  studies
• Topics Covered
• Review of the Q2 Conceptual Framework
• Methodological Features of Analyses of Poverty
  Correlates/Characteristics
• Methodological Features of Consumption Poverty
• Methodological Features of PPA Poverty
• Case Study 1 Putting Together PPA vs.
  Consumption Poverty
• Case Study 2 Methodological Integration
  Enhancing Comparability in PPA Poverty
• Case Study 3 Methodological Integration
  Enhancing Representativeness in PPA Poverty
• Case Study 4 Methodological Integration Using
  People Perceptions in Poverty Line Determination

3
GROUP WORK 2

• 1. Divide into three groups (North, Centre,
  South)
• Q Give five characteristics of a poor individual
  in urban and rural areas in your region.
• 2. Take 5 minutes
• 3. Nominate 1 person to present your answer to
  the group

4
GROUP WORK 2 - Rationale

• The Group Work had three objectives
• 1) to illustrate that the dimensions of poverty
  used by the different groups tend to differ
  (Basket Consistency)
• 2) to illustrate that the poverty cut-off used by
  the different groups may not be the same (Cut-off
  Consistency)
• 3) to illustrate that as a result of points 1 and
  2, results may be sensitive to whoever happens to
  be conducting the ranking exercise (Reliability).

5
1. Review of the Q2 Conceptual Framework

• Identification Stage. Analysis of
  characteristics/correlates of poverty. Addresses
  question Who are the poor?
• Ways of Combining distinguishes between
• Putting Together. Results of different
  approaches conducted separately are put together
  with a view to enrich or confirm/refute each
  other.
• Methodological Integration. The outputs of one
  approach feed into the design or methods of
  another (e.g. using ethnographic results to
  inform household survey questionnaire design) or
  the integration of techniques typically used in
  one approach in another, (e.g. selecting PRA
  sites probabilistically and calculating standard
  errors for the numerical data generated).

6
2. Review of Module 1 A Few Identification
Issues
There are three separate issues in the
Identification Stage 1) Determining the
dimensions/ conception of poverty (Poverty
Basket 2) Determining the poverty line (Poverty
Cut-off) and 3) Adding up those below the
Poverty Line (Aggregation). The Consumption and
PPA Poverty Approaches do these differently.
7
2.1 Methodological Features of Analyses of
Poverty Correlates/Characteristics
The following seven criteria distinguish
different traditions of analysis of poverty
correlates/ characteristics 1. Peoples
Perceptions. Is there an attempt to include
peoples perceptions of poverty in the
analysis? 2. Basket Consistency. Are the same
dimensions of poverty/well-being used when making
inter-personal or inter-group comparisons? 3.
Cut-Off Consistency. Is the poverty line being
drawn at the same level of well-being across the
domain of the comparison.
4. Aggregation Consistency. Is there an attempt
to add up those below the poverty line using the
same poverty basket and cut-off. 5.
Representativeness. Can findings be
extrapolated to other population groups? 6.
Reliability. Would the same results occur if data
collection and analysis were replicated by
different persons under identical conditions? 7.
Validity. Are research results accurate or
characterised by bias? The move to the Q2 in
poverty analysis can be thought of as an attempt
to satisfy those criteria not met singularly by
consumption or PPA poverty.
8
3. Methodological Features of Consumption Poverty
Consumption Poverty Key Features 1. The
dimension of well-being used is economic welfare,
i.e. utility or preference fulfillment
represented by money (consumption exp.) 2.
Consumption is adjusted across regions, using
price indices, and across household types, using
equivalence scales, to facilitate comparability.
3. The poverty line is set as the consumption
level at which dietary energy (caloric) needs are
met (food-energy methods) or corresponds to a
basket of basic need goods which satisfy caloric
requirements. 4. Consumption modules in household
surveys are the main sources of data.
1. Peoples Perceptions. No. The well-being
metric is set as preference fulfillment
represented by money 2. Basket Consistency.
Yes. 1 implies that the well-being basket is the
same for all (e.g. consumption) 3. Cut-Off
Consistency. Yes. The same poverty line is
usually applied to all in the dataset (though in
some methods (e.g. food share) an urban and rural
poverty line are calculated separately to reflect
different preferences. 4. Aggregation
Consistency. Yes. Points 2 and 3 imply
aggregation consistency. 5. Representativeness.
Yes. Most household surveys rely on probabilistic
sampling which allows for the calculation of
standards errors. 6. Reliability. Reliability is
promoted through standardization of questions and
questionning in enumerator training and the use
of standard protocols for data cleaning and
analysis. Problems remain re. enumerators,
etc. 7. Validity. Consumption is hard to measure,
sensitive to recall period, there are problems of
respondent fatigue, etc. Measurement error is a
problem but less clear if systematic bias will
result. We will ignore the problem of
utility inconsistency, whereby rankings are
sensitive to the reference consumption basket
used to adjust for price differences between
regions.
9
4. Methodological Features of PPA Poverty
(First Generation) PPA Poverty Key Features 1.
The relevant dimensions of well-being/poverty are
determined in consultation with people and vary
between sites. 2. As above, the poverty cut-off
is determined in consultation with people and
varies between sites. 3. PRA techniques including
focus groups, social maps, well-being rankings,
are the main sources of information.
1. Peoples Perceptions. Yes. The well-being
metric is set as preference fulfillment
represented by money 2. Basket Consistency.
No. Different dimensions of well-being/poverty
are found in different areas. 3. Cut-Off
Consistency. No. Different poverty cut-offs are
used in different areas. 4. Aggregation
Consistency. No. Points 2 and 3 imply
aggregation inconsistency. 5. Representativeness
. No. First generation PPAs did not use
probabilistic sampling and any claims to being
representative relied on casual empirical
generalisation. 6. Reliability. Reliability was
undermined by the lack of standardisation of
questions across sites, the lack of rigorous
training of PPA facilitators, the difficulties of
standardisation in focus group/semi-structured
interview settings, the failure to check conduct
multiple well-being rankings exercises (using
different respondents) to gauge inter-coder
reliability, etc. 7. Validity. PPAs provide more
opportunities for validation, cross-checking and
triangulation, are less impersonal, time
consuming and boring than standard HHSes. On the
other hand, there are problems of monopolisation
of dialogue by dominant groups silence on
sensitive subjects lying self or group report
biases, etc.
10
5. Case Study 1 Putting Together PPA vs.
Consumption Poverty
Q Do Peoples Perceptions Matter? Comparing
Consumption Poverty and PPA Results in the
Republic of Guinea (West Africa)? Research
Question Are females as a group more likely to
be poor or worse off than males? Methodology
Comparison of Results of national household
survey data (World Bank SDA-Integrated), and PRA
data from one village (1 month stay in 1997 in
Kamatiguia in North of Guinea). Source Shaffer.
1998. 'Gender, Poverty and Deprivation Evidence
from the Republic of Guinea.' World Development.
Volume 26. Number 12
11
5.1 Gender and Consumption Poverty in Guinea

• The following three issues addressed
• Comparison of poverty status of MHHs and FHHs
• FGT Poverty measures of each
• Stochastic dominance tests
• Representation of females in poor households
• Intrahousehold distribution of food

12
5.2 Poverty Status of MHHs and FHHs
Summary FHHs are not more likely to be poor than
MHHs. Result is not sensitive to choice of FGT
poverty measures or poverty lines (see next page
stochastic dominance tests)
13
5.3 Poverty Status of MHHs and FHHs
14
5.4 Representation of Females in Poor Households
Summary Females do not appear to be
overrepresented among poor and ultra poor
households.
15
5.5 Intrahousehold Distribution of Food
Summary Nutritional outcome Indicators do not
suggest intrahousehold discrimination against
females in food consumption
16
5.6 Summary Gender and Consumption Poverty

• Summary
• FHHS are not more likely to be poor than MHHs
• Females are not overrepresented in poor household
• Nutritional outcome indicators do not suggest
  gender-based intrahousehold discrimination in
  food consumption

17
5.7 Gender and Deprivation

• The following three tools used to address the
  question
• Focus Group Discussions
• Social Mapping/Well-being Rankings
• Follow-up Discussions

18
5.8 Focus Group Discussions
Village assembly was convened with around 60
people A number of issues were addressed
concerning well-being meaning and problems in the
village Gender question asked whether they felt
that women as a group were worse off than men as
a group. Result A large majority of both men
and women felt that women were worse off for two
reasons 1) Work Load/physical fatigue 2)
Decision-Making Authority
19
5.9 Well-being Rankings and Follow-up Discussion
Two Rankings Conducted Separately with groups of
men and women selected to reflect diversity of
wealth and age. Both rankings came to the
identical conclusion putting all village women
below all village men with the exception of two
(see ranking results next page). Why Men
Decision Making Authority (Cest lhomme qui
decide) Women Work Burden and Fatigue
(moso ye dyon ne di)
20
5.10 Well-being Ranking Results
21
5.11 Gender and Poverty in Guinea Summary
Women/Females are not more likely to be
consumption poor but may be worse off in terms of
a broader definition of poverty Peoples
perceptions do matter for 1) how poverty is
defined (the basket of poverty goods) 2) who is
considered to be poor 3) policy (should public
policy directly address aspects of female
deprivation.)
22
6. Case Study 2 Methodological Integration
Addressing Consistency and Reliability in PPA
Poverty
Issue Are there ways of addressing consistency
and reliability problems in PPAs? Research
Objectives i) to generate a quantitative
household economic status index that was directly
linked to qualitative statements about economic
well-being collected during participatory wealth
ranking, ii) to use participants descriptions of
what constituted poverty in their setting to
apply poverty lines to that index. Methods Focus
Group Discussions Well-being Rankings Location
8 rural villages of Limpopo Province, South
Africa
23
6.1 Methodology

• Data Collection
• A social map is constructed at a public assembly
  including all village households
• Focus groups discussions are held during which
  participants are asked to describe the
  characteristics of three groups very poor,
  poor, but a bit better off and those that are
  doing OK. The number of times that different
  terms are used to describe these groups is
  recorded.These resulting information is referred
  to as General Statements.
• Well-being Rankings are then conducted on
  households in the village and respondents are
  asked to provide characteristics of the different
  ranking groups. The resulting information is
  referred to as Pile Statements. The ranking is
  conducted on three occasions with different sets
  of villagers.
• Data Analysis
• Statement Coding (The General and Pile Statement
  are Coded into a number of categories, e.g. lack
  of food, shelter, jobs, etc.)
• Scoring of the Ranking Piles (according to the
  following pile n 100((N-n)/(N-1)), where n
  is the pile number and N is the total number of
  ranking piles).
• Scoring of the Pile Statements (weighted
  according to the 1) the number of times the
  statement was made in relation to the different
  ranking piles and 2) the score of the ranking
  piles (e.g. a pile statement mentioned 8 times
  in relation to the poorest pile in given rankings
  (scoring 100), and twice in relation to piles
  that were ranked second in ranking sessions that
  generated 5 wealth ranks (scoring 75 each time).
  In this case, the average pile statement score
  for this statement would be ((1008)(752))/(10)
  95.
• Calculating a Household Wealth Index. A household
  wealth index is based on the mean of the pile
  statement scores for all piles statements
  corresponding to the pile in which the household
  was ranked. Remember all households were ranked
  on three separate occasions.
• Generating a Poverty Line. An attempt is made to
  see if a visual relationship exists between the
  pile statement scores (listed in descending
  order) and the number of times these statements
  were mentioned in the General Statements. The
  resulting visual cut-off pile statement scores
  are then applied to the household wealth index to
  create different poverty groups.

24
6.2 Illustration of the Methodology
Data Collection
Data Analysis
Step 2 Pile Scoring (0-100)
Well-Being Ranking
Step 3 Pile Statement Scoring (times said re.
different piles)

• Pile Statements
• Not enough food
• Mental Illness
• Widowhood
• Etc.

Step 4 Household Wealth Index Average Pile
Statement Score for Piles in which Household is
Found
25
6.3 Illustration of Step 5 Determining the
Poverty Line
26
6.4 Using the Method to Determine Poverty
Incidence
27
6.5 Significance
The Methodology addresses a number of
methodological shortfalls of typical PPAs. 1. It
doesnt resolve basket inconsistency. It makes
interpersonal comparisons on the basis of pile
statement scores which are based on the different
characteristics attributed to the ranking groups.
It assumes an intrinsic link between these
characteristics and underlying wealth levels
across sites. 2. It addresses cut-off
inconsistency by applying a unique cut-off level
to all in the data set
corresponding to levels of pile statement/wealth
index values where a visual gap appears (though
the cut-off is acknowledged to be arbitrary. 3.
It only partially addresses aggregation
consistency because it uses a single cut-off but
different dimensions of poverty. 4. It addresses
reliability by conducting 3 ranking exercises of
all households and averaging the value when
calculating the wealth index. Big improvement
over standard PPAs in that it attempts to
facilitate interpersonal comparison of results,
is transparent about the basis of the comparison,
and makes an attempt to apply a single poverty
line across the domain of the comparison. Hinges
on the empirical soundness of the intrinsic
link between pile characteristics and underlying
material wealth across different groups.
28
7. Case Study 3 Methodological Integration
Enhancing Representativeness in PPAs
Issue Making PPA Statistically
Representative? Research Objectives i) to
estimate the proportion of poor household
eligible for targeted assistance (Targeted Inputs
Program (TIPS)) Methods Probabilistic Sampling
Well-being Rankings Location Malawi
29
7.1 Methodology

• Villages were sampled using a two-stage random
  process. In stage 1, districts were selected
  randomly from within each region. In the second
  stage, a village was selected at random from
  within each district. The work conducted within
  each village was equivalent to a full village
  census (see below), so the data at village level
  is free from sampling error.
• To enhance basket and cut-off consistency,
  food security was used as the poverty indicator
  in all sites. The followign three categories were
  used Food Secure (FS) Households that have
  enough to eat throughout the year from harvest to
  harvest.Food Insecure (FI) Households that have
  enough food to last from harvest up to Christmas
  but not between Christmas and the next harvest.
  (The harvest in Malawi is in April/May).Extremely
  Food Insecure (EFI) Households that have a
  longer period of not having enough to eat. These
  households start facing severe food shortages
  before Christmas.
• Food security was selected in that existing
  participatory studies found that it is perceived
  as a key indicator of poverty in rural Malawi
• A social map was constructed and key informants
  were asked to rank all village households into
  the three categories.
• Sampling design allowed for the calculation of
  standard errors of population proportions within
  the above three categories.

30
7.2 Significance
The Methodology addresses a number of
methodological shortfalls of typical PPAs. 1. It
resolves basket, cut-off and aggregation
inconsistency by restricting the poverty
categories to the extremely food insecure, food
insecure and food secure and adding up those
categories. 2. It incorporates peoples
perceptions by justifying the food security
emphasis on the findings of previous PPAs. 3. It
resolves representativeness by sampling using a
stratified random design and calculating standard
errors for poverty estimates.
Major remaining questions concern reliability,
i.e. if results are sensitive to the choice of
key informants used to group households into the
food security categories due to 1) imperfect
knowledge 2) different interpretation of the
food security categories, etc.
31
8. Case Study 4 Methodological Integration
Using People Perceptions in Poverty Line
Determination
Source Pradham and Ravaillion, 2000, Measuring
Poverty Using Qualitative Perceptions of
Consumption Adequacy, Review of Economics and
Statistics, vol. 82 Lokshin, et. al. , 2004,
Robustment of Subjective Welfare Analysis in a
Poor Developing Country Madagascar
2001 Objective Comparison of food share poverty
line (FSPL) with poverty lines derived from
subjective questions in HHS (Consumption
Adequacy Question (CAQ) Minimum Income Question
General Satisfaction Question) Methodology
Estimate an ordered probit model to predict the
Subjective Poverty Line (i.e. the consumption
level which just satisfies adequacy norms) and
then compare poverty headcounts with food-share
poverty headcounts for different
groups/regions. Results 1) Some versions of the
SPL accords well with FSPL overall, some do not
2) Even among those that accord well, overall,
there are differences for particular subgroups
(e.g. rural/urban, household size differences )
32
7 Summary Conclusion

• Poverty Identification is about determining the
  characteristics/correlates of the poor and the
  relevant dimensions of well-being to use.
• Standard analyses of poverty (consumption and PPA
  poverty) can be distinguished according to seven
  criteria 1) whether or not they take into
  account peoples perceptions, 2) whether or not
  they are comparing the same things (basket
  consistency) 3) whether or not they apply a
  consistent poverty line across all persons
  (cut-off consistency) 4) whether or not they
  are able to add up those below the poverty line
  in a consistent way (Aggregation) 5) whether
  or not they provide representative data 6) the
  measures they use to enhance reliability 7) the
  measures they adopt to enhance validity.
• The use of Mixed Methods in Poverty Analysis can
  be seen as an attempt to satisfy those criteria
  not met singularly by consumption or PPA poverty
• Examples were given of mixed methods of the
  putting together (Shaffer) and Methodological
  Integration (Hargreaves, Barahona, Ravallion)
  types.
• The choice of poverty indicator and the
  methodological implications of that choice have
  bearing on all other aspects of applied poverty
  analysis.