Category Archives: summerschool 2011

Unidimensional Poverty Measurement – right sidebar

Download materials

Lecture Slides

Exercise

Answer Key

Download Video

Download Audio

Useful resources:

For supplementary reading on this topic view

Reading List

Key readings covered in these lectures

Alkire, S., Foster, J. E., Seth, S., Santos, M. E., Roche, J. M., and Ballon, P. (2015). Multidimensional Poverty Measurement and Analysis, Oxford: Oxford University Press, ch. 2.

Video 1:

Foster, J.E. (2006). Poverty Indices. In Janvry, A. and Kanbur, R. (Eds). Poverty Inequality and Development: Essays in Honor to Erik Thorbecke. New York: Springer

Foster, J.E. and Sen, A. (1997). On Economic Inequality: After a Quarter Century. Annex to the Expanded Edition of A. Sen. On Economic Inequality. Oxford: Clarendon Press. Section A.6.

Video 2:

Foster, J.E. and Sen, A. (1997). On Economic Inequality: After a Quarter Century. Annex to the Expanded Edition of A. Sen. On Economic Inequality. Oxford: Clarendon Press. Section A.6.

Foster, J.E. and Shorrocks, A. (1988). Poverty Orderings and welfare Dominance. Social Choice and Welfare. 5: 179-198

Atkinson, A.B. (1987). On the Measurement of Poverty. Econometrica. 55:749-764

 

AF Measure Analysis Issues IV: Redundancy, correlation, complementarity, subjective scales validation (uses of principal components & factor analyses)

Jose Manuel Roche

  • Factor analysis; latent variable analysis help to define weights and final indicators.
  • Two types of factor analysis; exploratory and complementary
  • Principal component analysis.
  • Subjective scale validation.

Watch the video (includes video guide)

Listen to audio

Video

Guide to video

00:00 Introduction

02:42 Outline of the lecture

03:51 Definition of factor analysis, relate to latent variable analysis

08:28 Example of Rustein and Jonston (2004) Wealth Index, principal component analysis

11:37 Other options besides factor analysis (data reduction)

16:58 Example of Lelli (2008), factor analysis vs. fuzzy set theory

Part 1: Exploratory and Confirmatory Factor Analysis

22:08 Definition of exploratory (unrestricted) factor analysis (EFA); calculation steps

27:54 Example of principle component analysis (Klasen, 2000), here weight are based on the correlations between indicators – a statistical solution (see also Normative Issues in Multidimensional Poverty Measurement for choice of weights)

33:30 Confirmatory factor analysis – restricted analysis. Look in Brown’s book both for exploratory and confirmatory reference Brown (2006)

39:50 Literature overview on the goodness of fit

Part 2: Empirical Implementation of Exploratory Factor Analysis

41:16 The different step of exploratory factor analysis (Brown (2006))

43:03 Step 1: the example of Roche 2008

44:46 Step 2: extraction method

46:30 Step 3: determination of appropriate number of factors; sources – Kaiser criterion, Analysis of Scree plot, parallel analysis, normative judgement

52:07 Step 4: method of rotation to obtain your simple structural model, orthogonal and oblique rotation

61:40 Example of Roche (2008), rotation results – construct 3 indices based on factor analysis, discuss weight (link to lecture on normative issues)

70:04 Step 5: interpretation an evaluation of the quality of the solution

70:19 Example of Roche (2008), results of different models

75:28 Example of Gallo, Cesar and J.M. Roche (2011), the implicit weight arising from different clusters of variables

76:24 Tetrachoric correlations

77:29 Example Gallo, Cesar and J.M. Roche (2011), results

Part 3: Subjective Scale Validation

81:11 Subjective scale validation

81:50 Psychometric evaluation

81:32 Example of Gagne et al (2009), autonomy and work

84:33 The process of subjective scale variation

84:57 Techniques for the different stages of scale validation

85:22 Crombach Alpha

86:05 Example of Steger et al (2006), subjective scale validation, the meaning of life

87:30 Example, subjective scale validation, psychological needs

88:57 Exploratory factor analysis results

91:20 Complementary factor analysis results

92:10 Strength and weaknesses of factor analysis

Listen to audio

Ongoing Debates and Research Topics

Sabina Alkire

  • Debates about the AF method.
  • Clarification of misunderstandings in the debate.
  • Discussions about the global MPI in particular.
  • Review of different research questions requiring further work.

Overview of Research Topics

Watch video (includes video guide)

Listen to audio

Overview of Research Topics

The list below is not comprehensive, but covers some of the key research issues that have been identified by OPHI researchers. Please watch Part 2 of the video to get a further explanation by Sabina Alkire.

Research Issues: Data for MPI

  • Can we improve international data for MPI calculations?
  • Should we ‘call for’ a survey having a specific set of dimensions and indicators? And which?

Research Issues: General data

  • Data Constraints: Most criticisms address these (why don’t you include _____?).
  • identify ‘new’ questionnaires on standard surveys
  • identify ‘new’ dimensions
  • survey problems: individual level data, excluded populations
  • the consequence of the combination of surveys, administrative data, mapping
  • the consequence of the combination of data for different reference groups

Research issues: implementation of multidimensional measures

The example of the international MPI

  • Robustness checks (weights, bootstrapping, household size, indicators)
  • Aggregation within households – what biases exist?
  • Households without eligible populations – the impact on the measure?
  • Comparisons across surveys from different years – sample, respondents
  • Thorough investigation into the assets indicator in the living standard dimension
  • Indicator scrutiny: Stunting, Quality of Education, etc
  • Which are the ideal indicators to use for an international MPI?
  • Development of MPI-2 appropriate for countries with higher human development.
  • Rural/Urban poverty? Should we adjust urban indicators?
  • The impact of overlooked populations: elder poverty

General empirical implementation of the AF method: (example of questions)

  • Statistical interrelationships among indicators – Systematic.
  • The relationship between income/consumption poverty and deprivations
  • Combining individual and household level data
  • Investigate the issues of intra household inequalities, household composition, predicting household deprivations on multidimensional measures
  • Weights (ordinal ranking)
  • Development of group-specific indicators (child, ethnic minority)
  • Multiple cut-offs with ordinal data (extreme poverty, poverty)
  • Develop multidimensional health indicators, quality of schooling, governance, child poverty, etc.
  • Use of biomedical data to set weights – nutrition measure

Research issues: the development of the methodology

  • Develop time series and panel data methodologies, expand to include chronic poverty
  • Robustness tests (weights, cut-offs, indicators)
  • Test statistics, measurement error, uncertainty, inference
  • Appropriate validation ‘tests’ for national measures
  • Algorithm for adjusting weights for ordinal data and M1 M2
  • Complementarity and Substitutability
  • Chronic Multidimensional Poverty
  • Axioms – for Chronic Multidimensional Poverty
  • Characterisation of AF
  • Axiom – post-identification decomposability
  • Use of ordinal data and the depth of poverty

Video

Guide to the video

00:00 Introduction and outline

Part 1: The Different Debates on the AF Method and its Implementation in the Global MPI

00:49 MPI in the Human Development Report 2010 fostered debate and critique from Ravallion and national governments [see Alkire and Foster (2011)]

03:05 Ravallion’s critique

03:58 Critique 1: a single index

10:01 Critique 2: why aggregate to later break the index down afterwards

13:37 Critique 3: attainment vs deprivation space

16:21 Critique 4: prices as weights

19:21 Alternative to index suggested: a dash board (see Identification and Aggregation in the Alkire Foster Method)

28:11 Critique 5: data constraints (can this be avoided?)

30:10 General critiques of international data by OPHI and others, OPHI’s response to data issue critique

32:55 Critique 6: lack of policy consultations/participation of the poor

35:09 The discussion and debate on weights

Part 2: Review of Research Issues

38:20 Research issues 1: data constraints, the ideal data, surveys etc.

44:34 Research issues 2: implementation of the Alkire Foster method. The example of the global MPI

57:55 Research issues 3: empirical questions

68:17 Research issues 4: development of methodology

Listen to audio

Review of the Course & Communicating Your Multidimensional Measure

Sabina Alkire

&

Joanne Tomkinson

  • Recapping key lessons and topics covered during the course
  • Communicating your multidimensional poverty measure.
  • OPHI’s experience in communicating the global MPI.
  • General communication advice.

Watch the video  (includes video guide)

Listen to audio

Video

Guide to video

Part 1: Overview of the Course

00:00 Introduction

03:13 Axioms in the multidimensional space

06:00 Why new emphasis on multidimensional measures?

09:00 The challenges to multidimensional poverty measures

09:55 The Alkire Foster method – short overview

11:45 Informal glossary of terms in the AF method

12:34 Examples of case studies applying the Alkire Foster method: the global MPI, Mexico, Colombia

(18:00 – Technical interruption in the video)

17:55 Sen, data constraints, and the capability approach

18:58 The six essential choices which you have to consider when constructing an Alkire Foster based measure, and the source of information that can guide you in these choices

19:54 In practise – justify your measure

23:33 The normative issues

24:27 Stochastic dominance and justification for the method

26:58 Standard errors, vital for comparisons

27:35 Decomposition by subgroup and indictor/dimension

30:00 Factor analysis and redundancy, strengths and weaknesses

33:00 Missing dimensions

34:08 The political context

Part 2: Communicating Your Measure – the Actions after you have Computed Your Measure

34:55 Introduction

35:42 Why communicate your measure?

36:49 Find the human angle to your measure – it is not simple to communicate complex measures, find your killer statistics

39:02 A starting point and the building blocks (goals, audiences, messages, and products)

41:00 Audience type

42:11 Outputs and channels

44:38 Choose your messages carefully

45:14 Think about media diversity

47:40 Media tactics – e.g. press releases

48:51 Examples of media coverage of the global MPI 2010

49:53 Tips to communication

Listen to audio

Introduction to the Capability Approach

Sabina Alkire

  • The history and motivation for Sen’s work.
  • Definitions of capabilities, functionings, freedom, and agency.
  • The connection between the Capability Approach and issues for poverty measurement.

Listen to audio

Video (with guide)

Guide to the video

00:00 Introduction to an overview of Sen’s approach and complementary initiatives

02:56 An overview of the key publications in Sen’s work, note that there is a large secondary literature.

08:49 Sen’s background

Part 1: Understanding the Terms of the Capability Approach

10:10 Introduction to the capability approach, unpacking the terms included in the capability approach (freedoms, functionings, agents, capabilities).

14:25 A definition of functionings.

18:42 The interpretation of the space of resource, capability, functioning, and utility with an example of a bike and food

28:00 A definition of freedom

33:07 A definition of agency, brief, not the focus of the lecture

34:58 The link between capability and agency

36:55 Sen’s process freedom

40:18 Common misunderstandings of Sen’s capability approach

49:37 Introduction to other key authors’ work on Sen; M. Nussbaum and I. Robeyns

52:38 D. Thomson’s application of the capability approach to her own family

Part 2: The Capability Approach and Poverty Measurement

55:08 The link between the capability approach and measurement, the notion of space, indicators are often achieved functioning, and not capabilities (see also Normative Issues in Multidimensional Poverty Measurement)

58:55 Question asked to reflect on your own multidimensional poverty measure

60:17 Link between MPI and the capability approach, and a discussion on the practical implementation of the capability approach – how the AF methodology allows for diversity with valued functionings (a k cut-off larger than union)

66:21 The capability approach’s relation to human development (and the HDR), they have the same objective

70:14 The capability approach’s relation to other conceptual framworks (MDGs, human rights, human security, happiness)

A short introduction to the computation of standard errors for AF Measures

Gaston Yalonetzky

  • Computing the standard errors of averages in simple surveys.
  • Computing the asymptotic standard errors for ratios (like the average deprivation score (A)).
  • Asymptotic standard errors fro percentage changes over time (cross-sections and panel datasets).
  • Discussion of the importance of complex survey design for standard errors.

Watch the video (includes video guide)

Listen to audio

Video

Guide to video

00:00 Introduction to standard error and the two different types; statistical and analytical standard errors, here the analytical (sub population) approach is applied

10:45 Outline of the lecture

Part 1: The Formulas of Standard Error Computations in the AF method

12:57 Define and describe standard errors in general (sub-population approach)

17:04 Computing the standard errors of the censored headcount

24:15 Computing the asymptotic standard errors of the average deprivation score (Yalonetzky (2011))

33:17 Computing the standard errors for percentage change over time

35:16  Computations in cross section data

36:21 Computations in panel data

Part 2: The Importance of Survey Design for Standard Error Computations

39:17 Computing standard errors in complex surveys (Deaton (1997))

40:52 Define strata

43:08 Define cluster

45:48 Define sample weight

48:18 Explain the STATA command: SVY for computing standard errors

54:16 An example that shows that sample design is not a trivial matter (Lynn (2007))

59:40 Concluding remarks, comparing the sub and final population approach to standard errors.

Listen to audio

Multidimensional Stochastic Dominance

Gaston
Yalonetzky

  • Unidimensional versus multidimensional stochastic dominance.
  • Introduction to the four conditions for bivariate dominance.
  • Extending the bivariate case to the multidimensional case.
  • Illustration of the link between ALEP property and dominance conditions.
  • The application of dominance in well-being to poverty measurement.
  • Statistical inference techniques

Key readings

Related lectures

Watch the video (includes video guide)

Listen to audio

Downloads

Key readings covered in this lecture

Atkinson, A. and Bourguignong, F. (1982). The comparison of multi-dimensional distributions of economic status. Review of Economic Studies XLIX, 183-201

Barret, G. and Donald, S. (2003). Consistent tests for stochastic dominance. Econometrica. 71(1): 71-104

Yalonettzky, G. (2011). Stochastic dominance for ordinal variables: conditions and tests. Forthcoming in Econometric review.

Related lectures

Unidimensional Poverty Measurement: Axioms, Measures and Dominance

Video

Guide to the video

00:00 Introduction and outline of lecture

05:45 Defining dominance (see also Unidimensional Poverty Measurement: Axioms, Measures and Dominance)

Part 1: Unidimensional Dominance

15:21 Unidimensional dominance

Part 2: Bivariate Dominance Conditions

24:09 The bivariate case of dominance, first order conditions, survival functions, ALEP complements and substitutes

32:07 Dominance condition 1 – A condition for monotonically increasing functions with ALEP substitute arguments

40:50 Dominance condition 2 – A condition for monotonically increasing functions with ALEP complement arguments

45:32 Dominance condition 3 – A condition for monotonically increasing functions with ALEP neutral arguments

52:22 Dominance condition 4 (strongest condition) -A condition for ALL individual welfare functions that are monotonically increasing

Part 3: Multidimensional Dominance Conditions

58:52 Extending the bivariete dominance to the multivariate case (see Yalonetzky (2011))

66:40 The relationship between ALEP properties and dominance conditions

Part 4: From Well-Being to Poverty Measurement

78:26 the application of well-being dominance to poverty measurement

Listen to audio

Case Studies – the international MPI, and national multidimensional poverty measures for Mexico (2009) and Colombia (2011)

Sabina Alkire

  • An introduction to the choices and limitations behind the international comparable MPI and the many types of possible analyses.
  • An overview of two national A&F measures in Mexico and Colombia.

Watch video (includes video guide)

Listen to audio

Video

Guide to the video

Part 1: The Case of the global MPI

00:00 Introduction to the global MPI

03:10 Introduction to the components of MPI: surveys used, dimensions and weights chosen, and data restrictions on international comparable survey data

05:20 Explanation of the health dimension – variables and deprivation cut-offs

06:37 Explanation of the educational dimension – variables and deprivation cut-offs

07:58 Explanation of the living standard dimension – variables and deprivation cut-offs

11:54 Data constraints, a call for better data, and a note that MPI is not appropriate for national policy

15:46 Explanation of the equally (nested) weights in MPI – same as HDI, pasted robustness check, (most importantly) they are easy to communicate (Atkinson)

18:35 Identification (z-cut and k-cut offs); the debate/choice of the poverty cut-off (k).

21:48 Aggregation and limitation to the adjusted headcount (M0), as MPI is based on ordinal data.

23:10 Examples from qualitative work done on the global MPI.

26:10 Present the MPI 2010 results, comparison to $1.25/day, introduction to the many research questions relating to construction of a multidimensional poverty measure and MPI work within the topic (see lecture on Ongoing Debates and Research Topics).

33:56 Short introduction to subgroup and dimensional decomposition and contribution by indicator   (see lecture on Decomposition by Dimension and Subgroup)

37:03 Discuss the need to look at different poverty cut-offs

39:00 Short introductions to standard errors and time comparisons (see lecture on Standard Errors and Time decompositions)

40:36 Robustness tests done on MPI for k cut-offs and weights (see lecture on Robustness Analysis)

Part 2: The Case of Mexico and Colombia

43:00 Mexico’s multidimensional poverty measure based on the country social development law and introduced in 2009. (see Foster (2006))

46:37 Colombia’s MPI – a complement to income measures. Dimensions, weights and analysis are presented. (see OPHI website page on Colombia)

Listen to audio

Identification and Aggregation in the Alkire Foster Method

Sabina Alkire

  • Key literature on multidimensional poverty measures
  • The steps of identification and aggregation in Alkire Foster method
  • The importance of a joint distribution vs a marginal distribution

Audio

Watch the video (includes video guide)

Video

Guide to the video

00:00 General introduction

06:00 Introduction to A&F; measurement methodology; the lecture focuses on the identification and aggregation

Part 1: Why Multidimensional Measures?

10:30 Review unidimensional measures, as the concept of identification and aggregation can be translated into the multidimensional space.

14:20 Challenges of unidimensional measures

16:20 Why multidimensional measures of poverty?

Part 2: The Dual Cut-off Approach: the Main Question Being “Who is Poor?” Sen (1976)

20:07 The first step: identification using the deprivation matrix and z-cut offs

25:04 The second step: aggregation (censoring of data)

27:12 Explanation of the censored headcount, H

28:16 Explanation of the average share of deprivations among the poor, A

29:16 Explanation of the adjusted headcount, M0, including the properties of the measure

32:54 Explanation fof M1 and M2, in the case of cardinal data

37:35 The importance of normative issues (see also Normative Issues in Multidimensional Poverty Measurement)

42:27 The importance of axioms in doing methodological research

47:47 The axioms/properties of M0

48:35 Application of weights to the identification and aggregation steps to get H, A and M0 with weight applied (see also paper based exercise for this lecture)

53:27 Example of USA (decomposition, contributions of deprivations, dominance)

56:07 Example of Indonesia

58:46 More empirical examples

Part 3: Marginal vs Joint Distributions (see also Ongoing Debates and Research Topics)

59:45 Important points: marginal vs joint distributions

64: 15 Value of a joint distribution (marginal does not identify who is poor)

69:15 Censoring process

70:14 Terminology used in the AF method, which is different from income poverty measures due to the dual cut-off.

Why Multidimensional Poverty Measures?

Gaston Yalonetzky

  • Conceptual and philosophical arguments for multidimensional poverty measures.
  • Overview of public debates, Stiglitz Sen Fitoussi Commission.
  • The challenges for, and debates of, multidimensional poverty measures.

 

Listen to the audio

Watch the video

Guide to the video

00:00 Introduction

Part 1: Why a Renewed Interest in Multidimensional Poverty Measures?

06:28 Overview

11:06 1: Relevant data is increasing

13:12 2: Multiple poverty measures are exploding

14:50 3: The 2010 Human Development Report – three new indices

16:50 4: Technical resources do not reflect human development

17:33 5: The political demand is increasing

24:18 Introduction to the Stiglitz Sen Fitoussi Commission

32:33 Focus of the Stiglitz Sen Fitoussi Commission

Part 2: Why Multidimensional Poverty Measures?

34:28 Income measures are incomplete

38:45 Justifications of multidimensional poverty measures

Part 3: The Debates of Multidimensional Poverty Measures

40:57 The two major challenges for multidimensional poverty measures; why replace incomes? why a composite index?

45:09 First challenge posed by (among other Ravallion): monetary poverty is multidimensional

47:29 The answer to the challenge: the problems with monetary poverty

55:23 Second challenge: why not use the dash board approach to multidimensional poverty measures?

61:23 The answer to the challenge: the problems with the dash board approach to multidimensional poverty

70:17 The empirical challenge to multidimensional poverty

74:47 Challenges, improvements and best practices of multidimensional poverty measures

83:00 An overview “the reasons for using a multidimensional poverty measure”