James Foster George Washington University and OPHI 2 nd Conference on Measuring Human Progress 4-5 March 2013, New York Reflections on the Human Development.

James FosterGeorge Washington University and OPHI

2nd Conference on Measuring Human Progress4-5 March 2013, New York

Reflections on the Human Development Index

Introduction

• HDI – Remarkable accomplishment– Attracted criticism• Arbitrary decisions

• Purpose of this short talk– Focus on aspects of HDI that might be improved– Consider alternatives– Offer suggestions• View as less arbitrary

Desiderata

(Ia) It must understandable and easy to describe(Ib) It must conform to a common sense notion of what is being measured (II) It must fit the purpose for which it is being developed (III) It must be technically solid (IVa) It must be operationally viable(IVb) It must be easily replicable

Desiderata

• (I) corresponds to strong policy needs– Understandable at a deeper level including cutoffs– Measuring absolute size of HD

• Atkinson’s independence of other countries – not relative

– Anchored in underlying variables– Numbers mean something

• (II) concerns the intended purpose of the measure– Compete with GNI per capita– Compare HD achievements across countries – Monitoring progress across time for a given country – Drilling in or out (subgroups or dimensions)

Desiderata

• (III) is is theoretical justification– Axioms to make sure measure conforms to purpose– Theoretical framework (say within welfare economics)

• (IV) concerns practicality– Does it work with existing data?– Can it be updated in time?

• These are benchmarks to compare and evaluate particular versions of HDI

• Contrast with GNI per capita– Poor in (II), better in others?

Outline of Paper

• Reconsider frequent recalibration of top and bottom goalposts– Confounds understanding over time– Gives countries unclear signals

• Reconsider HDI demarcations into relative groups– Purely relative to deflect criticism– But ends up deflecting incentives

• Reconsider functional form– New geometric has certain characteristics– Old arithmetic has others– Which is best?

Outline of Paper

• Other themes– How to anchor HDI values

• Through normalized variables or through original variables?

– Purely data driven goalposts• Cause much confusion• Ought to have firm normative basis• Not just a (relative) function of observed achievements

– Differentiate purposes of goalposts• Upper (aspiration) vs. lower (natural zeroes)• Lower should stay fixed• Upper may change periodically

– Are irrelevant for new HDI, can be made so for old

Lower Goalposts: Natural Zeroes

• Fixed natural zeroes– Why should they change over time?– Conflicts with description of HDI as measure of absolute

size – Need to satisfy measurement properties– Analogous to poverty cutoff for countries

• Desiderata– Needed for measure to be simple and easy to explain– Needed to conform to measurement of HD– Fits purpose to compare over time and space

Understanding the HDIs

• Gist in 2-dimensional graphs– From original variables– To bottom normalized data (net variables)• Each measured from natural zeroes

Original Variable 1

Original Variable 2

natural zero

natural zero

Original Variables

Net Variable x1

Net Variable x2

00

Net Variables

Net Variable x1

Net Variable x2

00 Aspiration level

Aspiration level

Upper Goalposts: Aspiration Levels

Net Variable x1

Net Variable x2

00 Aspiration level

Aspiration level

New HDI

WA

Wx

x

Wx = x11/2 x2

1/2

Net Variable x1

Net Variable x2

00 Aspiration level

Aspiration level

New HDI

WA

Wx

x

HDIN = Wx/WA

Net Variable x1

Net Variable x2

00 Aspiration level

Aspiration level

New HDI

WA

Wx

x

HDIN = Wx/WA

Net Variable x1

Net Variable x2

00

New HDI

WA

Wx

x

HDIN = Wx/WA

Only uses reference level WA

Functional Form

• HDIN can be viewed as a social evaluation function of net variables, normalized by a reference level.– No need for upper goalposts or aspiration levels• But can use to set reference level

– New ref level yields a multiple of previous HDIN

• Same rankings• Same rates of growth• Other convenient properties

Functional Form

• HDIO: Arithmetic mean is used– But exactly how?– Return to graph

Net Variable x1

Net Variable x2

00 Aspiration level

Aspiration level

Upper Goalposts: Aspiration Levels

Net Variable x1

Net Variable x2

00 Aspiration level

Aspiration level

Upper Goalposts: Aspiration Levels

Set the slope of HDIO indifference

Net Variable x1

Net Variable x2

00 Aspiration level

Aspiration level

Upper Goalposts: Aspiration Levels

Set the slope of HDIO indifference

Net Variable x1

Net Variable x2

00 Aspiration level

Aspiration level

Upper Goalposts: Aspiration Levels

What if aspiration levels change?

Net Variable x1

Net Variable x2

00 New

Aspiration level

Aspiration level

Upper Goalposts: Aspiration Levels

What if aspiration levels change?Slopes change - inconsistent

Old Aspiration level

Net Variable x1

Net Variable x2

00 Aspiration level

Aspiration level

Upper Goalposts: Aspiration Levels

Always?

Net Variable x1

Net Variable x2

00 Aspiration level

Aspiration level

Upper Goalposts: Aspiration LevelsSlope unchanged if new levels are proportionate

Slope unchanged if new levels are proportionate

Net Variable x1

Net Variable x2

00 Aspiration level

Aspiration level

Upper Goalposts: Aspiration Levels

HDIO = Wx/WA

WAWx

x

Functional Form

• HDIO can be viewed as a social evaluation function of net variables, normalized by a reference level – if aspiration levels stay in proportion– New ref level yields a multiple of previous HDIO

• Same rankings• Same rates of growth• Other convenient properties

Summary of Suggestions

– Leave natural zeroes as natural zeroes– Alter aspiration levels only infrequently• 5 – 10 years normative targets• In a constrained way (or proportionate)• All past inconsistencies will then be caused by data

updates– Not by HDRO

Summary of Suggestions

– Fix absolute demarcation cutoffs for categorizing countries• Alternative 1: choose relatively, then fix absolutely• Alternative 2: Look within variables for natural cutoffs

– Note: They will be arbitrary• Like poverty lines, like middle class ranges• But if fixed over time, countries can progress• And given above recalibration methods

– Consistent cutoffs can be maintained over time

Summary of Suggestions

– Go back to original arithmetic formula• With fixed zeroes• With aspirations cutoffs constrained and updated

infrequently– Normative, not positive

Summary of Suggestions

– Why arithmetic?• Simple and understandable• With above calibrations, similar properties to HDIN

• Decomposability by dimension (as in MPI)• Potential for decomposability by subgroup if data

permits» Mean of log individual incomes» Rather than log of mean incomes

Summary of Suggestions

– What about inequality across dimensions?• HDIN accounts for it

• HDIO does not– Note the Atkinson Inequality measure: IA = (HDIO – HDIN)/HDIO

• But this form of inequality is not the priority!• Within dimension inequality is key!

– Focus: Need data to move to IHDI as the standard

Summary of Suggestions

– Alternative transformations for variables?• Careful to maintain simplicity!• Chakravarty: HDI0 form but with all variables

transformed by a common concave function• Of course this is possible

– Simplicity? Lose the close connection between variable and measure – it is a function of the normalized variable!» But can decompose by dimension

– No possibility for subgroup decomposition» Even if alter data

– Why transform income and others identically?

Thank you!