From:  Matt Berkley

To:  richard_zeckhauser@harvard.edu

Wed 02/07/03 15:04

Economics and poverty

Dear Professor Zeckhauser

Thank you very much for your reply via Jonathan.     I hope you will forgive me for answering your points in some detail, and for offering a fairly robust defence of the points in this letter.    The thrust of the letter is:   I agree that many economists are aware of the need to take certain (essential) variables into account in their assessments of aggregate economic gains to poor people;   but the most influential studies take no account of them.

I welcome your comments, because they help me to be more specific in my summaries for a popular audience.   I have great respect for social scientists studying deprivation who have developed methods for taking survival into account, and the smaller needs of children, and assets, and food prices.   I therefore have great respect for your own approach to these questions.    I am aware that many economists have tackled such issues at a theoretical level, and undertaken detailed studies which include these aspects of real life.   The church of economics is a broad one.  

However, the fact remains that in a debate such as that on growth and distribution  -  which is regarded as a key issue in development economics   -  there is almost no attention to these issues.   There is no attention to them whatsoever in most macroeconomists’ conclusions from empirical findings on income.   The findings’ relevance to costs and benefits to poor people is therefore unknown.

The income share of the poorest fifth, under conditions of zero change in average income, does not tell us whether the poor consumed more or less:   that is largely because of the mortality problem and  the inflation problem.    Has growth been good for the poor?    That question is not answerable from cross-sectional statistics on income.   

Studies using such data are, ostensibly at least, the most influential in the world today in relation to poverty.    They are promoted by the OECD and the World Bank as supportive of certain policies.    So are studies of the proportion below the dollar-a-day line, which suffer from both the mortality flaw and the inflation flaw. 

I absolutely agree that the points in my letter could have been more precisely targetted.   How to do this in a confined space and still get a letter published is a bit like the challenge of writing poetry.    I am learning.   My letter was in the context of the article to which it was replying:   Professor Sachs is Special Adviser to Kofi Annan on the Millennium Goals, and referred in his article to several goals which are, as I have been saying, defective in concept.    That is not to say that he is unaware of the need for multi-dimensional measures of progress.    I am sure that he is.    

There is certainly something very wrong with the Millennium Goal on poverty, and with the idea of reducing the proportion of poor people as laid down in national Poverty Reduction Strategy Papers;   I could have made clear in my letter to the paper that it is macroeconomists’ measures of outcome (and some statisticians’ measures on, for instance, the Millennium Goal indicator on nutrition) which are defective.   Sen’s theoretical work on the proportion of people in poverty did not deal with the problem, and the general use of cross-sectional averages to infer average gains suffers from the same problem.  

A similar consideration about targetting applies to the next point in my letter, about inflation.    A better formulation would have been:  “I hope that all macroeconomists’ studies of poor people will take inflation for poor people into account”.   It is true that no studies of the world’s poor have done this.    So both you and I are right:  many researchers take inflation into account in smaller-scale studies, but cross-country studies of countries where most humans live have not.   Nor have studies of the distribution of income in single countries.    Studies of the distribution of the numerical value of income are, in most countries, of no known relevance to the economic progress of poor people.

I should have said “I hope that social scientists will always recognise the existence of these variables in their work, because they are absolutely essential to the conclusions”.   Perhaps I had my audience in mind:   my observations do apply to all the major claims from macroeconomists on the progress of poor people which the readers of that newspaper, or any other, have read.  


There are two separate areas:  

1) those where official statements do not meet accepted norms

2) those where the theory of welfare economics is lacking.

In 2001 I called, among other people, Ravi Kanbur of Cornell to express my concern that macroeconomic statistics moved in the wrong direction if poor people died.    This concern arose from the use of data on “poorest fifths” in the World Bank document “Growth is good for the poor”, and from the use of the proportion in poverty (and other measures such as the poverty gap) to infer economic gains to poor people.   

Professor Kanbur has now written a paper saying essentially the same thing.   It is at <http://www.arts.cornell.edu/poverty/kanbur/Pov&Death.pdf> .    In this case the theory has clearly been defective.    Cross-sectional statistics are simply the wrong way to measure aggregate costs and benefits to people, unless there is reason to believe that survival rates had a particular trend.

Prior to my conversation with Professor Kanbur I wrote to Jeffrey Sachs making the same point:  the message is appended below.    Why did I not work through Ravi Kanbur to publish what I wanted to say in 2001?    Maybe because I sensed that my own approach was fundamentally different.   

At the time, he seemed to confuse questions about benefits to future generations with questions about longitudinal trends in poor people’s consumption during a period.    I regarded this as both intellectually confused and morally offensive.    You cannot even say how much later generations have benefitted in consumption terms if you ignore survival rates.    

An article by Erik Thorbecke, for the UN University conference on income inequality in May this year, sensibly notes that Kanbur’s point transcends other issues in poverty measurement, and calls for the use of lifetime poverty measures.    After my conversation with Professor Kanbur in 2001  -  in which I found myself in the unexpected position of answering his questions, rather than the other way round  -  I went through a very strange psychological process.    It is hard to describe the feeling that you understand something which no-one else in the world does.    A (perhaps small) determinant of my state of mind may have been that I sensed there was something extremely unscientific about all these macroeconomists’ measures, and in their modes of thinking.

Later, I contacted people such as Gary King and philosophers of science in the hope that they would see the issues more clearly.    Gary King does.   But the people who know the detail are those making the unfounded inferences.    One aspect in which they have done this is in relation to neglect of the cost of living.     That was so obvious that it took me three years to realise it.   

Where Professor Kanbur is still spectacularly wrong is over inflation.   He tells me that it is not economists’ fault that the data on prices are not available.   So does Kenneth Arrow.   They are right in what they say, but that is not relevant to the point which I made to them.   My point is this:  without data on the change in the cost of living for poor people, a social scientist studying income has no basis for a claim that poor people did better or worse.    When we talk about the cost of living for hungry people, we are referring to a very literal sense of the word “living”.   

There is no theoretical requirement for economists to take the inflation rate for the poor into account.    This allows official agencies to make unfounded statements on the progress of poor people, and wastes the time of academics in discussing statistics of no known meaning.

There should be such a requirement, because otherwise macroeconomics is in a state of indiscipline, or anarchy.     The current situation in economics is this:   a government or intergovernmental organisation can take action which increases mortality rates and/or puts up prices for the poor, and then use economists to say that the poor have done better.

In relation to the proportion of adults, UNFPA graphs show a marked decreased in the dependency ratio during the past couple of decades for Asia and Latin America, but not for Africa.    Unless they know that these trends did not apply to poor people, macroeconomists would therefore be wise not to treat percentage income rises per person as representing income gains.   This is a general problem in macroeconomics:   the work of David Bloom shows how great the effect of demographic shifts can be on economic growth statistics (about a third of the 6% annual growth rate in the “East Asian economic miracle”).    The rise in the average is not the same statistic as the average income gain.

Also, there is a greater proportion of adults per child in some countries than in others.   Per capita statistics do not tell us about income relative to need.     Here again, many statements about poverty are defective.    Economists are certainly aware of these problems:   Angus Deaton has been telling them to adjust for children for twenty years or more.    In relation to official statistics, Martin Ravallion of the World Bank wrote a paper drawing attention to the need to adjust accurately for children in the Economic Journal in 1995.    His own global poverty counts (and the World Bank claim about globalisation based on these counts) fail to adjust for children.    How the necessary adjustments would affect other conclusions, for instance about growth and poverty, is not known.    

Price data for poor people are not taken into account in any major cross-country studies on poor countries.    This illustrates an alarming tendency among macroeconomists to reject anything which is not “hard” data, but then to come to an unsupportable conclusion about what the data mean. The sane approach is to take whatever data there are  -  or anecdotal evidence  -  and lay out the assumptions honestly, and state that the conclusions are estimates based on assumptions.    Data on price trends are available from detailed smaller-scale studies and from a wide variety of other sources.    To infer economic gains to hungry people from the nominal value of income adjusted by the overall inflation rate is not a sane approach.

Further, the main dataset from the World Bank on income, the Deininger and Squire dataset  -  which is used for all such comparative studies from macroeconomists on poor countries  -   is itself unreliable for prima facie reasons:    these are sparse data (separated often by many years);   they use different methods of questioning (see again the work of Angus Deaton on recall periods);   they are a jumble of gross income, net income and consumption expenditure;    they use methods of valuation for consumption which vary from farm-gate prices to local market prices to national prices;    they may ignore the fact that poor people pay higher prices:   they have to buy in smaller quantities and cannot buy more when prices are low.    And these data come from national statistical offices in countries which are poor, and therefore underfunded.    Survey data leave out rich people and very poor people, who in the words of Martin Ravallion do not cooperate;   and destitute people, who are not reachable.  

This prima facie unreliability is borne out by the empirical work of Tony Atkinson and James Galbraith.   

In relation to mortality rates, it is hard to avoid the conclusion that all measures of annual income inequality as measures of relative welfare are defective.    Poor people die first, and this makes the figure for distribution among living people understate the extent of inequality of distribution to people during a period.    Rich people are still around to get the money.  

Now, I don’t expect social scientists to have perfect data, or to avoid any inferences.  

But clarity about what statistics actually measure, and in which respects social scientists have made inferences, is essential.    So is an examination of the theoretical and empirical foundations of those inferences.

Those empirical foundations have been the subject of enquiry from economists who take an approach which is detailed in its relationship to real life, such as your own.    The methods of macroeconomists, on the other hand, are extraordinarily simplistic.     The flaws appear in statistics from the United Nations;    in work originating from the World Bank and from its critics:   Xavier Sala-i-Martin, Surjit Bhalla, Foster and Szekely, John Weeks, and many others.

Oddly, in a recent book “Population Matters” edited by Nancy Birdsall, there is mention of the fact that increased child survival decreases economic growth, but not the fact that this alters the statistical relationship between economic growth and the average gain.    I find the lack of attention to this inevitable consequence of the idea in the book not only odd but also alarming.  

The problems are not confined to the work of economists

Several Millennium Goals besides the poverty goal suffer from the mortality flaw.    FAO statistics on the prevalence of undernutrition are not too good either:   the method is to take an average from the national food balance sheet, then to infer distribution of food consumption from the (unreliable) survey data on income, then to say how many people fell below a certain level in their consumption.    The problem  -  apart from the unreliability of the surveys  -  is that if food prices fall relative to the overall inflation rate, hungry people look as if they have eaten less than they really have.    Since we don’t know prices of food from these statistics, we don’t know the proportion of hungry people.  

There is a hole in my argument on survival rates which I have yet to fill

That is in relation to the reliability of survival-rate data.    The Chief of Statistical Services at the UN Statistics Division tells me that child mortality data are reliable, since they can be checked using demographic data (we can look at the number of four-year-olds against the number of one-year-olds three years earlier).     Data on life length are cheap and simple to process, and their meaning is clear.

Going back to your point about my overstating my case, I’m open to suggestions as to how to put the points both accurately and succinctly.     It seems to me clear that my observations are accurate in relation to the most influential statements from macroeconomists   -  those used by policymakers and those reported most prominently in the media.    

Yes, many economists have considered these issues in many respects.    But the theoretical issues have not been formalised into rules governing the use of inferences.    

One area where there is not enough clarity is in the concept of utility

I have written about this:   both Bernard Williams and Amartya Sen have been wrong, or at least unclear, to talk of utility as relating to states of affairs.    Utility, in the Benthamite sense, is about consequences, not static measures of welfare levels.    Some economists write of utility levels in the latter sense.    That is a fundamental confusion of two conceptual approaches.    It is related in a fundamental sense to the confusion between “the average rose” and “people on average had gains”.    

I.M.D. Little recently wrote a book in which he said that in practice these are the same thing.   I don’t know why.    Cross-country regressions may be influenced by such differences.   If poor people had more gains in life length in Cuba than in Brazil, did that influence the trend in average income, or the average in the poorest fifth?    I don’t know.    I wouldn’t use those statistics as indicators of average outcome in countries where the demographic profile is being changed by AIDS.    AIDS reduces both life expectancy and the birth rate.    

When social science develops appropriate rules  -  theoretical requirements  -  for practitioners, we will find that politicians find it harder to make unsupported statements about poverty

...In April 2001 I had a...conversation with Eric Swanson, head of the World Development Indicators project at the World Bank.     The sole subject was the mortality flaw.    In response to my concerns that the figures looked better if poor people died earlier, he said that I was interested in the micro level [which was not the case] whereas he was interested in the macro, long-term level.   In relation to poor people dying of AIDS, he said that some people had drawn an analogy with the plague, after which people were better off because the density of people on the land was lower. ....

...these possibilities for misuse of statistics exist.    I don’t think that’s good enough.     Well-known macroeconomists spend a great deal of time and money discussing statistical findings on income as if they related to proportional economic costs and benefits to poor people in the same direction.   No-one knows whether they do or not.   Perhaps the best way for me to express my points is not to say, or imply, that economists are unaware of the problems, but to call for an academic tradition to adopt discipline.   

Yours sincerely

Matt Berkley