Subj: Logic and inference in economics
Date: 9 January 2003
File: D:\Current articles\Logic and inference in economics.doc (72704 bytes) DL
Time (32000 bps): < 1 minute
Dear Professor Deaton
Logic and inference in economics: the implications of PPP flaws, and
demography, for analysis of economic gains
I am writing to you because of all the development economists who write about
theoretical issues, I have found the most comprehensive treatment in your
For the past two-and-a-half years I have been researching the inferential
procedures behind the most influential recent pieces of development research.
What I am doing is to look at the implications of all the major flaws, in order
to present an accurate picture of the limits to current knowledge in those
areas. I am aware that insiders know of many flaws in development economics,
while most outsiders do not.
This is a spare-time activity which has turned into a book-length document: my
main document is now 44,000 words and rising. My own training includes
philosophy and experimental psychology.
On PPP, I've pointed out to Thomas Pogge and Howard Nye that the flaws
invalidate not only the poverty counts but also the results of cross-country
regression analyses on poverty. I am quite keen to publish this fairly soon,
while there is attention on PPP. The document I attach presents this argument
in relation to analyses of per capita income and poverty, along with the
argument that the failure to exclude demographic factors also invalidates the
I've spoken to many people about some of the flaws I describe, including Frances
Stewart, Ravi Kanbur, John Broome, Dan Hausman, Gary King, Uskali Maki, and
officials in government, UN organisations and the World Bank. People often
mention some of the relevant flaws when arguing about conclusions, but not all
My work is intended for public consumption; and also to stimulate discussion
as to whether the theory of welfare economics needs revision. In my opinion
the theoretical problems which you have yourself raised about per capita income,
death rates, and PPP are issues for economic theory, not just for the analysis
of poverty. If economic theory cannot cope with a significant proportion of
the world's population, then it is inadequate. In any case, many of the
issues are not primarily about poverty, but about inferences - after all,
demographic change influences per capita statistics in rich countries as well as
poor. Also, the theory of poverty statistics today applies to the history of
There's one oddity which keeps coming back to me, and which I haven't yet
thought about very deeply: the cutoff point between income and
consumption-expenditure measurement. Where these two different measures are
used, either in combination or in comparison, or both, I think there may be some
strange statistical effects.
Another is the fact that there is a lower cutoff point to consumption. How can
we design meaningful measures of inequality - or even of aggregate welfare
levels - bearing this in mind? Perhaps by taking as zero the lower level of
possible consumption (something like what someone could just survive on for a
whole year), not zero consumption.
But then all these things are subject to both measurement error and the
inferential problems of dealing with aggregates. The more I know about
problems with population averages, the more I think that we can easily judge
someone's seriousness about helping/enabling poor people by their keenness to
raise and measure survival rates.
I'm also doing a review of the Millennium Goals, their targets and indicators,
from the point of view of their coherence and possible logical relationships
between them: e.g. if child mortality statistics are bad in a country, would
we expect the proportion of poor people to fall faster?
In my larger document I have various speculative ideas, such as that extended
lives of the poor have caused lower GNI increases in Cuba/Kerala/Sri Lanka, and
the perhaps more easily inferred one that "missing women" in Asia have made GNI
go up faster.
There are very few people who successfully separate sense from nonsense in
development economics. When I've mentioned to economists (prominent
professors) the problem of death rates making the figures look better, they have
generally gone blank. There are, in fact, ways to increase the chances that
social scientists make true statements about the progress of real people, even
where death rates are high.
So here is my short version, which I would be very grateful for any comments on.
It is harsh, but I can't see how I can tell the truth without stating it
clearly. I hope that I have not made gross errors. Various potential
titles include "Sense and nonsense in economics"; "Science and speculation in
economics" and "Logico-mathematical structure of arguments from economic
statistics to income gains and losses". In this version, I would ask you to
skip over if necessary the rather grandiose (but I think necessarily true)
points about Theil and game theory. The points about fraud have to be made by
someone, and I wondered whether to leave them out in this initial approach to
you. There is a distinction between work which is competent and open to
interpretation, and statements which are untrue from any reasonable standpoint.
How I can phrase this is something I need to think about. Thomas Pogge,
and Ray Thomas (secretary of the Official Statistics section of the Royal
Statistical Society) have read versions of my longer document.