From Growth To Equilibrium – The work of Jay W. Forrester

The number of times you read a paper which seriously affects the way you view the world is depressingly low in the development business. Papers focus on the small-scale (does buying more text books improve test scores?) instead of the big picture (why are so many people starving to death?).

Among the papers which do seriously change your view of things, very few of them are accessibly written (see this blog post on Sen) and almost none contain laugh-out-loud moments or prose well-written enough to verge on poetry.

Eric D. Beinhocker‘s “The Origin of Wealth” is a well-known example, and I have now found one more. Jay W. Forrester‘s paper Counterintuitive Behavior of Social Systems left me reeling, dazzled by beautifully expressed ideas and charged with a new enthusiasm for research just as much as Beinhocker’s book did, but is something in the order of 50 times shorter. It contains all the ideas which are currently doing the rounds at the Center for Global Development‘s supposedly state-of-the-art thinking on development and complexity (see Owen Barder‘s fantastic lecture on this subject) and yet was written in 1971. (Note that all quotes will be from this article unless I say otherwise.)

The economists among us may have heard of Forrester as the originator of the Macroeconomics 101 favourite “Beer Game“, which was an entertaining MIT-originated way of demonstrating how well-intentioned attempts to improve a business, even those made by people with full knowledge of the business and complete understanding of the policy levers at their disposal, can lead to booms, busts and the highly unpredictable and chaotic behaviour seen in financial markets. Forrester says in this paper that these unexpected outcomes are the result of the human brain’s inability to follow through the logical conclusions of its understanding, or “mental model”, of the system it is trying to fix.

“Ordinarily [people’s] assumptions about structure and internal motivations are more nearly correct than are the assumptions about the implied behavior.”

A common example of this is the arguments around the impact of giving cash to people asking for money on the streets. We’re told not to do it, for reasons which sometimes appeal to complex cause-and-effect chains, but it seems counter-intuitive and against some kind of moral urge. According to Forrester this is because social systems

“…are inherently insensitive to most policy changes that people select in an effort to alter the behavior of the system. In fact, a social system tends to draw our attention to the very points at which an attempt to intervene will fail.”

Forrester uses an example from his time working in urban dynamics at MIT. He says that building low-cost housing in depressed areas can have counter-productive effects by (temporarily) lowering housing costs, attracting inward migration without creating any new jobs. More people and no new jobs only serves to depress the area further. This and other measures (including financial aid in the form of subsidies) are concluded to

“…lie between neutral and detrimental almost irrespective of the criteria used for judgement.”

Forrester maintains that social systems are complex, and that solutions are not to be found in the same places as symptoms (c.f. Ben Ramalingam‘s leading-edge thinking on malaria). It is unsustainable for any area of a country to be fundamentally more attractive (across all possible considerations of attractiveness), and if local development programmes succeed in:

…[making] some aspects of an area more attractive than its neighbor’s, population of that area rises until other components of attractiveness are driven down far enough to again establish an equilibrium.

Instead, he makes a plea for programmes which think of the wider system holistically, taking into account so-called “general equilibrium effects” which describe how the wider system reacts to changes in a certain area.

“Programs aimed at improving the city can succeed only if they result in eventually raising the average quality of life for the country as a whole.”

This advice clearly applies to development projects too, and it’s astonishing to think how little it is still being heeded in the vast majority of today’s development literature: Angus Deaton‘s hugely entertaining diatribe “Instruments, Randomization and Learning About Development” makes similar complaints, and was written some forty years after Forrester’s.

Another piece of advice which could have been written yesterday (c.f. seemingly every British government policy since Attlee) is that:

there is a fundamental conflict between the short-term and long-term consequences of a policy change… [We should be] cautious about rushing into programs on the basis of short-term humanitarian impulses. The eventual result can be anti-humanitarian.

He goes on to extol the virtues of mathematical modelling over “contemplation, discussion, argument, and guesswork” and presents an enormous, Heath Robinson-esque model of the world which he uses to predict global catastrophe from a variety of all-to-contemporary concerns including food production limitations, a pollution crisis and natural resource limitations. Let me just restate: this paper was written in 1971.

A Heath Robinson-inspired model of the world (1971)
A Heath Robinson-inspired model of the world (Forrester, 1971)

He makes an ahead-of-his-time gloomy assessment of the outlook for international development, that there may be

“…no realistic hope for the present underdeveloped countries reaching the standard of living demonstrated by the present industrialized nations.”

Forrester then goes on to present possible ways of avoiding catastrophic population collapse cautioning that transitioning from the contemporary path of unsustainable growth to one of sustainable equilibrium will involve painful readjustments and unpopular policies (sound familiar? See this entertaining clip of Paul Krugman talking about fiscal consolidation during a recession, or any one of a million of his brilliant blog posts on the subject.) This stuff reminds me of the hugely impressive Diane Coyle, and her work on The Economics of Enough.

The paper ends with a little bit of more-optimistic poetry:

I suggest that the next frontier for human endeavour is to pioneer a better understanding of the nature of our social systems. The means are visible. The task will be no easier than the development of science and technology. For the next 30 years we can expect rapid advance in understanding the complex dynamics of our social systems. To do so will require research, the development of teaching methods and materials, and the creatino of appropriate educational programs. The research results of today will in one or two decades find their way into the secondary schools just as concepts of basic physics moved from research to general education over the past three decades.

I wonder if, over forty years on and given that almost no one is working on development as a complex system, Jay W. Forrester is disappointed that this is the one area where his predictive powers seemed to fail him so spectacularly.

The spatial distribution of human development

Let me get a quick admission out of the way, before I launch into this review of the freshly-published World Development article Using Census Data to Explore the Spatial Distribution of Human Development by IƱaki Permanyer: I don’t know much about human development indices. I know what a Gini coefficient is (in brief, a number which is close to unity if a few people have all the money, and close to zero if the money’s more-or-less evenly spread) but I haven’t read much from the long bibliography of this paper about the ins and outs of various ways of measuring human development. So, there’s a possiblity that I’m going to be over-enthusiastic about this approach compared to many other papers which I simply haven’t read.

But I must say, I’m hugely excited by the methods proposed by Permanyer in this paper.

In summary, he discusses a method of using very simple questions in the Mexican census to proxy for such unknowables as health, education and standard of living. He goes on to show that these proxies perform well in comparison to other, more difficult to obtain, metrics and that they allow a comparison between municipalities in human development terms.

The really nice thing about his method is that it doesn’t at any point rely on self-reported income or health; metrics which are famously inaccurately reported.

Instead he constructs an asset index to proxy for material welfare and a simple child mortality stat to proxy for health. These are incredibly simple to gather, and are not subject to misreporting.

A whole page of the paper is dedicated to the justification of using an asset index instead of income, included the hand-wringing worry that

…asset indices have been criticized because they might not correctly capture differences between urban and rural areas. Since many assets are cheaper, more easily available and more desirable in urban areas, urban households might appear to be wealthier than their rural counterparts.

This seems like a strange concern to me, since the whole point of using an asset index is to get away from traditional money-based definitions of wealth. As Amartya Sen (who seems oddly under-cited in this paper) would doubtless argue: if the assets in question add to the capabilities of the respondent, and the respondent happens to live in an area where the asset is cheaply available, then surely the respondent is indeed wealthier. This strong argument seems like an obvious omission to me.

The paper goes on to report the results of calculating these new human development indices using census data from 1990, 2000 and 2010 (data which the author needed “a special permit” to access) and the results are truly gripping. He shows a couple of fantastic choropleth maps (something which I’ve never seen in an economics paper!) and is able to calculate, using the fabulously titled kernel estimation, the change in distribution of these indices throught Mexico across each of the three census periods. This is where the real magic of the paper lies: the distribution graphs are wonderfully informative (they summarise tens of millions of data points into 6 curves!) and tell a powerful story of the success, and origins of, Mexico’s growth programmes over the last 20 years.

Finally, let me add that the paper is lucidly and well written (despite some English oddities which could have easily been ironed out at copy-editing: Permanyer insists on continually using the clunky construction that a method “…allows to…” do something cool) and that there is a nice example of disarming honesty.

Everyone in research, to some extent, bases their methodological decisions on what data are available but Permanyer makes this explicit:

The choice of municipality as unit of analysis has been basically determined by data constraints.

Great stuff, and a cracking good read the paper is too.

Sen is everywhere

I’m sure I’m not the only one to have had a hard time reading Amartya Sen’s Development as Freedom* on the tube or in a cafe. It’s serious reading and requires a suitably contemplative atmosphere to avoid the dreaded “I’ve read this sentence three times now, and I still haven’t taken anything in” effect. It’s very precisely written, with little wasted (although to be fair, a helpful amount of repetition of the main ideas) and feels a little like reading maths. (Or, perhaps, like reading philosophy which, apart from the obvious, I’ve never seriously attempted.)

Anyway, it cheered me somewhat to see an article in the New Statesman (11-17th January 2013) by Tim Wigmore, apparently nowhere to be found online, entitled “Give a little, but give it well”. The article is a plea for evidence-based targeted of money to charities and describes the work of givingwhatwecan.org who claim to be able to rank charities based on their ability to “provide the greatest return in terms of quality-adjusted life years” for each quid donated.

Two things strike me as interesting about this:

  1. Ranking charities in this way seems to me to be a direct application of Sen’s plea for a more rounded “capability approach” to assessing development: rather than focusing on how much incomes are increased by a given development intervention, he proposes we look instead at which things the people involved are able to do which they couldn’t do before. (Only things which the people themselves “have reason to value” are to be included in the list. These are things which Sen refers to as “substantive freedoms”.) It’s encouraging to see practical applications of Sen’s philosophy making it into a popular weekly magazine. It also makes me think it was worth the effort to read the book in the first place!
  2. Given the famous difficulty in knowing what NGOs are doing with their donation money, it seems incredible that one organisation can not only find this stuff out, but find it out in sufficient detail to be able to compare across organisations in terms of an effectiveness metric. Amazing.

So, with this in mind, I’m off to find out how “Giving What We Can” are able to do what Wigmore says They Can, and also off to continue to enjoy/battle through Sen’s heavy-hitting classic, the effects of which are clearly still being felt throughout the development research community.

* Sen, Amartya 1999 – Development as Freedom, Anchor Books

Behold! I show you a mystery

I’ve been looking into whether Google’s “search result count” facility can help direct the focus of my research a little bit, by recording the result count for a series of search strings relating to different countries, in different years, and for different development project types (e.g. education, health etc.)

Although the result count is only an estimate (and famously a rather poor one) I think that comparing result counts against each other should be an at-least-reasonable heuristic. I assume that although they’re wrong, they’re not randomly so, and hence a higher result count for one of a set of structurally similar search terms should say something about the number of pages found.

The other assumption that lies behind using result counts in this way is that the amount of ‘stuff’ on the internet is a good measure of how much the English-speaking developed world is interested in a given topic. I assume that if “Kenya development project water” comes up with more results than “Kenya development project malaria” then more English-speaking internet users are ‘interested’ in water projects than in malaria projects. I then make the leap of faith that this implies these projects are happening more. Debatable? Most certainly. I’d be interested to try and defend this against a well-informed doubter. Comments below!

Since this assumption, if true, would be more accurate post internet-era, I’ve restricted my searching to the years 2000 to 2012. I’ve both included the year in the search term, and restricted the search results to only those pages from that year.
Google Custom Range
Methodological quibbles (or more) aside, I was impatient to start looking at the results of this “Google-harvest” and have analysed the numbers for a subset of African countries (namely Burkina Faso, Congo, Egypt, Eritrea, Gambia, Ghana, Guinea, Guinea-Bissau, Liberia, Mali, Mauritania, Morocco, Mozambique, Sierra Leone and Tunisia (I realise that any search results for Guinea-Bissau will show up in those for Guinea as well. I also realise that “Congo” is two countries. I’ll gloss over these details for now).

Although this represents only a small number of all the countries in Africa, the results are already worth commenting on. Here’s a graph showing how the total result count across all project types was divided between the project types.

Graph of Google result counts
Graph showing the time trends for Google search results across 15 African countries for different types of development project. Search terms were, for example, “Niger 2001 Secondary School project development aid”

The sharp-eyed amongst my readers will have spotted something odd about this graph. In 2006, there are huge spikes education projects (18.3% up to 20.8%) and agriculture projects (10% to 11.8%) at the apparent expense of water, AIDS and malaria projects. So the mystery is this: what happened in 2006 that led to a huge (but temporary) increase in interest in education and agriculture projects, at the expense of interest in health projects? And will this trend still be visible once all the results are in? Only time will tell….

p.s. apologies to G.F. Handel for the title of this post.
Interesting econometrics to follow, this is just the before-party.

Wanted: people to exploit

The first rule of academia: never talk to anyone from the ‘real world’ about your research.

I was reminded of this rule yesterday when I spoke to my flatmate about what I’m trying to do with complexity science-style modelling and development. I told him of the need to stop seeing development as the search for a missing ingredient, and of how being a developed economy doesn’t imply that you know what it takes to become a developed economy any more than being healthy implies you know anything about medicine. In fact, the opposite could perhaps be argued: it’s through my illnesses that I’ve learned such physiology as I know, not through my wellnesses.

I told him of the prospect of investigating a more subtle model of society which recognised that institutions, policy, wealth and technology are all interacting systems and that sudden and radical changes in state (such as that which occured during the industrial revolution) are well understood in chaos theory, but terribly predicted by linear regression. I gesticulated wildly and my cheeks grew pink with enthusiasm.

He made a point, though, which entirely deflated my new-found sense of purpose in studying why some countries are rich, safe and just and others are poor, dangerous and corrupt. Surely, he said, there can be no study of development which doesn’t account for the fact that we are where we are because our ancestors successfully stole labour and resources from parts of the world they colonised for four hundred years. In our quest to make developing countries ‘more like us’, are we going to prescribe four centuries of stealing labour and resources from us? He thought not.

The problem with using the past of our development to study the future of others’ development is that we want to improve their lot without affecting our lot in any way at all. That’s what makes development hard, and it goes beyond the usual ‘world resource constraint’ argument which says that the world doesn’t have enough stuff for everyone to live like an American. I’ve never really bought into that argument. This new argument is more pernicious: we’re not going to run out of stuff for developed countries to consume, we’re going to run out of people for developing countries to exploit.

My first choropleth

I’ve managed to get some data onto my PostGIS database, and have got my first visualisation result.

Since the aid dataset is large (c. 1,000,000, source: aiddata.org, the dataset is called ‘aid data 2 short’) I’ve summed over donors and within years, to leave me with a total aid received (or rather, promised, since the dataset has amounts pledged, not amounts donated) per recipient country per year. This set is a far more manageable size.

Having added population data from the UN Department for Economic and Social Affairs (World Population Prospects, the 2010 revision) I’m able to draw a choropleth map of the world with the colour of a country representing amount of aid committed to that country in a the current year (the example uses 2000 as a randomly chosen benchmark). It’s work which is very similar to this.

Nice, isn’t it?

levels of aid choropleth
my first choropleth (QGIS)

Giving aid to non-existent countries

In the big, fat dataset I’m using to study what affects the global flow of development aid, I’ve come across a problem which researchers must have dealt with a million times from scratch, and never published the results of how they got around it.

The data set has donor/recipient pairs for over a million promised amounts of development aid, along with a date and a reason for giving the aid.

Sounds simple, but the problem comes from data relating to countries which, according to my separate dataset of country boundaries (the wonderful CShapes) didn’t exist at the time the aid was promised.

For example, aid was given by the World Bank to Botswana in 1965 even though that country only achieved independence from Great Britain in 1966. My world boundaries dataset reflects this accurately giving a “start date” for Botswana of 1966. Before this date, the country was simple considered not to have existed and a map drawn of all the countries as of 1965 just has a gap where Botswana should be.

This is a pretty serious problem in the field of development aid, since many of the relevant countries are former colonies and aid was often given to those countries before they were independent from their colonial masters.

This problem must have been solved a million times before. But by whom?