November 1, 2011 § Leave a comment

The Spirit Level shows that income inequality is significantly negatively correlated with the percent of the population who agree that most people can be trusted. This holds across countries and across US states: Mississippi and Alabama, the states with lowest trust, are unequal, while North Dakota and Montana, the states with highest trust, are relatively equal. There’s something to this. There probably is a fairly simple set of causal pathways that connect the variables “inequality” and “trust”. But are these cause-and-effect type pathways or confounding-type pathways?

The book asks: “But does inequality create low levels of trust, or does mistrust create inequality?” This is a false dichotomy. Variables may have a common cause and no effect on each other. To get a more realistic answer, we turn to the literature they cite.

Rothstein and Uslaner:

First, we show by a cross-national statistical analysis that inequality is a key factor in shaping generalized trust, but that there is no direct effect of trust on inequality (the causal direction starts with inequality).
That’s a strong statement. What does this cross-national statistical analysis consist of? A bunch of structural equations and a probit model, which I shall be generous and leave alone. The much stronger argument about causality is qualitative. There’s a short but careful retelling of 150 years of Swedish history that notes that country started off more equal than many, and that their civil service has been a trusted institution since its inception.
A detailed account of the historical development in the Swedish case indicates that it was the impartiality of the government institutions, especially those handling policies related to the labor market and social policy, that made the development of an “historical compromise” between labor and capital possible. This compromise was to a large extent built on “a spirit of trust”  that developed into the well-known “Swedish Model” that came to mark the Swedish society after 1936.
I can’t see how the argument is particularly strong evidence for the causal priority of trust to equality. What it seems to show is that there is feedback, that trust and equality evolved together. To show that one direction dominates the other requires faith in structural equations that I don’t have. To think about this properly, you need to consider interventions, which is tricky: how do you change trust? But that’s a question you need to answer in any case.
What’s frustrating about The Spirit Level is that it doesn’t delve deeply into any particular issue. The questions we really care about are “if we change X, what’s going to happen to Y?” Such questions can rarely be answered definitively. But the book displays so little curiosity toward working in the direction of answers, toward working out what’s causal and what isn’t. I know I’m not the target audience, but it could have been so much more convincing with even a hint of rigor.

Inequality cheat sheet

October 20, 2011 § Leave a comment

Non-quantitative post below the jump. « Read the rest of this entry »

Pairing: USA vs Japan FIGHT!

October 16, 2011 § Leave a comment

Let’s say you want to study the relationship between economic inequality and attitude toward the self, with social hierarchy perhaps a common cause. “Attitude toward the self” isn’t well-defined and hard to measure, and there’s little multinational data. What you could do is try to directly compare two countries. You could pick the USA, a highly economically unequal society, and Japan, a relatively economically equal society (forget that Japan is highly hierarchical for the moment).

What else could result in American and Japanese attitudes differing?

  • The US is much more multiracial than Japan
  • The US is mostly Christian, Japan Buddhist and Shinto
  • Japan has universal health insurance
  • Japan lost the Second World War
  • Evangelion was better in the original language

The most fundamental problem is sample size: even if you had identical twin countries, one pair ain’t enough. If you want to draw a strong conclusion from one pair, you need some strong qualitative arguments. If you want strong conclusions from quantitative data, you need quite a bit of data, and strong assumptions on top of that.

Rant: Poverty is a problem

October 12, 2011 § Leave a comment

Warning: Non-numeric post.

The view that social problems are caused directly by poor material conditions such as bad housing, poor diets, lack of educational opportunities and so on implies that richer developed societies would do better than others.

Ignoring the weaselly “directly”, this still isn’t true. You can have poor material conditions in the richest of countries, as anyone who’s spent a day in Oakland can tell you. Of course this is related to inequality, but the thesis of The Spirit Level is that inequality in itself leads to problems — that doubling everybody’s income would make negligible difference.

However, surveys of the 12.6 per cent of Americans living below the federal poverty line (an absolute income level rather than a relative standard such as half the average income) show that 80 per cent of them have air-conditioning, almost 75 per cent own at least one car or truck and around 33 per cent have a computer, a dishwasher or a second car.

For a lot of people, a car is necessary to do whatever job they have, and for any kind of quality of life. They are a lot of American cities in which public transport isn’t practical outside the main arteries. Saying that vehicle ownership is “usually a reflection of the strength of thir desire to live up to the prevailing standards” is vaguely insulting. Meanwhile, a third have either a computer, a dishwasher or a second car? The lucky duckies.

Although modern market democracies fall into neither of those extreme, it is reasonable to assume that there are differences in how hierarchical they are. We believe that this is what income inequality is measuring… The first pointer is that only the health and social problems which have strong social class gradients — becoming more common further down the social hierarchy — are more common in more unequal societies.

Or it could just be absolute or relative poverty. Over and above that, a bunch of things that are correlated are not the same thing.

The other pointer which suggests that income inequality reflects how hierarchical societies are, became clear when we reviewed nearly 170 academic papers reporting different pieces of research on the relationship between income inequality and health… While there was overwhelming evidence that inequality was related to health when both were measured in large areas (regions, states or whole countries), the findings were much more mixed when inequality was measured in small local areas.

Of course the larger the scale you consider, the more inequality you see. To the extent to which this is evidence for anything, it seems to be evidence against the authors’ hypothesis. If being the biggest fish in a small pond doesn’t help you, surely that means the effect of hierarchy is less than that of the material circumstances of the community you live in?

What problems might inequality cause? What else might cause them?

October 11, 2011 § Leave a comment

The “Index of Health and Social Problems” in The Spirit Level has been rightly derided as silly. Even if you believe that inequality has an effect on everything from imprisonment rates to mental illness, there’s no reason to believe that inequality works in the same way for all variables. The authors make the case that inequality harms everyone, and that’s perhaps plausible for mental illness. But are the rich in a unequal society more likely to be imprisoned than the average person in an equal society? Evidently not. Within a country, your position in the distribution is hugely important. Differences between countries might matter as well, but it’s a lot harder to attribute these to a particular factor or set of factors.

A major confounding variable would be level of government spending. Now, one could argue that government spending itself reduces inequality not only by direct transfers, but implicitly, since the benefits of, say, a public health service are spread more or less evenly, while the costs are not if taxes are progressive. But that’s not the argument The Spirit Level is trying to make — it’s trying to say that income inequality in itself causes these things. The problem is that government spending and income inequality are highly correlated, and from a set of 20-something countries, you’re not going to have enough data to get small standard errors for the effect of each. And that’s without considering any of the dozens of other variables that could have notable main effects.

Looking at data below the country level may help. But the set of causal relationships that hold at state level may not be the same as the set that hold at the country level. Now, if you had data for subregions of a variety of countries, you could build a Big Bad Multilevel Model of Doom. It would be even better if you had individual-level data. But if you don’t, then although additional data is not to be sneezed at, it doesn’t resolve the fundamental problems.

Absolute and relative income, or inequality?

October 6, 2011 § Leave a comment

Saying that average life expectancy isn’t related to average income is a strong statement, but it isn’t the same thing as saying that personal life expectancy isn’t related to personal income. Going to the individual level:

  1. Personal life expectancy in rich countries might be related to absolute income, but not relative income.
  2. Personal life expectancy in rich countries might be related to relative income, but not absolute income.
  3. Personal life expectancy in rich countries might be related to absolute and relative income.
If average life expectancy isn’t related to average income, then you really have to fudge numbers to get 1.  to be true. Stated as bluntly as it is, it’s a bit hard to believe 2. 3. is the easiest to believe.
If we’re going to talk about inequality, though, we need to clarify whether we’re talking about inequality as part of the above, or an national inequality effect on top of the above.

Aside: “We are the first generation to have to find new answers to the question of how we can make further improvements to the real quality of human life. What should we turn to if not to economic growth?” (Spirit Level, p. 10) I would have thought such new answers were at least as necessary as now for all of human existence prior to about 1750. Put another way, the constraints to growth provided by the environment now seem less stringent than those provided by feudalism. (Yes, yes, climate change; that’s another Big Project I’ll get to some other year.)

Health care spending

October 6, 2011 § Leave a comment

What does money get you in terms of health? If you look at a graph of life expectancy against per capita GDP — even if income is on a log scale — then the relationship gets weaker as you get richer. On the other, a more relevant x-variable might be health care spending:

This isn’t a surprising graph. The relationship is mostly increasing (with the US an outlier), but returns diminish. Even within the top group, though, three years of life expectancy is practically important. An interesting question is: what explains the differences within that top group? It’s unlikely we can prove causation, but with substantial background knowledge and good data, we should be able to formulate a reasonable working theory.

Where Am I?

You are currently browsing the inequality category at "But it's under .05!".