Wednesday, September 28, 2011

Simpson's Paradox in Texas

I am not a fan of Rick Perry. Unlike President Obama, Perry is not smart. Perry is the worst of both worlds: he unnecessarily alienates leftists and moderates by pushing all their cultural identification buttons, at the same time as he engages in ultra-liberal open border policies with taxpayer’s money.

The media might refer to Perry's policy of using taxpayer money to pay for the college of illegal immigrants as "moderate". But a policy which is opposed by 86 percent of the public even in a rather liberal state such as Wisconsin is hardly "moderate". Compared to the views of voters, Perry's policy is extreme-left. His defense that without having their college education paid for by taxpayers illegal immigrants would live on the "government dole" is unconvincing, since illegal immigrants are not legally entitled to welfare.

Though I feel no need to defend Perry, let me defend the state of Texas itself, which is currently under attack because of Perry's candidacy.

Texas has had rapid job growth. Leftists such as Paul Krugman argue that this is irrelevant, because we have to adjust for demography. Texas also has fast population growth (apparently supply does matter even during recessions). Fair enough, I wrote about the role of population growth for Texas a while ago.

At the same time, left-wing economists such as former U.S. secretary of labor Robert Reich attack Texas for being a state where workers are paid poorly.

“While Texas leads the nation in job growth, a majority of Texas' workforce is paid hourly wages rather than salaries. And the median hourly wage there is $11.20, compared with the national median of $12.50 an hour.”

Reich writes that the Texas model is one of “low wages“ and “lower incomes”, and therefore not right for the U.S.

However, if you want to make demographic adjustment for job growth, you need to make demographic adjustment for wages. Yes, Texas has lower average wages than the U.S. But in fact, Texas Whites, Texas Hispanics and Texas African Americans each earn higher wages than Whites, Hispanics and African Americans in the rest of the country.


The data is from 2011, hourly wages, from the Current Population Survey of the Bureau of Labor Statistics. I look at those ages 20-65. Indeed, I find that average for all groups is slightly lower in Texas than the rest of the country.

But as you see, each demographic groups earns higher hourly wages in Texas. The same is true for mean and median total salary, Texas is on average lower, but Whites, Blacks and Hispanics are each higher in Texas than the other States.

Each group earns more in Texas, yet the average is lower, because the state has a higher share of low-earning groups.

This is an instance of what statisticians refer to as Simpson's Paradox.

The lower wage rates in Texas are entirely due to demographic composition. Texas is 45% white, compared to 64% nationally, and Texas is 38% Hispanic, compared to 16% nationally.

Demographically adjusted, It would be more accurate to say that Texas is characterized by a high wage (and high ethnic diversity) model.

Robert Reich also complains “Texas schools rank 44th in the nation in per-pupil spending.”

Sure, but again, each ethnic group has higher high school test scores in Reading+Math in Texas than the average for the U.S. The data is from the Department of Education NAEP. The graph is truncated at 200 points, to make it easy to read.


The overall average is identical in Texas and the Nation as a whole. Yet all groups do better in Texas than the Nation as a whole. The only reason that the average is not higher in Texas is that the state has a higher share of lower than average performing Hispanics. Even though Hispanics in Texas outperform Hispanics in the rest of the country, their higher share of the overall population depresses the Texas mean and dominates the result.

The Simpson paradox in wages and the near-Simpson's paradox in test scores highlight the increasing importance of demographic composition for social outcomes.

Saturday, September 24, 2011

More on Bryan Caplan, ethnic diversity and the size of government.

Bryan Caplan has now responded to my post arguing that open borders will expand the welfare state. The discussion continues below:

1. Caplan almost entirely relies on Alesina, Glaser and Sacerdote (2001). The paper shows that globally, ethnic fractionalization correlates negatively with social expenditure. I have however shown that this result does not hold for rich OECD countries.

The Alesina et al. paper paper uses “Social expenditure”, a subcategory of the welfare state, which includes transfers but not things like public health care. Social expenditure only constitutes about 40% of total government spending in the U.S. The correct measure of the size of the welfare state is total expenditure, not just one subcategory of spending. I therefore used total public expenditure in my last post. But let’s not let this details distract us.

It turns out that for rich OECD countries, there is no statistically significant relationship with the degree of ethnic fractionalization even if we use OECD “social expenditure” data just as Alesina et al. did. I look at 1990-2000, to correspond with the years the fragmentation data was calculated.


It turns out Alesinia et al. relationship between diversity and a small welfare state is entirely driven by third world countries. But Caplan is discussing immigration to the west; not how tribal fragmentation impacts policy in Africa.

Caplan argues: “instead of throwing away most of the data and variation, it's better to keep the data and add control variables.“

Incorrect. Since we are specifically discussing immigration to the west, we need to throw away the variation that comes from poor African and Latin American countries. Adding dummies does not solve this (though a combination of dummies and interaction variables might have). Even with a regional dummy, the large number of African and Latin countries can drive the results, even when there is no relationship within rich OECD countries.

When I restrict the analyses to rich nations, it turns out that the relationship between diversity and small government does not hold. The results in the paper Caplan builds his entire case on are driven by how fragmentation effects third world countries.

It is logically flawed to have African data generate a relationship, and proceed to argue (as Caplan does) that this proves that more diversity will reduce the size of government in the United States, even though we know the relationship doesn’t hold in rich countries.

Alesina et al. have another misleading graph, which Caplan also relies heavily on. This shows that in the U.S, states with a high percentage of African Americans have lower welfare benefits per welfare recipient. But expenditure is not measured by benefits per recipient; it is measured by the benefit per recipient times the number of recipients. States with more minorities have more welfare recipients. You have to look at both effects, which Alesinas paper neglects to do. This is important, as empirically, it appears that the second effect dominates the first, so that the size of government is bigger in more diverse states (though it is not unproblematic to compare U.S states in this way).

I am glad that Caplan now acknowledges that we need to take into account the effect of demography on the number of welfare recipients, not only welfare per recipient. He has written about the Alesina paper numerous times, and never once pointed out that Alesina graph is only half the story. I might have pointed out this problem to him and his readers if I were not banned from his blog.

Caplan also claims that “AGS's graphs and regressions are better”.

Why? I am using the exact same ethnic fractionalization data that they use (since I downloaded it from their appendix) and standard social expenditure data from the OECD just like they did. The main difference is that I look at rich OECD democracies, since we are discussing policy in those rich countries, and since there are many obvious reasons to believe the mechanisms are not the same as in the third world.

What I do is very similar, so it is hard to escape the conclusion that Caplans objection are not methodological. He just prefers the old result because that produces his preferred ideological outcome.

Aside from all of this, I have to stress that the causal interpretation of the Alesina et. al story which Caplan has made repeatedly is problematic, since the authors only report correlations. There are plenty of differences between the United States and Scandinavia besides how ethnically fractured they are. There are plenty of differences between New England and the South besides the share of African Americans.

2. I write that if ethnic diversity increases poverty and social problems, the left gets more votes, since this increases the demand for government policies to fix the problems. Caplan writes that this is “theoretically confused”. He further writes “the entire political spectrum is indeed irrational, and habitually demands government intervention for no good reason”. (Caplan thinks there is no good reason because according to him poverty in rich countries is merely relative).

I see two problems with this assertion. First, I believe that Caplan is projecting his own atypical preferences to everyone else. People on average have an aversion to relative poverty. This is a preference, not something you can simply dismiss as “irrational”.

Second, more importantly, he is mixing a positive and normative debate. Even if you think that the tendency of voters to demand public intervention as a result of larger social problems of relative poverty is “irrational” or otherwise immoral or confused, this tendency is an objective fact of political reality, not something that you can assume away because you disagree with it.

3. Caplan writes:

“Sanandaji also ignores the possibility that immigrants vote “overwhelmingly for the left” primarily because right-wing parties treat them with such hostility. “

There is no evidence for this claim.

Hispanics did not vote for consistent amnesty supporter John McCain. By no reasonable definition does, say, the Swedish parties of the centre-right or the Republican party of 2008 show “hostility” to immigrants, unless you define opposition to illegal immigration (shared by two thirds of Americans) and to the welfare state as inter-ethnic "hostility". Indeed, that the pro-immigration Swedish centre-right is just as marginalized with third-world immigrants, as is the American right is a rather telling fact.

If you ask Hispanics themselves, it turns out that they are far to the left of American whites regarding economic policy. As I have written, “In the United States, where while only 35% of non-Hispanic whites prefer higher taxes in return for more government services, the figure is 65% for first generation Hispanic immigrants, and 66% for second generation Hispanics.”

This question has nothing to do with the Republican immigration policies (or, say, policing or drug policy), instead directly asking about attitudes regarding fiscal policies. The gap between Hispanics and whites regarding issues relating to the size of government is immense.

When polled, Hispanics are closer to the Democrats regarding public health care and public education than on questions regarding immigration. If Hispanic support for Democrats was "primarily" driven by immigration policies (or some other proxy for inter-ethnic hostility), we would not observe this pattern.

It should also be noted that the countries of origin of Hispanic immigrants, e.g Mexico, Argentina, El Salvador, Peru, Brazil, Bolivia and Venezuela have a long tradition of populism and left wing economic policy. Is this because of GOP “hostility” to illegal immigration? Hardly.

Immigration is a subcomponent of ethnic identity politics, where the left has an inherent advantage. If Hispanics are poorer than whites with worse social outcomes (currently they earn about half much as of whites), they will observe and resent this, and they will believe that the free market is unfair to them.

In addition, the prospect of enacting resource transfers from the majority population, in the form of cash transfers, government services and quotas and preferences, will be a fixed political advantage of the left, as a classically liberal right can never embrace such measures as enthusiastically as the left.

Poor minorities are loyal supporters of the left almost everywhere, including countries such as Sweden where immigration policy has frequently been more generous during right-of-center governments. Between 2000 and 2006, when the Social Democrats were in power, Sweden took an average of 60.000 immigrants per year. After the right took power in 2006, average immigration increased to 100.000 per year. Yet in the 2010 election 77% pf non-European immigrants voted for the left.

In 1986 Republican President Ronald Reagan enacted amnesty to illegal aliens. In the following presidential election, 70% of Hispanics votes for Democrat Dukakis (who got 40% of the white vote). I doubt that Ronald Reagan was somehow showing “hostility” to the millions of illegal Hispanic immigrants he granted amnesty to.

Bryan Caplan likes free markets, and he likes open borders. So he has constructed a story wherein free immigration of unskilled labor can be easily combined with a smaller welfare state. It would be more honest, and more interesting, if Caplan acknowledged that there is no guarantee that the universe is designed in such a way to remove tradeoff for his ideological principles.

Sunday, September 18, 2011

Bryan Caplan is wrong about open borders and the size of government

Bryan Caplan once again claims that open borders would reduce the welfare state. He uses two misleading pieces of evidence, based on Alesina et al. (2001):

1. First, internationally, more ethnically fragmented countries have less social spending as a share of GDP.

This compares underdeveloped third world countries like Peru and Guatemala with Western Europe. Ethnically fragmented societies tend to be poorer and less well organized, which makes a large welfare state hard to finance. But we are discussing immigration to the west, not to Guatemala.

Alesina et al. (2001) further use social spending, a narrow component of government spending which is even more biased towards rich European countries.

I use the total size of government (taxes as a share of GDP 1990-2001), from the OECD, and only for rich OECD countries, as well as fractionalization data from Alesina et al. (2001). The period is chosen to correspond with the years when ethnic fractionalization was calculated.

As you see, there is no statistically significant relationship between ethnic fractionation and the size of government among advanced OECD countries.


2. The second graph of U.S states which Caplan cites is equally flawed. This uses the generosity of welfare expenditure per recipient, controlling for state income.

As I pointed out in my earlier post, generosity per welfare recipient is an improper measure to evaluate diversity and the size of government. More poor minorities mechanically increases the number of those on welfare. Likely, the state will respond by making welfare less generous per recipient. The total cost may however still go up, since there are now simply more people on welfare. New Hampshire can afford to be more generous than Maryland.

This is why I use aggregate numbers, which suggest that American states with more minorities spend more in total. After all, we are discussing the effect of open borders on the overall size of government, which is a aggregate number. Even if open border forces us to be less generous per welfare recipient, increasing the number of poor people tends to expand government.

Caplan doesn't counter any of my other arguments. In particular, let me stress once again that even if open borders makes the majority population more anti-government, after a while their preferences will not matter, since they will inevitably become a minority of voters.

One more point. Looking at third world countries is also misleading for another more subtle reason. Poor countries tend not to have yet developed liberalism, an important mediating factor between ethnic diversity and the size of government.

Non-liberal voters probably become less generous if their taxes are going to poor people from other ethnic groups. This is especially true in poor countries such as Africa or Latin America. However, liberal western voters likely become more supporting of government intervention if poverty is concentrated among disadvantaged racial minorities.

This is because they view a society where the white majority is rich but where blacks and Hispanics are poor as racist and discriminatory, in addition to merely having an uneven distribution of income (I think this view makes some sense). Poverty which is spread out evenly between races is by contrast perceived as more likely to be the result of a reasonably fair market process, and therefore less morally offensive.

I know I can't persuade Bryan Caplan of this, because in my experience he is too ideologically dogmatic regarding open borders to listen to anything other than perhaps hard data. For instance, I was banned from Econlog after I debated the fiscal cost of unskilled immigration.

But the rest of us should simply read the arguments of the American left to test my claim. What are the overwhelmingly white Hufftington Post, New York Times and Daily Kos more bothered by, poverty among rural whites (their co-ethnics) or poverty among African Americans and Hispanics?

Or look at how movies and the media portrays poor Appalachian whites ("rednecks", "white trash") compared to how poor minorities are depicted.

For the same reason, I think using a definition of ethnic fractionation of different white groups (such as different language groups in Switzerland) which Alesina et al. use is misleading. We should calculate the share of racial minorities, since liberals view black or immigrant poverty more problematic than white poverty.

Racism carries a bigger moral punch in our society than the unequal distribution of earnings among whites.
Google Analytics Alternative