I’ve repeatedly argued that libertarians exaggerate the political externalities ofimmigration.  One of my arguments is that immigration increases diversity, which undermines solidarity, which mutes public support for the welfare state.  A well-known 2001 paper by Alesina, Glaeser, and Sacerdote (henceforth AGS) presents evidence that, on net, racial fragmentation makes the welfare state smaller.  AGS is heavily cited, but the best of my knowledge, no one has ever published a critique in an academic journal or book.  The most thoughtful response to AGS remains a blog post by Tino Sanandaji, Ph.D. student in Public Policy at the University of Chicago.  Consider this a belated point-by-point reply.

Sanandaji begins with a correct explanation of the AGS mechanism:

[E]conomists believe that solidarity is diminished in ethnically heterogeneous societies. According to this theory voters care more about people with the same race and ethnicity as themselves, and are less willing to help the unfortunate if they have a different skin color…

Some libertarians want to rely on this mechanism to tear down the welfare state through open borders and the ethnic tensions they believe that migration will cause.

Sanandaji’s first reaction:

[I]f that is the price of limiting the welfare state… I would oppose it… I understand that some free-marketers have turned against the very notion of “solidarity”, because the left has exploited the term so much. However this should not let us lose sight of the fact that solidarity and national cohesiveness are social goods, not something that we should want to destroy through an immigration shock doctrine.

This is a straw man.  The claim isn’t that open borders will “destroy” solidarity or the welfare state, but merely that open borders will undermine both.  And while free-marketers may well agree that some degree of solidarity is good, it’s also hard for free-marketers to deny that current levels of solidarity are excessive.  Solidarity stands in the way of free-market reforms in pensions, education, health care, taxation, agricultural policy, and much more.

Sanandaji then turns to the heart of his critique:

Ethnic diversity overall tends to expand the welfare state, not reduce it. While the research only focuses on one effect of unskilled immigration (reduced fellowship), there are at least three effects that go the other way. Here are the main effects of increasing the share of low income minorities:

1. Solidarity is diminished and social ties are weakened, so that the majority population becomes less willing to pay taxes to help “the other”. This limits the size of government. The ethnic-diversity-and-redistribution-literature has almost entirely focused on this sole effect.

This is a strange way to describe the state of the argument.  It would be more accurate to say that (a) Almost everyone assumes that immigration increases the size of the welfare state; (b) AGS identified a mechanism going in the opposite direction; and (c) AGS showed that on international data, the net effect of diversity on the welfare state is indeed negative.  There is a -.66 bivariate correlation between social spending as a percent of GDP and racial fragmentation, and this relationship persists controlling for per-capita income, region, and age distribution.

I freely admit that AGS’s results for the U.S. states are indeed subject to Sanandaji’s critique here.  AGS only show that states with larger black population shares spend less per person, not less in total.  But even this weaker result shows that the standard story is overstated.

Now let’s turn to Sanandaji’s “three effects that go the other way”:

2. Increasing the share of low income individuals increases the welfare state through a mechanic effect. This means even if you don’t vote for any changes to the welfare state, the use of preexisting welfare programs such as unemployment insurance and public health care increases.

True enough, but this “mechanic effect” is precisely the kind of thing that economics rightly teaches us to second-guess.  You could just as easily say that the “mechanic effect” of raising the price of iPhones is to increase profits.  In both cases, you need to consider how the mechanic effect interacts with behavior.

Sanandaji’s next mechanism tries to do just that:

3. More disadvantaged citizens increases the need for a welfare state. To the extent that the welfare state reflects a desire to reduce social problems, having more deprived individuals increases the demand for more government to solve problems.

This sounds good, but it’s theoretically confused.  The standard microeconomic story is: (a) the more poor people there are, the higher the cost of reducing the poverty rate; (b) voters will respond to this higher cost by spending less per person; (c) the net effect on total welfare spending (number of recipients times spending per recipient) is therefore ambiguous.

Sanandaji continues:

The welfare state exists largely because the middle classes and the rich feel sorry for the poor. The left is not stupid or irrational, they rarely demand government intervention where there are few problems.

Once again, this sounds good – until you remember that “poverty” is relative.  By global standards, almost everyone in the First World is rich.  But First World countries still have Social Security, Medicare, free public education, etc.  And few of these benefits are means-tested!  Contrary to Sanandaji, the welfare state exists largely because (a) people receive money whether they need it or not, and (b) the middle classes and the rich constantly redefine poverty to ensure its continued existence.  Not just the left, but the entire political spectrum is indeed irrational, and habitually demands government intervention for no good reason.

Sanandaji’s fourth point:

4. Though ignored by proponents of the ethnic-diversity-and-redistribution, minorities also get to vote, and they vote overwhelmingly for the left.

I’m puzzled by this claim.  AGS’s Table 10 econometrically confirms that blacks are much more in favor of redistribution than the general population.  Their whole story is precisely that the indirect effect more than offsets the direct effect.  AGS’s case would be even stronger if they had a measure of total spending on the y-axis rather than a measure of per-person spending.  But contrary to Sanandaji, they’re not “ignoring” this point.

Sanandaji also neglects the possibility that immigrants vote “overwhelmingly for the left” primarily because right-wing parties treat them with such hostility.  Bush’s share of the Hispanic vote in 2004 was probably overstated (39%, not 44%, is a reasonable estimate), but he still made impressive gains simply by treating Hispanics with respect and taking a relatively pro-immigration stance.

Sanandaji continues with what is arguably his best point:

This effect is dominant when we are discussing free migration, because with open borders in a world where 700 million people have told Gallup they would like to migrate right now, sooner or later the immigrants will become the majority of voters and make the political preferences of the natives irrelevant.

AGS show that in the United States, higher black population shares predict lower welfare benefits (but not total welfare spending), even controlling for median income.  But if you study their Figure 5 closely, you’ll notice that blacks are a minority in every U.S. state.  What would happen if blacks actually formed a majority?  Maybe, as Sanandaji suggests, there would be a U-shaped effect – beyond 50%, AFDC benefits would start to rise.  Indeed, I bet that is precisely what would happen.

However, there is a crucial difference between African Americans and immigrants.  African Americans have a shared group identity.  Immigrants do not.  They identify at most with other people from the same country.  And even there, internal divisions between “early” and “late” arrivals are common – see Jews or Cubans.  By the time that “immigrants” form a majority in any First World country, most of them will probably no longer consider themselves to be “immigrants.”  See the United States, where the grandchildren of immigrants often have trouble pronouncing the word “immigrant” without reproach.

In the conclusion of his post, Sanandaji provides two graphs in support of his skepticism.  For a blog post, they’re impressive graphs; if you read Sanandaji’s comments, he diligently responds to criticism with additional legwork.  Nevertheless, AGS’s graphs and regressions are better, especially their Figure 4 and Table 9, which show social spending as a function of diversity controlling for the leading confounding variables.  Until I see contrary evidence of comparable quality, AGS’s story remains the one to beat.

P.S. I just noticed that Sanandaji replied to my earlier post on AGS.  He argues that if we limit AGS’s sample to the OECD, their result goes away:

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.

A fair point, but instead of throwing away most of the data and variation, it’s better to keep the data and add control variables.  That’s precisely what AGS did, and found that their result held controlling for per-capita income.