More on Sanandaji’s book about social-democracy in Scandinavia

If you’re interested in economic policy, Matt and I have been talking about Sanandaji’s book more, so perhaps you will find our discussion interesting. It started in the comments of my post about the book, but continued in the comments of this other post. In the course of that discussion, we have been talking about several things, but the main point we have been discussing is Sanandaji’s claim that, although poverty is less prevalent in Scandinavia than in the US, people of Scandinavian ancestry in the US are even less likely to be poor than their counterparts in Scandinavia. Matt argued that the methodology used by Sanandaji to establish that claim was flawed and I think it’s fair to say that he conclusively showed that. However, I argued in response to him that even though Sanandaji’s argument was not sound, his conclusion was nevertheless true. Matt disputed that claim, but I argue in my latest reply to him that, in doing so, he made the same kind of mistake that Sanandaji made in his book. This is where things are for the moment, but perhaps Matt will have something to say in response.

EDIT: For those who are interested in this exchange, Matt just replied to my latest comment, so you should have a look at what he says. I’ll reply to him when I have some time, which may take a while again, since I have a lot of work. (By the way, his comment initially ended up in the spambox and the only reason I noticed it is because he emailed me about it, so if your comment still doesn’t show after a while please email me to make sure the same thing didn’t happen to you.)

6 thoughts

  1. I thought it would be useful to make a few updates to the most recent argument I made in the other thread, and to summarize where the dialectic stands at this point in time, from my perspective (spoiler: I’m winning).

    Before going on, both Philippe and I have been using the Census Bureau’s yearly poverty rates for the US as a whole as given in Table 2 on the CB’s website. But, for 2005-2015, we’ve been getting our poverty rates for the various Scandinavian-American groups from the figures given on this website (ask Philippe how to find the data), which is taken from the American Community Survey (ACS). But the ACS’s figures for US poverty overall are slightly different from the figures given in Table 2 linked to above. For example, in 2015, Table 2 says that the poverty rate for the US as a whole was 13.5%, but the ACS says it was 14.7%. That’s not a huge difference, but I think, for 2005 and after, we should use the ACS’s numbers because that’s where we’re getting our estimates of Nordic-American poverty. Table 2 and the ACS are using the same definition of poverty (the gross income Orshansky definition), as far as I can tell, but they use different data sets. Table 2 is getting its data from the Annual Social and Economic Supplement to the Current Population Survey (CPS-ASEC), which is different from the American Community Survey. The main differences between CPS-ASEC and the ACS are summarized here.

    For the year 2000, the data set on which the estimates of Scandinavian-American poverty are based is apparently the Census 2000 Summary File 3 (SF3), which is different from both the ACS (which didn’t exist yet) and the CPS-ASEC. The overall poverty rate in 2000 according to SF3 was 12.4%, whereas according to CPS-ASEC it was 11.3%.

    Recently, I pointed to Notten and de Neubourg (2007), in which one can find (in Figure 1 on p. 36) gross income Orshansky-poverty rates for Denmark and Finland (though not Sweden, and Norway was not included in the study). Notten and de Neubourg unfortunately do not give these data in a table in numerical form, but only in the form of line graphs that plot the poverty rate for each country by year. However, by eyeballing the graphs one can see that the gross income Orshansky-poverty rates of both Finland and Denmark were well below 5% for the year 2000.

    I argued that this allows us to make a relatively direct comparison between, on the one hand, the poverty rates of Denmark and Finland, and on the other, the poverty rates of Danish- and Finnish-Americans as measured by the Census Bureau. This is because both N&N (2007) and the Census Bureau are using the gross income Orshansky definition of poverty. According to the Census Bureau, for the year 2000, 5.0% percent of Danish-Americans, and 6.3% of Finnish-Americans, had incomes below the gross income Orshansky-poverty line. As I said above, eyeballing the graphs in N&N (2007: 36; Figure 1) tells us that the poverty rates of both Finland and Denmark were well below 5.0% (they seem to each be around 2.5%, roughly half of Danish-American poverty and around 40% of Finnish-American poverty).

    Philippe responded by pointing out that, though the Census Bureau and N&N (2007) are using the same definition of poverty, they are using different data sets. Moreover, this difference in data is apparently enough to make the 2000 poverty rate for the US as a whole, as measured by the Census Bureau, substantially lower than that given by N&N (2007) for the US as a whole.

    I replied that it is not at all clear that N&N’s figure is lower than the Census Bureau’s because they use different data sets. N&N’s (2007) definition of gross income includes one in-kind benefit, food stamps (p. 16), while the US government does not count food stamps as income for the purposes of measuring poverty. Although the ACS does appear to count food stamps as income (see the fact sheet linked to above), this is irrelevant to the question at hand, since the Census did not ask about food stamp income in 2000 (p. 19), the last year for which N&N have data. Therefore, the discrepancy between N&N’s US poverty rate and the Census Bureau’s may have nothing at all to do with the different data sets, and may be entirely the result of the fact that N&N counted food stamps as income, whereas the Census Bureau didn’t. (Nordic countries don’t have food stamps, so N&N’s estimates for Finland and Denmark aren’t affected by their decision to count food stamps as income).

    One might wonder why we shouldn’t count food stamps toward the reduction of poverty. I tend to agree with this proposal, but only if we also include all poverty-reducing in-kind benefits. It is quite likely that Scandinavian countries are considerably more generous in offering in-kind benefits to their citizens than is the US. If you look at Figure 1 on p. 906 of this paper, you can see that Sweden’s spending on in-kind transfers in 2001 was somewhat higher (as a percentage of GDP) than America’s, while Finland’s was somewhat lower. However, as the authors point out (p. 905), American spending on non-cash welfare is driven largely by education and especially health care expenditures. The large spending on health care is in turn driven primarily by the high cost of health care in the United States, compared to other developed countries. Furthermore, contrary to both liberal and conservative propaganda, education does not do much to reduce poverty. So, after adjusting for high education spending in the US, and the high cost of health care, I would expect that the inclusion of in-kind transfers would lower Scandinavian poverty rates relative to the American poverty rate.

    Another problem is that I have so far only been speaking of poverty rates as measured by gross income. Certainly we should want to take the effect of taxes into account. Since Scandinavians of all income levels (including the bottom) pay more in taxes than do Americans, the consideration of only pre-tax income may be deflating Scandinavian poverty rates relative to the American rate. Moreover, low-income working Americans receive a tax-based benefit known as the Earned Income Tax Credit (EITC) (a sort of negative income tax), which is not counted as income for the purposes of measuring poverty, as it is technically after-tax income. So we should certainly want to compare the after-tax poverty rates of Scandinavian-Americans with the after-tax poverty rates of Scandinavian countries. Unfortunately, we cannot directly do this, as the Census Bureau’s estimates of poverty rates do not account for taxes, and the Census Bureau is our only source for the poverty rates of Scandinavian-Americans.

    However, N&N (2007: 35; Table 2) do tell us what the “net” (after-tax) income Orshansky-poverty rates of Denmark, Finland, Sweden, and the US are. For the year 2000, they are:

    Denmark – 3.4%
    Finland – 4.9%
    Sweden – 5.7%
    US – 8.7%

    Philippe thinks that, in order to roughly estimate what N&N would calculate the poverty rates of Nordic-American groups to be, we should determine the ratio of each Nordic-American group to the overall poverty rate for the US as determined by the Census Bureau, and multiply it by N&N’s figure for overall US poverty. Then, we can compare this result to N&N’s poverty rate for the respective Nordic country.

    According to Philippe, in general, when we are trying to guess how a given author, who uses different methodology from the Census Bureau, would estimate the poverty rate of an American ethnic group, we can use the following equation. Here E is an American ethnic group, A(E) is the imputed poverty rate (as it would be determined by the author of the study) of E, C(E) is the poverty rate of group E as determined by the Census Bureau, C(US) is the overall US poverty rate as determined by the Census Bureau, and A(US) is the overall US poverty rate as determined by author A. I call this equation (P).

    (P) A(E) = [C(E)/C(US)]*A(US)

    This equation rests on the assumption that, no matter the definition of poverty or the data set used, the ratio of the poverty rate of an American ethnic group to the overall American poverty rate will be roughly the same. As you shall see, I am going to challenge this assumption in a moment, but for now, let’s see how it applies in this case.

    The poverty rates, as calculated by the Census Bureau in 2000, for Danish-Americans, Finnish-Americans, Swedish-Americans, and the US as a whole, are 5.0%, 6.3%, 5.5%, and 12.4%, respectively. If we use equation (P) to calculate the poverty rates of Scandinavian-American groups as they would be calculated by N&N for the year 2000, we get the following results:

    Danish-Americans – 3.5%
    Finnish-Americans – 4.4%
    Swedish-Americans – 3.9%

    In each case, the poverty rate of the Nordic-American group in question is either lower than that of its ancestral country (as calculated by N&N), or only slightly higher (in the case of Denmark). Therefore, Philippe concludes that Nordic welfare states are not doing much to reduce poverty.

    Before I get to my main critique of Philippe’s argument, I want to draw attention to a few limitations of N&N (2007)’s data that weaken this argument. Note, in equation (P) above, that as A(US) rises so does A(E). Therefore, if, for each Nordic country N, A(US) – A(N) is smaller than what N&N have estimated it to be, this gives us reason to resist the conclusion of Philippe’s argument. As it happens, there are several biases in N&N’s estimates, all of them favoring the US over the Nordic countries (see N&N 2007: 31).

    As I have already mentioned, N&N (2007) count food stamps as income, but do not count all poverty-reducing in-kind benefits as income, and I gave above what I consider good reasons for thinking that doing the latter would favor the Nordic states.
    Furthermore, N&N (2007) assume a 100% take-up rate for EITC, but, as they acknowledge, this cannot be accurate, since not everyone who is eligible for EITC benefits receives them (p. 18). This assumption deflates after-tax US poverty below what it actually is. There is also the problem that the poverty rates of Denmark, Finland and Sweden are based on register data, whereas those of every other country (including the US) are based on survey data, and N&N believe that register data may give higher poverty estimates than does survey data (pp. 15-16). This would artificially increase Scandinavian poverty rates relative to the US poverty rate. So, I do not think we should take N&N’s poverty figures for the US and Scandinavia at face value, apply equation (P) to the US figure, and compare the result to the Scandinavian figures, as Philippe thinks we can do.

    However, leaving these quibbles aside, I am now going to offer a deeper critique of Philippe’s argument. The gist of my counterargument is that Philippe’s argument, when applied to different studies that compare absolute poverty rates across countries, produces different, inconsistent conclusions, sometimes favoring his position, but sometimes favoring mine. Therefore, we cannot trust Philippe’s argument. The details follow below.

    Gornick and Jäntti (2011) give estimates for Danish, Finnish, Swedish, and American absolute poverty (based on disposable income) for the year 2004. The figures they give are as follows (Table 1, p. 26):

    Denmark – 2.9%
    Finland – 5.5%
    Norway – 3.8%
    Sweden – 4.4%
    US – 9.4

    The Census Bureau’s American Community Survey (ACS) does not have data for 2004 The closest year for which they have data is 2005. I am assuming that a one-year discrepancy will not be much of a problem. In the Census Bureau’s other yearly measure of poverty (the one based on CPS-ASEC), the poverty rate changed by only one-tenth of one percentage point between 2004 and 2005 (from 12.7 to 12.6).

    Here are the ACS’s estimates of the poverty rates of Danish-Americans, Finnish-Americans, Norwegian-Americans, Swedish-Americans, and Americans in general, for the year 2005:

    Danish-Americans – 5.9%
    Finnish-Americans – 7.2%
    Norwegian-Americans – 6.8%
    Swedish-Americans – 6.8%
    US – 13.3%

    Applying equation (P) to get the poverty rates of Nordic-American groups, as the would have been calculated by G&J (2011), we have:

    Danish-Americans – 4.2%
    Finnish-Americans – 5.1%
    Norwegian-Americans – 4.8%
    Swedish-Americans – 4.8%

    By Philippe’s method as applied to G&J (2011) Denmark and Norway have substantially lower poverty rates than do Danish- and Norwegian-Americans, respectively. Swedish-Americans have a somewhat higher poverty rate than does Sweden (though this difference was just outside the margin of error as calculated by the ACS). Finnish-Americans do have a somewhat lower poverty rate than does Finland (however this difference was just barely within the margin of error).

    So, on the whole, it looks like Philippe’s argument, when applied to G&J (2011), supports my position more than it supports his.

    Now let’s take a look at the Stanford Center on Poverty and Inequality’s 2016 report (the chapter on poverty was also written by Gornick and Jäntti). This gives estimates of absolute poverty rates, based on disposable income, for the year 2010, for Denmark, Finland, Norway, and the US.* They are as follows:

    Denmark – 3.2
    Finland – 4.0
    Norway – 3.4
    US – 9.2

    The ACS gives, for 2010, the following poverty rates for Danish-Americans, Finnish-Americans, Icelandic-Americans, Norwegian-Americans, and Americans as a whole:

    Danish-American – 7.2
    Finnish-American – 8.0
    Norwegian-Americans – 8.1
    US – 15.3

    Now, applying equation (P) gives us the following imputed figures, for Nordic-American ethnic groups:

    Danish-Americans – 4.3
    Finnish-Americans – 4.8
    Norwegian-Americans – 4.9

    All of these are substantially higher than the poverty rates of the three Scandinavian countries as given by G&J (2016), and all the differences are outside the margin of error. So equation (P), when applied to G&J (2016), supports my position much more than it supports Philippe’s (see also the footnote about Iceland and Icelandic-Americans).

    Let me be clear about what I am arguing here. I am not using equation (P) as applied to G&J (2011, 2016) to support my position. Rather, I am trying to undermine Philippe’s argument for his position. Equation (P), when applied to N&N (2007, 2011), produces a conclusion that is inconsistent with that yielded by (P) when applied to G&J (2011, 2016). Therefore, either there is something wrong with equation (P), or there is something wrong with at least one of the studies we have been looking at. If the former, then Philippe’s argument is unsound. If the latter, then, since we have no more reason to trust N&N than we have to trust G&J, we therefore have no reason to think Philippe’s argument is sound. In either case, Philippe does not succeed in supporting his conclusion.

    If there is something wrong with equation (P), then it must be because the assumption on which it is based is false. That is, different measures of poverty do change the ratio of Scandinavian-American poverty to overall poverty. Why might this be the case? I can think of at least one hypothesis. Perhaps Scandinavian-Americans are less likely to avail themselves of welfare benefits even when they are eligible, maybe because their stellar culture and work ethic (the one Sanandaji likes so much) makes them ashamed to take such benefits. In particular, eligible Scandinavian-Americans are less likely to receive the EITC than other eligible Americans. So, the after-tax poverty rate of Americans in general will fall relative to the pre-tax rate, whereas the after-tax poverty rate of Scandinavian-Americans will be roughly the same as their pre-tax rate, or perhaps even go up. Therefore, when switching from a pre-tax income definition (like the Census Bureau’s) to an after-tax definition (like the ones used by N&N and G&J), the ratio of the percentage of Scandinavian-Americans in poverty to Americans in poverty will increase.

    Since Philippe’s argument is either unsound, or is such that we have no reason to think it is sound, and my comparison of pre-tax Orshansky-poverty rates based on the Census Bureau and N&N (2007), for all its problems, provides a more direct comparison anyway, I conclude that the weight of the evidence supports my position considerably more than it supports Philippe’s. This conclusion is strengthened when we remember that all the Nordic countries have immigrants and descendants of immigrants, and that there is some evidence of positive selection of Danish emigrants (though there is also some evidence of negative selection of Norwegian emigrants, and the evidence, in both directions, on migrant selection, is scant and inconclusive).

    *The study also includes the poverty rate for Iceland, but there aren’t enough Icelandic-Americans for the ACS to have 1-year estimates of their poverty rate, though it does have 3-year and 5-year estimates. According to these estimates, Icelandic-American poverty is surprisingly high. For example, from the years 2009-2011, 10.7% of Icelander-Americans were living in poverty. The next highest Nordic group was Finnish-Americans at 8.3%, a considerable difference (since there were only around 50,000 Icelandic-Americans at this time, there was a higher margin of error for the Icelandic-American poverty rate. However, the difference between Icelandic-Americans and Finnish-Americans was still outside the margin of error). Sanandaji likes to favorably compare Iceland to other Nordic countries, noting that Iceland has a higher life expectancy and lower infant mortality than other Nordic countries and lower income inequality, despite its smaller welfare state and lower taxes: “Why is it, for example, that Iceland, with a moderately sized welfare sector, has outpaced the four major Scandinavian countries in terms of life expectancy and infant mortality? Why does Iceland top the income equality league?” (117). However, Sanandaji fails to note that Iceland, while having a smaller welfare state and lower taxes than other Scandinavian countries, is in other ways closer to the social democratic ideal than the other Nordics. For instance, Iceland has the highest trade union density in the OECD, and probably the world. It also looks as though the poverty rate of Icelandic-Americans (who should have a culture similar to that of Icelanders back home) is higher than that of other Scandinavian-American groups. If one takes the ratio of Icelandic-American poverty to total poverty for the years 2009-2011 (10.7/15.2 = 0.7) and multplies it by G&J (2016)’s (single-year) figure for the US in 2010 (9.2), one reaches the conclusion that Icelandic-Americans had a poverty rate of 6.5% by G&J’s methodology, which is 97% higher than G&J (2016)’s figure for Iceland in 2010 (3.3%), at least if we take Philippe’s procedure seriously. Since Icelandic-Americans presumably share a similar culture with Icelanders, and enjoy both a smaller welfare state, and much less protection from unions than do their cousins back home in Iceland, Iceland’s welfare generosity (relative to America’s) and high union density may go a long way toward explaining the apparent gap in poverty rates.

    1. By the way, Philippe, since I can anticipate that you will be converting to socialism after reading this, I’d like to wish you, in advance, a happy International Working Women’s Day, and a happy centennial anniversary of the February Revolution.

      In Solidarity,


  2. Hahaha, other people have tried to convert me, but I’m afraid they have always failed so far 🙂 So I won’t celebrate the February Revolution, but in 2 years from now, I will gladly celebrate the suppression of the Spartacist uprising in Berlin. Actually, I probably won’t, because I have a weird fondness for Rosa Luxemburg. Mostly because, unlike the socialists who had the uprising crushed, she actually had some balls… Anyway, I know I still need to reply to you and I will do it sometime this week or perhaps this weekend, but I have been horribly busy. (In part because I started dating a Norwegian-American, which I guess is pretty funny, considering the situation. I’ll be sure to ask her whether she grew up in abject poverty ^_^)

  3. Equation (P), when applied to N&N (2007, 2011), produces a conclusion that is inconsistent with that yielded by (P) when applied to G&J (2011, 2016). Therefore, either there is something wrong with equation (P), or there is something wrong with at least one of the studies we have been looking at.

    It occurs to me that there is another possibility: things could have been different in 2000 (which when the data N&N use was collected) than they were in 2004 and 2010 (when the data G&J use was collected). Support for this hypothesis comes from a paper I just found:Gornick and Jantti 2010, which looks at poverty rates for the US, Denmark, Finland, Norway, and Sweden roughly around the year 2000.

    Here’s the figures G&J (2010) give (Table 1, p. 352):

    Denmark – 3.9%
    Finland – 11.7%
    Norway – 3.4%
    Sweden – 9.0%
    US – 9.1%

    Already, we can see that the total US poverty rate around 2000 was lower than Finland’s and only just above Sweden’s. Remember, for year 2000, the Census Bureau gives the following poverty figures:

    Danish-Americans – 5.0%
    Finnish-Americans – 6.3%
    Norwegian-Americans – 5.9%
    Swedish-Americans – 5.5%
    US – 12.4%

    Applying equation (P), we get:

    Danish-Americans – 3.7%
    Finnish-Americans – 4.6%
    Norwegian-Americans – 4.3%
    Swedish-Americans – 4.0%

    So, we can see here that, according to (P) and G&J (2010), only Norwegian-Americans had a higher poverty rate in 2000 than their ancestral country. Finnish- and Swedish-Americans had much lower rates. Danish-Americans had a slightly lower rate, but when we look at child poverty (<18) the situation clearly favors Denmark. Denmark had a child poverty rate of 2.0% for the year 2000, and the US had a rate of 11.7%, in G&J (2010) (Table 2, p. 354). The Census Bureau has several different figures for child poverty, and I am not sure which one corresponds to G&J's definition. For completeness' sake I list all of them for the year 2000:

    Danish-American related children under 18 – 4.7%
    Danish-American families with related children under 18 – 4.8%
    US related children under 18 – 16.1%
    US families with related children under 18 – 13.6%

    Applying equation (P), we obtain:

    Danish-American related children under 18 – 3.4%
    Danish-American families with related children under 18 – 4.1%

    Both of these figures are much higher than G&J (2010)'s figure of 2.0% for Danish child poverty in 2000. So, since Danish overall poverty was only slightly lower than Danish-American poverty (a difference that could well have been within the margin of error), and Danish child poverty was considerably lower than Danish-American child poverty, Denmark clearly did better than Danish-Americans when it comes to poverty in 2000 (especially because child poverty is worse than poverty in general).

    But it's clear that Finland and Sweden, according to Philippe's argument, were doing much worse in 2000 then Finnish and Swedish-Americans (though this was reversed in Sweden by 2004 and in both countries by 2010). What's going on here? Well, Finland and Sweden suffered severe economic recessions in the 90s, much worse than the relatively tame one in the US. In Finland’s case it was an outright depression, even worse than the 1930s depression was for Finland. Although both countries recovered pretty quickly, they started further behind than they would have if their recessions/depressions had been as mild as the US’s recession. (It’s also the case that the Swedish and Finnish welfare states were savagely rolled back during the 90s, although that was also true to some extent for the US welfare state, much smaller to begin with).

    This speaks to a more general problem with the kind of comparisons Sanandaji and Philippe want to make between Nordic countries and Nordic-American groups. If the only two inputs that determined Nordic-American poverty were (A) Nordic culture, and (B) transfers, then it would be relatively easy to determine the relative importance of these factors by comparing Nordic-American outcomes to Nordic ones, since they presumably have much the same culture, but vastly different tax-and-transfer regimes. But there is (at least) one more factor: general economic health. Nordic-American economic success is inextricably tied to the success of the US economy as a whole, whereas Nordic economic success is not (or is much less so). If you select a year in which the US was doing better than some or all of the Nordics, and had been for some time, you are going to skew the results. It’s better to pick a year in which both the US and the Nordics are doing well, or both are doing poorly. This is why I prefer G&J (2011) and especially (2016), since these studies use data from 2004 (when both the US and the Nordics were doing well) and 2010 (when both the US and the Nordics were doing very badly). Both of those studies support my position rather than Philippe’s.

    1. Now that we have G&J (2010) in hand, we can verify a point Bruenig made in the post I originally cited: the Labour governments of Blair and Brown in the UK made tremendous strides in reducing child poverty.

      Take a look at Table 2, p. 354, in G&J (2010). Disposable income absolute (US line) child poverty was 22.9% in 2000. Now look at Table 2 on p. 19 of G&J (2016). Disposable income absolute (US line) child poverty was 8.1% in 2010. Child poverty dropped to almost a third of what it was just a decade before.

      We don’t have to worry too much here about definitions of poverty and what not, since these are the same researchers using the same definition and the same data set (the LIS), measuring the poverty rate in one country over a decade.

      The most parsimonious explanation for this is the dramatic increase in transfers to families with children under the Blair and Brown governments. It looks like Labour cut child poverty in the UK by two-thirds. That’s really, really good. Unfortunately, it doesn’t begin to make up for all the Iraqi children impoverished and murdered…

      1. In fact, I think we can be basically certain it was the transfers. It certainly wasn’t the economy. If it was, you’d expect to see the “market income” poverty rate fall by a lot. But look at G&J (2010), Table 2, 354. The market income (before taxes and transfers) US line child poverty rate for the U.K. in 2000 was 34.9%. Now look at G&J 2016, Table 2, 19. The market income US line child poverty rate was 32.8%. That’s barely a change at all compared to the precipitous decline in post-taxes-and-transfers poverty.

        What else could explain it? Ah, it must have been that British kids adopted Scandinavian values over the course of 10 years….

Comments are closed.