Tuesday, February 17, 2015

Does social democracy / a large welfare state kill innovation?

Image: At The Next Web, Brad McCarty documents the history of the smartphone, including this 2001 state-of-the-art Nokia Communicator.

[Updated--see below]

As I often do with viewpoints I don't like, I find the most reputable expression of that idea so I can't be accused of choosing a poor representative for my critique. Enter Daron Acemoglu (MIT), James A. Robinson (Harvard), Thierry Verdier (Paris School of Economics), and their 2012 non-peer-reviewed paper Can't We All Be More Like Scandinavians? People with Big Ideas about welfare states eagerly point to this paper as proof that their evidence-free Big Ideas were right all along. Let's see what this paper really says.

The thrust of the paper is this: it's a well known fact that large welfare states stifle innovation, mostly due to limiting inequality and perhaps also by limiting economic insecurity. These are the "cuddly capitalists" (their term). In the "cutthroat capitalist" (their term) countries, inequality is much greater and economic security more tenuous, and--as a direct result--innovation is greater. Since the rewards are so great for success--and the consequences for failure so severe--incentives line up perfectly to maximize innovation.

So far, this isn't anything new. But Acemoglu, Robinson & Verdier take this logic one step further. Not only is there greater innovation in the cutthroat capitalist countries, they posit, but the innovation of the cutthroat capitalists makes cuddly capitalism possible. Were it not for the greater innovation and resulting economic growth which spills over from the cutthroat to cuddly capitalist countries, the large welfare states of the cuddly capitalist countries would not be possible.

But wait! There is a slight bump in the road to solemnly dismantling the welfare state in the name of innovation. It's always taken as fact that the United States, the most cutthroat of the capitalists of the developed world, handily beats the rest of the world in innovation and technology. But where is the evidence?

Fortunately, Acemoglu, Robinson & Verdier have marshalled the strongest evidence available to demonstrate the superior capacity for innovation in the United States compared to the social democracies:
The United States is also widely viewed as a more innovative economy, providing greater incentives to its entrepreneurs  and workers alike, who tend to respond to these by working longer hours, taking more risks and playing the leading role in many of the transformative technologies of the last several decades ranging from software and hardware to pharmaceuticals and biomedical innovations. Figure 1 shows annual average hours of work in the United States, Denmark, Finland, Norway and Sweden since 1980, and shows the significant gap between the United States and the rest.
Sure enough, in a graph with a y-axis that doesn't begin at zero, Figure 1 indeed shows that people in the United States work more hours than in Scandinavia, except for the weird part in the early 80's (and earlier) where the Finns worked longer, that I guess we're just ignoring:

Some problems should be immediately obvious.

First, Acemoglu, Robinson & Verdier use the United States to represent all of the cutthroat capitalists. Might their conclusions be stronger if they considered other countries with high inequality and a fragile social safety net? This issue will be taken up below, but it shouldn't be too hard to guess why they omitted other cutthroat capitalists, like New Zealand and Ireland, from their analysis.

Second, Acemoglu, Robinson & Verdier assume that more work hours at the aggregate level results in more innovation. It doesn't. Longer aggregate work hours are simply indicative of higher poverty rates. It's well established that people work longer hours in countries with higher poverty rates. Thus, since the United States has such a high poverty rate while the Scandinavian countries have such a low poverty rate, it's no surprise that the average number of hours worked is lower in the Scandinavian countries. Clearly, average hours worked each week isn't a proxy measure of innovation, but a proxy measure for a country's poverty rate.

Third, productivity research is unequivocal that working more hours results in less innovation, not more. For a striking example, productivity experts estimate that if the early Macintosh engineers had worked 40 hours per week instead of 90, the first Macintosh computer would have been ready for release a full year earlier. In short, all available research argues that long hours lead to less innovation--meaning that Acemoglu, Robinson & Verdier could scarcely have chosen a worse proxy measure for innovation.

In sum, Acemoglu, Robinson & Verdier's first measure of innovation--a country's average weekly work hours--does not actually measure innovation. Fortunately, Acemoglu, Robinson & Verdier don't trouble their readers with confounding variables like poverty or productivity research and charge right along to exhibit B demonstrating the superior innovation of the United States.

Exhibit B examines patents. Surely countries with most patents per person must be more innovative than the rest--and wow does the United States ever have more patents per capita than any other country:

Already, however, this data is suspicious. The welfare states of both Scandinavia and the United States had been established for decades prior to the mid 1990's; yet Sweden and Finland apparently had more per capita patents filed in the mid 1990's than the United States. How do the authors explain this temporary lull in American innovation? Or, perhaps Acemoglu, Robinson & Verdier would like to explain what policy of the Clinton Administration caused the sudden explosion of American innovation beginning clearly in 1996.

Or maybe patent trolling can explain much of the difference in patents per capita.

Indeed, not even Acemoglu, Robinson & Verdier can ignore the fact that the inanity of United States patent law creates extraordinary numbers of worthless patents. They acknowledge this fact and attempt to control for patent trolling:
These differences may partly reflect differential patenting propensities rather than differences in innovativeness, or may be driven by "less important" patents that contribute little to productive knowledge and will receive few cites (meaning that few others will build on them). To control for this difference, we adopt another strategy. We presume that important--highly-cited--innovations are more likely to be targeted to the world  market and thus patented in the US patent office (USPTO)...Figure 3 plots the numbers of patents granted per one million residents for  Denmark, Finland, Norway and Sweden relative to the United States between 1980 and 1999.  Each number corresponds to the relevant ratio once we restrict the sample to patents that obtain at least the number of citations (adjusted for year of grant) specified in the horizontal axis.
The problems here should be obvious. To restate their strategy, Acemoglu, Robinson & Verdier try to control for American patent trolling by only looking at patents filed with the United States patent office, as this is likely the best patent office for innovators looking to sell on the global market. They also weight the resulting patents per capita by number of citations, with the idea that a patent that is cited by many other patents must have made a more important contribution than a patent that isn't cited by very many other patents. They also weight them by "year of grant" without explaining what/how they are weighting or what this actually means.

But there's two major problems with only looking at patents in the United States, even if they are weighted by number of citations from other patents. Acemoglu, Robinson & Verdier reveal the first issue, apparently without any irony, in a footnote (emphasis added):
Another plausible strategy would have been to look at patent grants in some "neutral" patent office or total number of world patterns. However, because US innovators appear less likely to patent abroad than Europeans, perhaps reflecting the fact that they have access to a larger domestic market, this seems to create an artificial advantage for European countries, and we do not report these results.
So--per Acemoglu, Robinson & Verdier--European innovators have an "artificial advantage" in filing patents in Europe. But the opposite must also be true; "US innovators" have a similar "artificial advantage" filing patents in the United States. Perhaps European firms really are more likely to file patents in the United States than American firms are to file patents in Europe, but that doesn't mean the patents of European firms filed in Europe can be ignored. By only looking at filings in the American patent office, a significant number of European inventions and innovations--those patents filed in Europe--are thus being disregarded. Obviously, looking at patents filed worldwide would be a better--albeit not perfect--measure. Interestingly, Acemoglu, Robinson & Verdier actually had these data but they "[did] not report these results." One wonders why they didn't report these results. Maybe the results didn't support their thesis.

In other words, Acemoglu, Robinson & Verdier expect us to take them at their word that it never occurred to them that American firms might have an artificial advantage filing patents in the United States. It occurred to them that European firms have an artificial advantage filing patents in Europe, but it never occurred to them that American firms might have an advantage filing patents in the United States.

The second major problem is the unresolved question of whether or not a large numbers of citations actually indicates a more important or innovative patent. Acemoglu, Robinson & Verdier assume this is true, but productivity researchers have not resolved the question of whether or not number of citations is a useful measure of the quality of patents (for an example, this issue is discussed in several places in this paper). A patent might receive many citations simply because it is famous; a patent might receive very few citations because its full potential was not realized for years; and applied science science patents might be expected to have fewer citations than basic research patents of equal importance. There is significant debate over what type of research is most important. It is doubtful that a state-of-the-art final product would have many citations. Rampant American patent trolling could inflate the number of citations of unimportant or even useless patents. Clearly, the number of citations is an obviously dubious measure of patent quality.

In sum, Acemoglu, Robinson & Verdier's second measure of a country's innovation--patents per capita, weighted in some way that they didn't explain by number of citations and year--is not a measure of innovation. It ignores a large fraction of European inventions and innovations, and it rests on the faulty assumption that patent quality can be measured by number of citations from other patents. In addition to several other key problems, out of control American patent trolling should make this measure extremely suspect.

To recap, Acemoglu, Robinson & Verdier base their assumption that the United States is the world's leading innovator on two pieces of data--one completely flawed and one extraordinarily dubious. They had a third piece of data, patent filings in worldwide patent offices, that they chose not to report.

And that's it. That's the best evidence they could find to argue that the United States leads the world in innovation.

I've limited my criticism to data Acemoglu, Robinson & Verdier actually cite. But Lane Kenworthy dismantles their work even further by scrutinizing their underlying assumptions that they don't even bother to provide data for. For example, Acemoglu, Robinson & Verdier take as a given that innovators in the United States have an especially high incentive to innovate due to high income inequality. But Kenworthy points out that the richest 1% in Sweden and the richest 1% in the United States really aren't as different as the authors assume. Presumably, a successful innovator would find herself in the richest 1%, whether in Sweden or the United States. But there is very little difference between wealth and tax rates of the American and Swedish top 1%. Indeed, the richest 1% of the social democracies have been getting richer and richer in lockstep with the richest 1% in the United States, wholly inconsistent with their thesis that inequality uniquely drives innovation.

Additionally, their strict division of cutthroat and cuddly capitalism collapses under even basic scrutiny. Government spending in Sweden and the United States was not so different until around 1960--in other words, Sweden was cutthroat until the 1960's. If cutthroat capitalism drives innovation, which in turn drives economic growth, then by the logic of Acemoglu, Robinson & Verdier, Sweden's switch from cutthroat to cuddly capitalism should surely have slowed economic growth. But it didn't. Acemoglu, Robinson & Verdier's facile model would predict that the economic growth rates of the United States and Sweden should diverge after 1960, when they have actually been steadily converging.

Finally, Kenworthy takes on the issue of Acemoglu, Robinson & Verdier using the United States to represent all cutthroat capitalists. This is extraordinary cherry picking of data. As Kenworthy points out, if being cutthroat results in greater innovation, then other cutthroat nations of the developed world (particularly New Zealand, Australia, and Ireland) with similarly low levels of government spending and high inequality should be innovation powerhouses. They're not; there's simply no relationship between being cutthroat and innovation. It's little wonder Acemoglu, Robinson & Verdier chose to ignore other cutthroat capitalists.

These are just the problems in the background of Acemoglu, Robinson & Verdier's paper. The rest collapses under the weight of their flawed assumptions. They assume that the cutthroat capitalists are more innovative, then set up a theoretical model under that assumption...and then unsurprisingly derive the result they originally assumed to be true. Put more simply, their abstract could have read:
We assume that A is true and that if A is true, then B must also be true. We set up a theoretical model where we assumed A was true and that B results from A. Indeed, our model found B.

Other measures of innovation
One might wonder why Acemoglu, Robinson & Verdier didn't choose more straightforward measures of innovation--like R&D investment as a percentage of GDP, or the number of science and engineering personnel per capita actually working in R&D. Though imperfect, these are clearly better measures of innovation than the roundabout ways Acemoglu, Robinson & Verdier attempted to measure innovation.

These data do exist, and they don't agree with the thesis of Acemoglu, Robinson & Verdier. The World Bank collects data on aggregate per capita spending on R&D. By this (albeit rough) measure of innovation, the United States is by no means peerless:

By this measure, the United States is at best middling compared to the social democracies. Finland, Sweden, and Denmark all invest a higher percentage of GDP to R&D than the United States, and the United States is roughly on par with Austria. However, there is a serious problem with these data. The United States' R&D budget is grossly inflated since the World Bank defines R&D expenditures, in part:
"R&D expenditures include all expenditures for R&D performed within a country, including...wages and associated costs of researchers, technicians, and supporting staff..."
Obviously, the American R&D budget is inflated by the "associated costs" of researchers, since they include benefits like health insurance that are more expensive in the United States than any other country. If health insurance costs were comparable to the rest of the world, the United States would surely fall in the rankings. Additionally, the United States private sector tends to have more managers per worker than European countries. Since managers would count as "supporting staff", this again artificially inflates the United States' aggregate R&D budget.

Indeed, if we look at the number of research personnel per capita, the United States ranks below each one of the social democracies:

Obviously, if the United States is were truly peerless in innovation, it should have more science and engineering personnel per capita than any other country. It doesn't.
For what it's worth, the 2014 Global Innovation Index, which examines 81 innovation indicators, ranked Sweden #3, Finland #4, the United States #6, Denmark #8, Norway #14, Iceland #19, and Austria #20. And--for what it's worth--the World Economic Forum's 2014 Global Competitiveness Report ranks Finland #3, the United States #5, Sweden #6, Norway #11, Denmark #15, Austria #16, and Iceland #31.

Clearly, there is no honest reading of this data that argues that the United States outperforms the social democracies in innovation. Obviously, there is no direct measure of innovation, but these proxy measures are surely better than those used by Acemoglu, Robinson & Verdier. There is no great difference in innovation between the social democracies and the United States.

Finally, as a group, the social democracies outperform the rest of the cutthroat capitalists (countries with high inequality and low government spending) in these same measures of innovation. For R&D expenditures as a percentage of GDP, the social democracies were ranked (1) Finland; (2) Sweden; (3) Denmark; (4) Austria; (5) Norway. If we take Denmark and Austria as representatives of the social democracies and compare them to the cutthroat capitalists outside of the United States:

--Denmark and Austria easily outperform the cutthroat capitalist group. Indeed, among the cutthroat capitalists, only Australia's R&D expenditures exceed the world average; New Zealand, the United Kingdom, and Ireland all fall below the world average for R&D expenditure.

Repeating this exercise for number of research personnel per capita:

--and Denmark and Austria again outperform the cutthroat capitalist group.

Obviously, none of these measures are able to compare the quality of innovation. Because it is impossible to define or measure the newness or importance of ideas, a systematic measure of innovation is simply not possible. Any comparisons will necessarily be anecdotal and therefore useless for reaching conclusions in the aggregate.

However, for an interesting anecdote, the early smartphone market was defined by competition between Nokia, a Finnish company, and Ericsson, a Swedish company.

Finally, research is typically treated as a sacred cow in the United States. It shouldn't be. Most research is of stunningly low quality ("We know that about $200 billion — or the equivalent of 85 percent of global spending on research — is routinely wasted on poorly designed and redundant studies. We know that as much as 30 percent of the most influential original medical research papers later turn out to be wrong or exaggerated.")

All of the data considered argue that social democracy does not encumber innovation. Compared to the United States, the social democracies make as great or greater per capita aggregate investments in research and development, and support as many or more per capita science and engineering personnel. By these same measures, the social democracies easily outperform all of the "cutthroat capitalist" countries outside of the United States. There is no honest reading of the data that argues that the social welfare and tax policies associated with the United States results in greater innovation; there is no honest reading of the data that argues that social democratic policies stifle innovation. We do not have to choose between universal child care and new inventions and technology.

It's very interesting that Acemoglu, Robinson & Verdier and others focus on high inequality and low government spending as the only serious candidates for drivers of innovation. Scientists and engineers need an education in order to innovate; much government spending funds basic research; a strong infrastructure network is the basis of communication and movement of materials needed to conduct research; etc. Yet the focus is overwhelmingly on high inequality and low government spending. It's almost as though Acemoglu, Robinson & Verdier were never actually concerned about maximizing innovation; it's almost as though their primary concerns are justifying high inequality and advocating cuts to government spending.

UPDATE 02/19/2016
Let's return to the reason Acemoglu, Robinson & Verdier believe the American economy should be more innovative than those of the social democracies (emphasis added):
The United States is also widely viewed as a more innovative economy, providing greater incentives to its entrepreneurs and workers alike, who tend to respond to these by working longer hours, taking more risks and playing the leading role in many of the transformative technologies of the last several decades
But Matt Bruenig points out that the social democracies actually have higher rates of entrepreneurship (number of new firms started in a year) than the United States (there was no data available for Norway):

Clearly, the social democratic system does not dull incentives to take risks and start new companies because more people are taking risks and starting new companies than in the United States. This is by Acemoglu, Robinson & Verdier's own definition of innovation: taking risks to start a new company and bring ideas to market. Surely some of these new companies will fail, but this risk taking, Acemoglu, Robinson & Verdier tell us, is what makes for a healthy entrepreneurial system.

Bruenig also takes a swing at innovation quality:
Stockholm is home to the hottest tech sector in the world after Silicon Valley (which is itself located in tax-loving California). In recent years, the Nordic countries have given the world Spotify, Skype, Mojang (maker of Minecraft), Rovio (maker of Angry Birds), Supercell (maker of Clash of Clans), and Klarna (cutting-edge fintech firm). The Nordic countries, led by Volvo, are also on the forefront of driverless car research and, led by Finland, on the forefront of mobile phone networking technology. The Nordics are also home, in Sweden, to the most successful pop music industry in the world.
There's nothing sluggish about the Nordic startup and innovation scene. Nothing at all.
Also, the social democracies have higher rates of venture capital investment (social democracies marked in red, United States in blue):

A very clever way to use patents as a proxy for innovation is to compare rates of "triadic patent family" filings. A triadic patent family is an invention that is filed in three patent offices: United States, European Union, and Japan. Clearly, this is a far better measure than patent filings in the United States or European Union patent office alone, and it eliminates the inanities unique to the United States patent system. I think I first saw this idea on Matt Bruenig's Demos blog and the Research Institute of the Finnish Economy might have been the first group to suggest it. Here is John Hsu's graph of traidic patent filings by country of origin:

Denmark, Finland, and Sweden have higher per capita triadic patent filings than the United States.

In the end--as with most concepts--there is no single indicator that can capture innovation. Using a wide range of measures, there is no indicator of innovation that the social democracies do worse on.

Update 03/16/2016
Bruenig at it again:
Consider this March 2015 article in the Financial Times titled "Stockholm: the unicorn factory." Unicorn is the name for startups that reach a $1 billion valuation. In the article, Murah Ahmed tells the tale:
For Stockholm, the focus of Sweden’s tech industry, Truecaller is not an outlier. In the past decade, this city of 800,000 inhabitants has churned out more billion-dollar tech companies than any in Europe, beating metropolises such as London and Berlin. According to a study by Atomico, “on a per-capita basis, Stockholm is the second most prolific tech hub globally, with 6.3 billion-dollar companies per million people compared to Silicon Valley with 6.9”.

No comments:

Post a Comment