The Race Between Education and Technology [Tapa blanda]

Three stars? This book is undervalued by the current amazon reviews. In my opinion, a majority of books about education are downright bad. Of those that are good, many are not uniquely good - you can find the same information in other popular books on education. This book is good and unique.

This book is excellent on the links between education, productivity, and income inequality over the last century. It makes the case that increasing human capital through education is a very important goal - the claim is not original but the way they argue it is.

The policy recommendations are less well backed up.

Google for an excellent commentary written by Arnold Kling & John Merrifield, entitled "Goldin and Katz and Educational Policy Failures in Historical Perspective."

I found this book fascinating and would recommend it although I found it frustratingly flawed, and therefore, without the authors' further comments, will eventually have to reduce my 4 star rating. The book's core thesis is that the rate of technological change in the 20th century has been constant, while the supply of skilled workers has been uneven, leading to expanding and contracting wage differentials between skilled and unskilled workers over the course of the century. The variation in supply and correlation to changes in relative wages seems large enough to be convincing. But while I find the case persuasive, I couldn't help but feel it was, in part, reserve engineered to reach its conclusion. It seemed to me that three critical issues remained either unwittingly or intentionally overlooked.

In part, large part perhaps, I would guess, the "skill" of a worker transcends their education. So a high school drop out today, when most everyone graduates from high school, represents a much less skilled worker than a drop out at the turn of the century, when very few graduated. To suggest the ratio of wages between high school graduates and drop outs today can be compared to the past without some adjustment, or even mention of an adjustment, that take this into consideration, seems lacking. The same is true of college graduates, where the meaning of the term has been averaged down. The analysis seems to suggest that a college or high school graduate is equivalent no matter what (changing) percentile of the population it encompasses or, more complicated mathematically, that the relative curve across percentiles is such that any point is logarithmically proportional to any other. While perhaps this is a second order issue on the margin, over the large shifts that have occurred in the education of various percentiles of workers over the course of the 20th century, leaving it unmentioned is disappointing.

While the mathematical logic of relative wages vs. relative supply (of skilled vs. unskilled workers) seems reasonable on the surface, it's not clear why expansion in the supply or demand of one would affect the ratio between them proportionally. It would seem, for example that the US has absorbed a large numbers of unskilled immigrants, it has offshored an unprecedented volume of "unskilled" goods, and it has accelerated productivity improvements and yet the unskilled wage has remained flat. That suggests that the marginal product of unskilled labor is largely flat over large shifts in volume. To argue that the expansion or contraction of supply drives (relative) wages up or down requires a downward sloping marginal product of labor, unskilled labor in this example. I don't see why that would necessarily be the case to the degree necessary to explain the data. Quite the contrary, it seems unlikely based on the prevailing flat wages. In that case, an expansion of unskilled workers would have no effect on the wages of unskilled or skilled workers and their subsequent ratio, whereas, in this example, the expansion of skilled workers could have such an effect. To assume they are wholly substitutable seems to be a stretch.

Similarly, the book overlooks both productivity issues and hours worked in analyzing the supply of labor. We know the most productive workers are now working longer hours than the rest of the working population. And it's hard to believe that the information age hasn't, to a very large extent, affected the productivity of the most productive workers. This would suggest that their supply has expanded far more than their headcount. If that's true, then if supply has expanded and wages have expanded, exogenous technological breakthroughs are likely driving up wages and not merely reduced relative headcount as the book argues.

Piecing these together, it seems easy to imagine significant changes in the relative productivity and resulting supply of skilled vs. unskilled labor, along with differing marginal product of labor, highly circumstantial substitutability, and varying rates of exogenous technological changes all significantly affecting these ratios in ways not considered by this analysis. Perhaps the authors' argument is better stated this way: if one assumes that the rate of technological change is constant, and that relative productivity changes between skilled and unskilled are similar (although the book argues that in the past that was not the case), and that substitutability is generally high, then relative headcount should drive relative wages; because it appears to be highly correlated, these assumptions are likely to be true. That's a pretty big leap of faith. And, if the authors are correct that it is merely a matter of relative supply, it's not at all clear to me why diminished relative supply of skilled workers has led to an increase in the percentage of overall income paid to skilled workers, as has been the case. Apparently there was previously far more of them than was optimal! I find that hard to believe and apparently so too the authors. And the recent 10 to 15 year growth in output per US worker relative to Europe and Japan, despite unfavorable demographic shifts in the US workforce (toward less educated Hispanic immigrants), seems greatly at odds with the book's analysis.

Lastly, the book is very weak on recommendations and in many cases what recommendations it does offer, a steeper income tax for example, hardly stem from the 300 pages of analysis that comes before. In another example, it suggests that testing is bad because it may increase drop out rates. But, I sit on the board of a charter school that today has outstanding test scores where previous testing and low scores put enormous pressure on the school to make positive changes, changes that were made because of the demand for higher scores. Again, there is little in the analysis to warrant their suggestion that more forgiving testing would be a net positive. I find the simplistic solutions largely unsupported by their analysis to be disappointing.

This book epitomizes everything that is wrong with social science. The modus operandi is to pull together a series of charts showing correlations, assume that the correlation is due to causation, and ignore any discussion of alternative explanations of the trends.

Goldin-Katz spend the bulk of the book hammering away on two points that everyone already knows: years of schooling on a national level correlates with industrialization, and years of schooling on a personal level correlates with income. Goldin-Katz spend precious few pages actually dealing with the causation issue, and never address any of the best arguments against their thesis. Nor is there any attempt to actually talk to people working in technology in order to understand more deeply why the correlation exists.

Let's examine in detail some of the flaws.

a) Goldin-Katz's base hypothesis is that years of schooling should continuously rise over time, as technology increases. But the very definition of technology is that you get more output for a given amount of input. Thus we should not expect a proportional increase in education to take advantage of new technology. Indeed, this is what we see on the ground. As a programmer in 2009, I no longer need to learn a huge amount of information that my father needed to know. For my job, I do not need to assembly language, register hacks, memory allocation, pointer arithmetic, etc.

b) Goldin-Katz's hypothesis is at odds with the experience of all the recent college graduates I know. No one believes that education teaches job skills. A quick check of the top 10 most popular college majors shows that these majors have little to do with technology. Clearly if there is an income bonus from college education, it cannot be from teaching technology, because colleges do not actually teach technology.

c) Goldin-Katz's hypothesis is at odds with the life experience of most engineers I know. If you ask the typical, engineer, "How many years would it take, starting from the beginning of high school, and working efficiently, to reach an amount of knowledge where you could be a productively employed?" the answer is usually something like 1 to 3 years. If you look at the actual skills to do high tech jobs, you simply notice that very few require 8 years of full time schooling. You'll also notice that engineers universally deride schooling, and that they learn most of their skills by avoiding school work (this is especially true in high school). For more details Google the essay "Why nerds are unpopular" by Paul Graham.

d) The standard government economic growth statistics have so many methodological problems that's it's impossible to draw any conclusions from them (for more details, Google "Economics needs a divorce" ). It's unclear both a) that growth has actually been declining and b) that the decline has to do with lack of technological innovation ( it might have a lot more to do with the increasing portion of the GDP taken up by bureaucratic sectors that are impervious to technological change - like the education sector itself!). Chinese, Japanese, and Korean mercantilism have also played a great role in the decline of America's technological-industrial base. Never do Goldin-Katz address either of these points.

e) I do agree that 19th Century America derived great benefits from its strong primary schools and high literacy rates. But I believe this is primarily a threshold effect. After students have the tools to find books and self-educate, further formal schooling has diminishing returns. So I might agree that 19th century America derived an advantage from averaging something like 5 years of schooling rather than the 0-2 years of schooling that was common in other countries. But it does not follow that modern America would derive an advantage from raising the average years of schooling from 13 to 15. In fact, 13 is almost certainly above the point where opportunity costs exceed returns to schooling.

Goldin-Katz never address any of the competing explanations for the correlation between industrialization and education or income and schooling:

a) Richer countries can afford more years of schooling. The experience of my peers and I in college is that college is primarily a luxury good.

b) Academics have greatly increased their influence on politics in the past century, first with the Wilson administration then with FDR's brain trust. Prior to 1900 academics had neither involvement with politics nor control over policies. Today, virtually all major policy advisers are academics. Not coincidentally, there has been a concurrent increase in government money spent on schooling and on total years of schooling. Thus part of the rise in education over the last century was likely simply two unrelated but concurrent events - the continuing industrial revolution, and the increasing political power of the academic class.

c) On an individual level, selection effects plays a major role in creating the link between college and income. Completing college requires a threshold level of intelligence and diligence. Colleges select for people with high earning potential, because such people are more likely to make money, and donate it back to the school. I was talking to my friend who does hiring for Bain Consulting: "Bain likes to recruit econ majors from top schools, but because they learn anything valuable in the major, but because it means the person is smart and care about business." I hire programmers at a startup, and I care little about the degree, and a lot more about how smart the person is and what they have done. This does correlate with college and major, but the actual knowledge gained in the college major is a tiny portion of what is needed to be a successful engineer.

The selection/signaling effect is even more important considering the that the 1971 Griggs Supreme Court case made it illegal for employers to use IQ tests for hiring purposes. As a result, companies have to rely more on educational attainment as a proxy for IQ.

d) Finally, and perhaps most importantly, Goldin-Katz completely ignore the impact of credentialing laws. There are now legal degree laws for professions such as: lawyers, architects, doctors, teachers, civil servants, military officers, nurses, and education administrators. These professions receive relatively high salaries because they have either direct government subsidies, or they have monopoly rights to perform certain tasks (prescribe medicine, defend the accused in court, etc). Yet there is no evidence that requiring a degree is a credential is a greater indicator of ability than simply using a test or requiring apprenticeship. Most architects of the 19th century learned via apprenticeship, yet the quality of the buildings was much higher back then than today.

Searches of the book for "signaling", "credentials", "credentialing", "Spence" return zero hits. To write a book about the school about the education wage premium and not discuss these issues is completely egregious. In a just world this failure alone would be enough to ruin the reputation of Goldin and Katz as being serious scholars and to impugn the reputations of the academics who offered such fawning reviews.

Goldin-Katz's book is fundamentally about policy. It is about how to manage a countries economy to maximize technological growth. You would think that the first thing that anyone would do when writing such a book, is to talk to dozens of people in high technology. You would talk to engineers, entrepreneurs, workers at high tech firms, current college students, recent college graduates. Yet Goldin Katz do none of this. They sit in their ivory tower, plot some regressions and engage in chart-ism of the worst sort. Their statistics add nothing to the stock of knowledge that already exists about the correlations between education and income. And they ignore addressing all the possible arguments against their case. This book is only interesting the way that a car wreck is interesting.