Month: October 2019

The Fall 2019 issue of Daedalus is on the subject \”Improving Teaching: Strengthening the College Learning Experience,\” edited by Sandy Baum and Michael S. McPherson. There\’s a lot to digest in the issue, and I\’ll list the table of contents below. But I found myself especially interested by the comments on online education in \”The Human Factor: The Promise & Limits of Online Education,\” by Baum and McPherson, as well as in \”The Future of Undergraduate Education: Will Differences across Sectors Exacerbate Inequality?\” by  Daniel I. Greenstein.

It was now seven years ago, back in 2012, that companies like Coursera, Udacity, and edX announced their plans to revolutionize higher education with \”massive open online courses,\” or MOOCs. While the use of online tools has clearly spread, it seems fair to say that the revolution has not yet arrived. Where does online higher education stand at this point?

On the spread of online classes to this point, Baum and McPherson write (footnotes omitted):

But MOOCs, as attention-getting as they have been, have never been the main source of online education. For-profit, career-oriented institutions and large public universities have been the major providers at the undergraduate level, although several private nonprofit institutions now enroll thousands of online students. Today, more than 40 percent of all undergraduate students take at least one course that is offered purely online; 11 percent–including 12 percent of those in bachelor’s degree programs–study entirely online.

What\’s the evidence on how well online courses teach? A key difference here seems to be that hybrid courses with high online content can work well, but pure online courses have some problems.  Baum and McPherson:

But studies that focus on course completion rates as opposed to test scores generally show weaker outcomes when courses are entirely online.Moreover, recent randomized controlled trials of semester-long college courses have found lower test scores for students in fully online courses than for similar students in traditional classroom settings–but no significant difference in outcomes between those in settings that mix technology with classroom experience and students in fully face-to-face courses. Economist David Figlio and colleagues compared a fully online course to a classroom course; economists William Bowen and Ted Joyce each had teams comparing traditional courses to those replacing some live instructor time with online learning; and labor economist William Alpert and colleagues studied all three models.  The results of these studies are consistent. Classroom instruction time can be reduced without a negative impact on student learning. But eliminating the classroom and moving instruction entirely online appears to lead to lower course completion rates and worse outcomes, even when guidelines are followed for best practices for generating online discussion. The weaker results for students listening to lectures online instead of in a classroom with other students suggests that it may not be just personal attention, but being in a social environment that contributes to student learning. It is also possible that the more structured scheduling of classroom courses is important for some students.

The other big change in online higher education in the last decade or so has been a shift in who is most likely to be offering these courses. Back in 2009, it was mostly for-profits, but that has changed. Greenstein offers a comparison:

Unsurprisingly, by 2009, online instruction outside the for-profit sector was highly concentrated in a relatively small number of outlier institutions. In that year, Western Governor’s University (WGU), established in 1997 by the governors of nineteen states and with a significant grant from the Bill and Melinda Gates Foundation, offered fully online courses to over fifty thousand students, Penn State’s World Campus served twenty-five thousand (9,500 full-time equivalent) students, University of Maryland’s University College had twelve thousand online students, and there were one or two others operating outside the for-profit sector at something bigger than fledgling scale. There were also a number of headlining failures in the not-for-profit sector to point to, failures that reflected outright resistance to the genre, notably at the University of Illinois, where the Global Campus effort announced with great fanfare and with an investment of $10 million collapsed after only three years. By comparison, in the very same year–2009–the for-profit University of Phoenix was nearing its high watermark enrollment of nearly four hundred thousand online students.

Within a decade, the tables had turned. For-profits, under enormous pressure resulting from the Great Recession and a hostile regulatory environment, collapsed, losing as much as a half of all enrollments. Several of the biggest for-profits went out of business (Corinthian Colleges), were bought out by private equity firms (University of Phoenix), merged with not-for-profit institutions looking to accelerate their own online learning initiatives (Kaplan and Purdue Universities), or transitioned from for- to not-for-profit status. Large public universities and community colleges, meanwhile, moved in to pick up some of the slack. WGU grew to one hundred thousand enrollments and continues achieving 10 percent year-on-year growth. Arizona State University serves nearly the same number annually, and the University of Central Florida has grown to nearly sixty thousand students with almost one-third of all student credit hours taken online. Other evidence collected annually since 2002 has demonstrated how online learning has become part of the mainstream in higher education. Large public universities and colleges are particularly likely to offer a large share of student credit hours online. 

One of the hopes of online higher education was that it would be a low-cost way to make college classes widely available to underserved and at-risk student populations. This hope has gone largely unfulfilled. Baum and McPherson:

Two rigorous large-scale studies of community college students by the Community College Research Center (CCRC) found lower course persistence and program completion among students in online classes. These studies found that students who take online classes do worse in subsequent courses and are more likely than others not only to fail to complete these courses, but also to drop out of school.Males, students with lower prior GPAs, and Black students have particular difficulty adjusting to online learning. The performance gaps that exist for these subgroups in face-to-face courses become even more pronounced in online courses.

According to the CCRC, the differences are even greater for developmental courses than for college-level courses. In a study of online developmental English courses, failure and withdrawal rates were more than twice as high as in face-to-face classes. Students who took developmental courses online were also significantly less likely to enroll in college-level gatekeeper math and English courses. Of students who did enroll in gatekeeper courses, those who had taken a developmental education course online were far less likely to pass than students who had taken it face-to-face.

Thus, many of the current successes of online learning in higher education are for students who are pre-screened for  high admissions standard, and or highly motivated, or both. As one example, Baum and McPherson write:

Georgia Tech’s widely cited computer science master’s degree program is getting very positive reviews and appears to be opening opportunities to new students, rather than diverting them from face-to-face programs. Since this is a graduate program, all of the students have already earned bachelor’s degrees and, in the case of Georgia Tech, passed rigorous admission standards. Evidence about success in MOOCs confirms the reality that students from higher-income and more-educated backgrounds are most likely to participate and succeed in these courses.

Greenstein offers some other examples:

Two potentially very promising trajectories are beginning to take shape. The first is the use of hybrid modalities: modalities that mix face-to-face and online instruction. Where implemented well, they appear to lower costs and improve student outcomes. This at least is the experience at the University of Central Florida (UCF). With undergraduates taking nearly one-third of their credits online, UCF shows the best course outcomes for students in hybrid courses (with outcomes for face-to-face and fully online falling behind in that order). A second very promising development is seen in adaptive technology platforms and courseware that integrate data science to make machine-assisted learning directly responsive to individual students’ needs and their progress and pace in mastering explicitly specified course competencies. By the mid-2010s, results were more rather than less promising for the technology demonstrating improved student outcomes for students from all demographic groups.

I\’ve heard enthusiasts for online education point out more than once that the possibilities for innovative technological progress in the form  of a human delivering a live lecture are somewhat limited. In contrast, it\’s easy to imagine all kinds of potential for improvement in online higher education. It remains true that most online higher-ed involves lecture-based presentations followed with online quizzes and tests. One can easily imagine over time that the interaction of an online class with a student will become more adaptive, flexible, and responsive. The methods of group participation online with other students and faculty will become more sophisticated. But after some years of watching online classes not cause a revolution in higher education, some hard questions are emerging.

1) It\’s easy to imagine online higher education getting better, but it\’s not going to happen easirly or on the cheap. It\’s clear at this point that just recording some classroom lectures and linking up students to a multiple-choice online test-bank will work for a highly motivated few, but not for the many. The investment needed for really good online courses may be large, and it may be ongoing. The old model of finding a professor who teaches a course well, and then having the professor record some lectures or write a textbook, isn\’t going to suffice. Instead, there will be a need for experts in computer programming, psychology, artificial intelligence, and more. A highly-evolved version of an online education class is also not a one-time project, but instead is going to require ongoing cycles of learning, and respecting differences across topics. Teaching statistics online may look very different from teaching a foreign language or writing or chemistry or economics. Before we\’re too quick to assume that online higher education will soon and quickly get a lot better,, it\’s important to remember that creating the highly evolved online education courses of the future isn\’t just a matter of jumping a few hurdles, but of overcoming a multidimensional obstacle course. It\’s not about a few incremental gains to the existing courses, but of evolution into a different kind of online experience that barely exists–or may not yet exist.

2) Who is going to make these costly, risky investments? Maybe it will be a few very well-to-do schools. It would be an interesting irony of those who attend huge-endowment highly-selective schools also ended up with access to much better online courses! Another possibility is that it will be schools with extremely large enrollments–probably larger than the enrollment of any specific campus.  It would be interesting to see if some conferences, like the Big 10, SEC, Pac-10 or the ACC could put together a team along these lines. It\’s not at all clear how community colleges, smaller schools, or  schools with lower levels of funding can afford to make large and ongoing investments in a dramatically better version of online education.  Thus, it\’s not at all clear that these online courses of the future will be focused on at-risk or nontraditional students.

3) There are times when the discussion of online education seems to be based on a vision of education as something that can be downloaded or viewed online by individuals in isolation, who then absorb the necessary information. But most education has traditionally happened in groups, and the social and emotional structures of the group may matter–at least for most learners most of the time.  Thus, a challenge is to make online learning into a genuinely shared experience. I\’ll give Baum and McPherson the last word:

Behind the successive would-be revolutions in the technology of delivering college education seems to lie a desire to minimize, if not eliminate, the need for messy, often inconvenient, and always costly human interaction in the college-going experience. This desire is particularly evident when the concern is for mass higher education. A purely automated delivery system for much of higher education would appear to be very cheap and efficient, and perhaps even higher quality than traditional higher education because everyone could be exposed to the best lecturers. Unfortunately for this dream, developments in psychology and learning theory over the last two decades have made ever more clear how central the social, emotional, and interactional dimensions of learning are.

Here\’s the Table of Contents for the issue, with links to the articles:

• \”Becoming a Better College Teacher (If You’re Lucky),\” by  Harry Brighouse
• \”Catalysts for Achieving Sustained Improvement in the Quality of Undergraduate STEM Education,\” by Mary Sue Coleman, Tobin Smith, and Emily R. Miller
• \”Expertise in University Teaching & the Implications for Teaching Effectiveness, Evaluation & Training,\” by Carl E. Wieman
• \”Together and Alone? The Challenge of Talking about Racism on Campus,\” by Beverly Daniel Tatum
• “Now Is the Time”: Civic Learning for a Strong Democracy,\”  by Sylvia Hurtado
• \”The Future of Undergraduate Education: Will Differences across Sectors Exacerbate Inequality?\” by  Daniel I. Greenstein
• \”CREATE a Revolution in Undergraduates’ Understanding of Science: Teach through Close Analysis of Scientific Literature,\” by Sally G. Hoskins
• \”The False Dichotomy between Academic Learning & Occupational Skills,\”  by Thomas Bailey and Clive Belfield
• \”Mitigating Ethical Costs in the Classroom,\” by Jennifer M. Morton
• \”Financial Constraints & Collegiate Student Learning: A Behavioral Economics Perspective,\” by  Benjamin Castleman and Katharine Meyer
• \”Toward a 2.0 Compact for the Liberal Arts,\” by Earl Lewis
• \”The Human Factor: The Promise & Limits of Online Education,\” by Sandy Baum and Michael S. McPherson

Here\’s a description of the Arab \”social contract\” and \”development model\” according to a recent report by Adel Abdellatif, Paola Pagliani, and Ellen Hsu, \”Leaving No One Behind Towards Inclusive Citizenship in Arab Countries\” (July 2019). It an Arab Human Development Report Research Paper, written for the Regional Bureau for Arab States in the UN Development Programme. They write:

The social contract that emerged from and continues to evolve as a result of contesting and bargaining stemmed from the state-building and formation after Arab states won their independence in the 1950s–1970s. The emergence of independent states was associated with a strong nationalistic sentiment and the idea that the state should be the provider and engine of social and economic development. Despite considerable variation across countries, which was affected by natural resources endowments, the dominant model of development from the 1950s onward was having limited political participation and civil and political liberties in exchange for material benefits such as services, subsidies and employment. The model was based on strong central states overseeing and driving economic and social priorities while implementing wide-scale policies for redistribution and equity. It rested on four main pillars:

• Establishing a large bureaucracy to provide and deliver services.
• Expanding security services and the army.
• Setting up a large public sector of factories and companies.
• Subsidizing basic foodstuffs and energy products.

In  keeping with the development model, a large share of total employment–often more than 20%–is in the public sector.

This model can claim some successes. For example, life expectancy at birth in Arab countries was about 55 years in 1970, below the world average of 58 years. Now, life expectancy in Arab countries is about 76 years, above the world average of 73 years. However, human development gains for countries in the Arab world have generally slipped back since 2010. The big drop in global oil prices back around 2014 has meant a reduction in resources for oil-exporting countries in the Middle East, and lower spillover buying power for the non-oil exporters in the region.

The emphasis of the report is that many of the countries in the region do not have \”inclusive citizenship. For example, females in Arab countries lag males by more than the usual average for emerging-market economies in areas like education and political representation. The report notes:

The greatest measurable disparities are economic: globally women’s income is 57 percent of men’s, but Arab women’s income is only 21 percent of Arab men’s. Unequal gendered division of labour—both in unpaid care and domestic work and in the labour market—is a major characteristic of gender economic inequality across the Arab region. Women’s participation in the formal labour market remains among the lowest globally because of both cultural norms and weak incentives

There are big gaps between rural and urban areas, and within urban areas, \”[i]n at least seven countries with data, more than half of the urban population lives in slums.\” The concerns just keep coming:

Unaccountable and unresponsive public institutions as well as perceived widespread corruption often drive exclusion and disenfranchisement for large segments of the population. … A substantial number of citizens believe that the institutions meant to take care of their needs are leaving them behind … Trust in elected bodies, those that should be in charge of redesigning the social contract, is particularly low. Lack of trust is also reflected in low electoral turnouts—below 50 percent in most countries … Perceptions of ineffective institutions seem confirmed by stagnating or narrowly based economic structures, high unemployment, young people facing difficult prospects to secure their future and uneven provision of social services and social protection nets. Unemployment, averaging 10 percent, almost double the world average, disproportionately affects young people, at 25 percent. … 84% of the population is affected by or at risk of water scarcity. The decline of arable land and the dependency on food imports expose the population to risks of food insecurity …

Part of what makes the report interesting is that it is from Regional Bureau for Arab States in the UN Development Programme. And the unmistakable theme is that the Arab development model and the associated social contract isn\’t working very well. Another part of what makes the report interesting is its hesitancy about suggesting alaternative policy directions.

Yes, there\’s some discussion about how subsidies for energy prices that end up mainly benefiting the well-to-do, who after all use more energy, could be converted to to support for the poor. This is a problem in a lot of countries (for an overview, see this IMF working paper). But the challenges facing the Arab development model aren\’t about recalibrating some subsidies. The problem is that a \”development model\” based on high public employment, along with lots of social services and subsidies, needs substantial numbers of firms in a solid underlying economy to provide jobs and tax revenues and growth. 

For some earlier posts on the economic outlook for the Middle East, see:

• \”Middle East Economic Challenges (as the Role of Fossil Fuels Declines)\” (September 8, 201
• \”Youth and the Economic Future of Arab States\” (December 1, 2016)
• \”Next Steps for Economies of the Middle East\” (March 11, 2013)
• \”Economic Underpinnings of Arab Spring\” (January 27, 2012)
• \”The Middle East: Slow Growth and Stagnant Job Creation\” (June 8, 2011)

If you think about an economy as fairly stable and static, you would expect that any two companies within an industry would be fairly close in terms of productivity. After all, if Company A and Company B are selling similar products, and A has much higher productivity than B, it should drive B out of business. Thus, one might expect that at the end of this process, the competitors we observe within an industry in the real world should be fairly close in productivity level.

However, this expectation is dramatically wrong. Within an industry, it is a standard pattern to find a wide dispersion of productivity across firms in the industry. Academic researchers have been familiar with this pattern for at least 15 years. But now (pulse rate accelerates) there is systematic time series data across industries from 1997-2015!  \”The Dispersion Statistics on Productivity (DiSP) is a joint experimental data product from the U.S. Bureau of Labor Statistics and the U.S. Census Bureau. The DiSP provide statistics on within-industry dispersion in productivity.\”

For example, here\’s a figure from Cheryl Grim of the US Census Bureau. The bar graphs show that if you take a firm in the 75th percentile of the shoe or the cement industry and compare it with a firm in the 25th percentile of the shoe or cement industry, the firm in the 75th percentile will be about 150% as productive. In the computer industry, a firm in the 75th percentile is 400% more productive than a firm in the 25th percentile.

The existence of such differences in productivity across industry have been known for some time.   Cindy Cunningham, Lucia Foster, Cheryl Grimm John Haltiwanger, Sabrina Wulff Pabilonia, Jay Stewart, and Zoltan Wolf explain in \”Dispersion in Dispersion: Measuring Establishment-Level Differences in Productivity\” (Center for Economic Studies Working Paper CES 18-25R, September 2019).

They point out that research by Chad Syverson back in 2004, looking at data from manufacturing industries in 1977, found that firms in the 90th percentile of a certain industry were about four times as productive as firms in the 10th percentile. In the more recent data: \”Illustrating the properties of the new data product, we find large within-industry dispersion in labor productivity: establishments at the 75th percentile are about 2.4 times as productive as those at the 25th percentile on average.

Why do such differences exist? The reasons are obvious enough, as Grim explains?

Producers within industries differ in many ways. They produce different products of varying quality and have different customers and markets. They use different technology and business practices to combine different amounts of materials and equipment to produce their products. Some businesses are also larger and/or older than other businesses. Their ability to adjust their scale and mix of operations may vary due to these differences. Experimenting with new products and processes can also contribute to productivity differences. Businesses that have successfully adopted new technologies are likely to be more “productive” (as measured by these differences in revenue per hour) compared to businesses that have not yet adopted such technologies. All of these factors can contribute to enormous variations in this measure of business performance.

The fact that firms in the same industry be so different in productivity levels, and that these differences don\’t seem to fade away, has a number of interesting implications.

First, the pattern suggests that productivity growth doesn\’t always mean cutting-edge gains; indeed there is enormous potential for economic growth if the firms now lagging in productivity can be brought up to speed, perhaps by merging with higher productivity firms. In addition, one way that productivity growth happens for the economy as a whole is when high-productivity firms put low-productivity firms out of business.

Second, the persistence of these gaps suggests that some firms are protected from competition. For example, cement is not very transportable, and so competition in the cement industry is often limited to local firms. The potential reason why productivity differences may persist in other firms is worth considering.

Third, there seems to be some evidence that productivity diffusion is widening, as \”superstar\” firms in various industries pull further ahead. Indeed, this may be an important factor contributing to growth of inequality of wages, because workers and managers at high-productivity firms are typically much better-paid than those at low-productivity firms. 

CDOs, or \”collateralized debt obligations,\” were at the heart of what broke down in the US financial system and helped put the \”Great\” in the \”Great Recession.\” Is there another financial instrument out there that raises similar concerns? CLOs, or \”collateralized loan obligations,\” have a similar structure and have now reached a similar size to the CDOs circa 2008.

 How much should we be worried? As I\’ve noted in past discussions of the subject, several Fed officials including  Lael Brainerd of the Fed Board of Governors and Robert Kaplan of the Federal Reserve Bank of Dallas (who will rotate on to the membership of the Federal Open Market Committee in 2020) have raised concerns.  Sirio Aramonte and Fernando Avalos offer a nice short discussion of this comparison in \”Structured finance then and now: a comparison of CDOs and CLOs,\” which appears in the BIS Quarterly Review (September 2019, pp. 11-14). They write: \”The rapid growth of leveraged finance and CLOs has parallels with developments in the US subprime mortgage market and CDOs during the run-up to the GFC. We examine the CLO market in light of that earlier experience.\”

Here\’s some backstory. The collateralized debt obligation of concern back in 2007 were a set of financial securities that were based on pools of subprime mortgages. There\’s nothing wrong with collecting mortgages into a pool, packaging them into a security, and then reselling them to investors like insurance companies, pension funds, hedge funds, and banks.

But the problem with creating a financial security based on subprime mortgages was that–by the definition of \”subprime\”–a relatively high percentage of these mortgage were going to default, so a financial security based on these subprime mortgages would be fairly risky. For example, banks would not be allowed by regulators to hold such securities. However, some financial wizardry solved that problem.  The CDOs were divided up into sections, called \”tranches,\” with some of the tranches being very risky and some being very safe. For example, if losses on the underlying subprime mortgages were in the range of 0-10%, then all of those losses would fall on one set of investors in the highest-risk tranche. If losses on the underlying subprime mortgages fell in the range of 10-20%, then those losses would fall entirely on another set of investors in the next highest-risk tranche. With several of these tiers built into place, so that any losses would be concentrates on a subset of investors, the other tranches of the CDO appeared to be very safe: indeed, those tranches were rated AAA and banks were allowed to hold them.

The current wave of collateralized loan obligations are also financial securities based on pools of debt–but in this case, the debts are corporate loans rather than subprime mortgages. Again, there\’;s nothing wrong with collecting debt into a pool, packaging it into a security, and reselling it to investors. This kind of corporate debt is called  \”leveraged loan.\” As Aramonte and Avalos write:

CDOs and CLOs are asset-backed securities (ABS) that invest in pools of illiquid assets and convert them into marketable securities. They are structured in tranches, each with claims of different seniority over the cash flows from the underlying assets. The most junior or so-called equity tranche is often unrated and earns the highest yields, but is the first to absorb credit losses. The most senior tranche, which is often rated AAA, receives the lowest yields but is the last to absorb losses. In between are mezzanine tranches, usually rated from BB to AA, which start to absorb credit losses once the equity tranche is wiped out. The larger the share of junior tranches in the capital structure of the pool, the more protected the senior tranche (for a given level of portfolio credit risk).

For both CDOs and CLOs, strong investor demand led to a deterioration in underwriting standards. For example, US subprime mortgages without full documentation of borrowers’ income increased from about 28% in 2001 to more than 50% in 2006. Likewise, leveraged loans without maintenance covenants increased from 20% in 2012 to 80% in 2018. In recent years, the share of low-rated (B–) leveraged loans in CLOs has nearly doubled to 18%, and the debt-to-earnings ratio of leveraged borrowers has risen steadily. Weak underwriting standards can reduce the likelihood of defaults in the short run but increase the potential credit losses when a default eventually occurs.