Mariana Mortágua on why crypto-assets are not money

Economists typically identify money by the extent to which an asset satisfies three functions. First, it needs to be a medium of exchange, meaning that you give it up when you buy goods or services, and you can receive it when you sell goods or services. Second, it needs to be a unit of account, meaning that you can measure the value of something using the amount of money it is worth. And third, it needs to be a store of value, meaning that you can keep it and expect it to retain a reasonable proportion of its purchasing power into the future. These functions of money go back to William Stanley Jevons in 1875.[*]

There has been a lot of debate about whether crypto-assets such as Bitcoin or stablecoins are money (see the posts linked at the bottom of this post for some examples focused on Bitcoin). This recent article by Mariana Mortágua (ISCTE-University Institute of Lisbon), published in the Cambridge Journal of Economics (sorry, I don't see an ungated version online) wades into the debate, broadening the scope to consider crypto-assets more generally.

Mortágua concludes that crypto-assets are not money. However, unlike much of the debate, she doesn't focus on the volatility of crypto-assets and how that impedes the 'store of value' function. Instead, she adds to the argument that crypto-assets are not a unit of account. Specifically:

Reserves, cash, and bank deposits are titles of debt, claims denominated in one unit of account. Although the issuer of such a claim is not always a State entity (like a central bank), privately issued money is apparently indistinguishable from central bank money because it is issued using the same denomination (the generally accepted unit of account)...

While claiming that cryptocurrencies, like Bitcoin, can become money, their enthusiasts fail to explain how a system of bilateral transactions in one medium of exchange would give rise to a unique unit of account or why people should accept that medium of exchange in the first place...

Indeed, though cryptocurrencies have their own self-attributed trademarks, their prices are denominated in one already established unit of account. Even the most known and utilised cryptocurrency, Bitcoin, is very far from serving as a measure of value of services or assets. On the contrary, Bitcoin is priced in USD, and... only central (political) authorities have the power to determine the ‘meaning of a dollar’.

So, because we don't measure the value of things in terms of crypto-assets, they aren't money. It isn't enough that you can buy or sell things using crypto-assets like Bitcoin. That is an example of a crypto-asset serving the function of a medium of exchange. The transactions are priced in dollars (or other currency), which means that the unit of account is not the crypto-asset, but the currency.

The argument that cryptocurrencies like Bitcoin are not a unit of account is not new (see the links at the end of this post for more). What Mortágua adds is a broader theoretical argument about the nature of money itself, based on the idea that money represents claims denominated in an established unit of account (like dollars). She argues that this distinction is important not just for Bitcoin, but for crypto-assets more generally.

There is, however, an interesting complication here. The dollars in your bank account don't have their own unit of account either. A bank deposit is a claim on the bank, denominated in dollars, but we still count bank deposits as money. Mortágua's argument rests on the credit theory of money, which sees money as part of a system of debts and claims, denominated in a commonly accepted unit of account. Bank deposits are widely accepted as money because the banking system allows them to be used to settle debts.

Stablecoins are not as straightforward. While a dollar-backed stablecoin may also represent a claim on its issuer for US dollars, the arrangements supporting that claim are different from those supporting bank deposits. So, the question is whether those stablecoins have the institutional backing and general acceptability of money. Mortágua argues that they don't.

So, it's not just Bitcoin. Mortágua concludes that crypto-assets in general are not money. However, stablecoins show us that the distinction between money and other financial assets is not always straightforward.

*****

[*] Jevons noted a fourth function, a standard of value (a way of valuing debts), which we now consider to be much the same as being a unit of account.

Read more:

• You may be able to buy Coke with Bitcoin, but that still doesn't make Bitcoin money
• No, El Salvador hasn't just turned Bitcoin into money
• Bitcoin still isn't money, because people aren't using it as if it was

Inequality is not robustly associated with lower subjective wellbeing or worse mental health

When I cover inequality with my ECONS102 class, we talk about some of the many problems that have been associated with inequality, such as unequal access to political representation, lower social cohesion, and less interaction between increasingly segregated social groups. I also talk about the theory that links higher inequality with lower economic growth, though noting that the research is far from settled on that issue (see here and here, for example).

In that class, I don't discuss the research linking inequality with lower physical or mental health. That goes back to my objections to the Wilkinson and Pickett book The Spirit Level (which I discussed here). It turns out that might have been a good choice, at least in terms of mental health.

This recent article by Nicolas Sommet (Université de Lausanne) and co-authors, published in the high-profile journal Nature (ungated earlier version here), presents the results of meta-analyses of the relationship between inequality and subjective wellbeing, and inequality and mental health. Meta-analysis is a way of combining the results of many studies to generate an estimate of the overall effect size.

Their meta-analysis of inequality and subjective wellbeing included 152 estimates from 100 different studies, and they found that inequality was not significantly associated with subjective wellbeing. Their results were qualitatively the same whether subjective wellbeing was measured as life satisfaction or happiness, and did not depend on study quality. That is worth noting, given that about 80 percent of the studies included in the two meta-analyses were judged to be at high or very high risk of bias.

Sommet et al.'s meta-analysis of inequality and mental health included 99 estimates from 75 different studies, and they found a small negative association between inequality and mental health. Specifically:

A 0.05-point increase in the Gini coefficient predicted a 4% (1/0.961) increase in the odds of experiencing a mental-health issue.

The Gini coefficient measures inequality on a scale of 0 to 1, and a 0.05-point increase is similar to the difference between New Zealand (0.326, above the OECD average) and Denmark (0.276, close to the lowest inequality among OECD countries - see Figure 6.1 in this report). Similar to subjective wellbeing, the results were qualitatively the same whether mental health was measured as depression or 'other mental disorders', and did not depend on study quality. The effect was larger when inequality was measured at the 'macro level', meaning the country level (or large subnational areas with populations over one million), although the effect was also statistically significant when inequality was measured at the 'meso level' (small subnational areas, such as counties).

Not so fast, though. After correcting for publication bias (the tendency for statistically significant results to be more likely to be published), Sommet et al. find that, for mental health:

The median bias-corrected estimate was OR+0.05 = 1.018 (0.986–1.051), P = 0.266, thereby confirming the lack of a significant association after correcting for publication bias.

Moreover, Sommet et al. conduct a specification-curve analysis, which tests how sensitive the results are to different modelling choices. They find that their conclusions are robust across approximately 95 percent of 768 alternative model specifications.

The overall result is that inequality is not statistically significantly associated with either subjective wellbeing or mental health, at least not robustly so. However, Sommet et al. then dig a little further, finding some interesting differences in the estimated effects across countries with different characteristics. First, for mental health, they find that the negative association is significant in lower-income samples, but not in higher-income samples. They replicate this result using data from the Gallup World Poll, using average national income to distinguish between lower- and higher-income countries. They explain this as suggesting that:

...inequality may be particularly harmful to low-income populations, possibly by undermining community cohesion... fostering adverse social comparison... or fuelling perceptions of unfairness...

Second, for subjective wellbeing they find that there is a negative association with inequality in high-inflation countries, but a positive association with inequality in low-inflation countries. They also replicate this finding using data from the Gallup World Poll. They explain this in the following way:

Rising costs of living drive distress... and inequality may exacerbate the burden of inflation by increasing social discontent and resentment. Conversely, when prices are stable, inequality might be interpreted as a signal of future economic mobility and foster hope... That is, inequality may only harm well-being under degraded economic conditions (here, high inflation) and have positive effects under others (low inflation).

Their explanations are speculative, and I would be cautious about over-interpreting the mechanisms underlying these latter results. Nevertheless, the fact that both findings were replicated using the Gallup World Poll should provide some comfort about the results. Overall, the results of this meta-analysis can be taken to mean that, while there may be a relationship among some subgroups or some countries at some points in time, there is not a robust overall association between inequality and lower subjective wellbeing or worse mental health.

[HT: Marginal Revolution, last year]

Two papers on inequality in pre-industrial times, that extend our view over two millennia

Back in 2015, I wrote a post about some research by Branko Milanovic (City University of New York) on inequality in ancient times. Milanovic is a prolific inequality researcher, but I only recently got time to read this 2019 article of his, published in the journal Review of Income and Wealth (ungated version here). The article is something of a sequel to the earlier article, tracking changes in income inequality from 14 to 700 C.E. in Europe and the Mediterranean.

This involves combining data from a variety of other studies, first to estimate changes in mean income, then overall and regional inequality, before turning to individual inequality. The resulting estimates are summarised in Figure 2 from the paper, which shows upper and lower bounds on the estimated Gini index:

The 'social table' referred to in Figure 2 comes from this earlier paper by Scheidel and Friesen, with an estimated Gini index of 41.3. The decline from there to a Gini index of between 3 and 13 represents both an unprecedented decline, and an extraordinarily low level of inequality by 700 C.E. Milanovic notes the apparent disconnect between the low inequality he estimates and the impressions of contemporary accounts of poverty, and explains that:

We believe that it is possible to reconcile the conjectured low measured inequality with the impression of a high post-300 inequality because the latter may, in reality, refer to the speculation (or a fact) that the elite was appropriating proportionally much more of the surplus above subsistence than before... Those who observed poverty of most contrasted to the wealth of the very few tended to conclude that inequality must have unambiguously risen. But they might have failed to account for the fact that the numerous poor were now all equally impoverished, and that measured inequality (calculated across all individuals) could not, on that account alone, have been very high...

It is worth noting that these low estimates depend on an assumption that average incomes were barely above subsistence. When almost everyone is extremely poor, even a rich elite that appropriates most of the surplus above subsistence might not produce a particularly high overall Gini index. So, we should be careful that we don't mistake low inequality for prosperity, because it certainly isn't implying that. Nevertheless, the results remain somewhat extraordinary, and it will be interesting to see other estimates of inequality during that time, taking advantage of different data sources or methods.

Milanovic's paper finishes with 700 C.E. What happened to inequality after that? This 2021 article by Guido Alfani (Bocconi University), published in the Journal of Economic Literature (ungated earlier version here), picks up the story in 1300 C.E. and takes it through until the end of the 18th Century, synthesising estimates from a number of other studies of inequality in various parts of Europe. Alfani makes note of Milanovic's results (from the paper I noted above), but doesn't really critically appraise them, which is fine because Alfani is more interested in the period from the late Middle Ages to the early modern period. The studies he synthesises are summarised in Figure 2 Panel A from his paper, which again focuses on the measures of the Gini index, although note that this is for wealth inequality (rather than income inequality):

Alfani uses this to suggest that the evidence supports two stylised facts about inequality in this period: (1) that inequality in Europe increased from 1450 or 1500 until 1800; and (2) before 1450, there was a period of declining inequality that coincides with the Black Death. The decreasing inequality during the Black Death is consistent with an explanation that my ECONS102 students will be familiar with - agricultural wages increased dramatically during that period, alongside a decrease in land rents (as discussed in the book The Rise of the Western World by North and Thomas, which I reviewed here). Alfani then goes on to explore reasons that might explain the overall increase in inequality. He dismisses economic growth as the main explanation, noting that inequality appeared to increase even during periods of economic decline (aside from during the Black Death). Instead, Alfani points to the importance of institutions, and in particular the rise of the fiscal-military state, alongside generally regressive taxation, as well as inheritance systems that protected wealth within families. He concludes that:

The different, and sometimes partly contrasting, explanatory factors for long-term inequality change discussed above lead to a clear conclusion, which can be summarized as follows:

(i) In preindustrial times, there was no necessary cause of inequality growth (like for example, economic growth).

(ii) There were instead a number of sufficient causes of inequality growth, among which we could name economic growth, demographic factors (including urbanization growth), institutional change and politics, “proletarianization” (i.e., the crisis of small ownership), and so on. When one or more of these potential causes became active, inequality grew.

(iii) The bottom line being, in the long preindustrial period it was seemingly much easier for inequality to grow than to decline—and indeed, we have very few examples of long-lasting inequality decline, and almost all of them can be connected to large-scale catastrophes.

Finally, Alfani takes a longer view of wealth inequality through to modern times, by comparing his estimates (based on an average of the various studies included in his synthesis) to those of Thomas Piketty, which begin in 1800. The results are shown in Figure 8 in the paper, which shows the share of wealth held by the richest 10 percent, rather than the Gini index:

So, now we are getting a longer-term picture of nearly two millennia of changes in inequality in Europe, albeit with some important gaps. Milanovic's results suggest that income inequality declined dramatically after the fall of Rome, while Alfani's results suggest that wealth inequality declined around the time of the Black Death, before increasing through to the early 20th Century. In the 20th Century, wealth inequality saw another period of rapid decline (but nothing like the decline in income inequality after the fall of Rome), before increasing again in the second half of the century. Of course, income inequality and wealth inequality are not directly comparable, even if they tend to be correlated. So we should be cautious about drawing conclusions about the overall trajectories of both types of inequality. More studies are needed to confirm these long-run patterns, as well as to fill in the gaps and to extend these results to consider other regions outside of Europe.

Read more:

• Inequality was similar, or higher, in ancient times than it is today
• The last two centuries of global inequality
• Changing global inequality, migration and open borders
• More evidence of declining global inequality
• Chinese growth and global inequality
• Several research articles on inequality
• The global inequality boomerang

This week in research #148

Here's what caught my eye in research over the past week:

• Greenaway‐McGrevy and So (open access) find that upzoning reform in Auckland decreased rents by 23 percent eight years after the reform
• Hess and Thomas (open access) find that state-level recreational marijuana legalisation in the US was associated with near‐zero initial college enrolment growth in STEM majors as out-of-state enrolments increased but were offset by a decrease in in-state enrolment, but that total enrolment increased in later periods as out‐of‐state gains persisted and in‐state enrolment returned to baseline
• Stephenson finds that the 2015-2018 NFL drafts had negligible effects on room rentals and hotel revenue in the host city, while the 2019-2024 drafts (excluding 2020 and 2021) had statistically significant but small effects
• Hillman (with ungated earlier version here) finds no lasting or statistically significant effects of unilateral‐no‐fault divorce laws on divorce rates, using state-level data in the US

Non-market valuation, and the economic value of birdsong

When economists want to know the value of something, we typically look at the price it is traded for in a market. That works well when you want to know the value of an apple, a haircut, or a rare Pokémon card. It doesn't work at all for goods and services that are not traded in markets (like clean air) or those that are not traded because they don't yet exist (like vaccines for HIV). To value those goods and services, we need to apply non-market valuation.

I've previously written several posts about the contingent valuation method and its problems (most recently here), but it is one of several methods that are used for non-market valuation. A similar method is to run a discrete choice experiment, where survey participants are presented with a series of choices between different alternatives that have different characteristics, and are asked to choose the alternative they prefer. By getting participants to make a number of choices between different alternatives, researchers can establish how important each of the characteristics are to the participants. And if one of the characteristics is a price, then researchers can estimate how much participants are, on average, willing to pay for each characteristic.

One of the issues with the contingent valuation method is that it is vulnerable to context. For example, in this 2015 post I describe research that shows that research participants are less rational in their choices if they complete the task late at night. However, researchers may also be able to use context to improve their estimates of what people are willing to pay.

Consider this new article by Anthony Higney (University of Glasgow) and co-authors, published in the journal Ecological Economics (open access). They use a discrete choice experiment, conducted in an online sample of 1094 people from the UK, to measure willingness to pay for ecological restoration. Specifically, they presented each research participant with a series of choices about habitat restoration for birds, where the characteristics of each alternative included the habitat type (woodland, grassland, or arable farmland), the size of bird habitat (compared with baseline), and the quality of the habitat (in terms of the number of bird species), as well as a one-time donation amount. The inclusion of the donation amount means that Higney et al. can use the results to estimate the willingness-to-pay for each of the other characteristics.

Research participants were randomised so that half of the participants had access to a sound recording of birdsong for each of the alternatives (the treatment group), while the other half had no birdsong (the control group). Higney et al. essentially argue that listening to birdsong increases the salience of the habitat quality in the participants' decisions about which alternative to choose. They found that:

...the treatment of playing birdsong causes respondents to be less sensitive to costs in making their choices and rate quality changes from the status quo more highly.

Higney et al. can't tell whether their respondents actually listened to the birdsong or not. So, what their results really say is that respondents who had access to birdsong were willing to pay more for habitat restoration than respondents who did not have access. How much more? The willingness-to-pay for medium-quality habitat (compared with low-quality) was £276.5 higher for those with access to birdsong than those without access, while the willingness-to-pay for high-quality habitat (compared with low-quality) was £274.6 higher. That compares with a baseline willingness-to-pay in the control group of £99.5 for medium quality, and £86.0 for high quality. Interestingly, much of the difference arose because respondents who had access to birdsong were less sensitive to the donation amount, rather than simply placing greater value on improvements in habitat quality.

One reason for caution about the size of these estimates is that the highest donation amount participants were offered in the experiment was £200, which is smaller than the estimated willingness-to-pay for medium-quality and high-quality habitat among the treatment group. So, participants were never actually asked to consider donation amounts as high as their estimated willingness-to-pay. This is not an unusual outcome in this sort of hypothetical exercise, but some follow-up research using a wider range of donation amounts would be useful.

Overall, there was no significant difference in willingness-to-pay between the medium-quality and high-quality habitats (either for the treatment or the control group), and Higney et al. don't have a good explanation for why people weren't sensitive to the quality of habitat restoration beyond medium quality. That would also be a good avenue for future research.

Why would listening to birdsong change what people are willing to pay for habitat restoration? Higney et al. suggest that this could be due to framing:

Listening to the sound clips might have reduced the use of cognitive shortcuts in the treatment group. This could include negativity bias, i.e. giving disproportionate weight to negative information such as the payment vehicle...

Alternatively, respondents in the treatment group may consciously place less weight on other aspects of their choice because the sounds make them feel more connected to nature.

Framing does seem like an obvious explanation, and Higney et al. show in a follow-up regression analysis that participants who had access to birdsong reported significantly greater feelings of connection to nature. That is consistent with the addition of the sound clip meaningfully changing how the alternatives are interpreted. Higney et al. intentionally created sound clips to match the quality of the habitat they were asking about. It would have been interesting to run the same exercise, but with every sound clip being the same (or very similar) generic level of birdsong. Or to have two treatments: (1) generic birdsong; and (2) birdsong that conveys the quality of habitat. That may have helped tease out whether the participants were receiving better information about the alternatives by having access to the sound clip, or whether they were simply feeling more connected to nature after hearing the birdsong.

When valuing the benefits of ecological restoration, we can't rely on market prices and so non-market valuation techniques are really important. The way that we undertake those non-market valuation exercises is important, and understanding how framing affects people's decisions in discrete choice experiments is important for designing research that will give us the most accurate assessment of value. This research can't tell us whether playing birdsong to research participants gives us estimated willingness-to-pay that is closer to the 'true' value. But it does tell us that the information that research participants are presented with, and how that information is presented, matters.

Nobel Prize for Robert Hall and Michael Woodford

New Keynesian economics came into its own in the wake of the Global Financial Crisis, and has provides the playbook for central banks' monetary policy decisions. So, it was fitting to see New Keynesianism recognised in the latest Sveriges Riksbank Prize in Economic Sciences in Memory of Alfred Nobel (aka the Nobel Prize in Economics), awarded to Robert Hall (Stanford University) and Michael Woodford (Columbia University), "for microeconomic foundations of macroeconomic stabilisation".

The Nobel Prize Committee's citation noted:

Robert E. Hall and Michael Woodford are awarded this year’s prize in economic sciences for demonstrating how economic models could be reconstructed to revitalise stabilisation policy. Rather than basing the models on statistical links between macroeconomic variables, such as inflation, production and consumption, they chose to build them on a microeconomic foundation. This encompasses an understanding for how the total mass of individual households and businesses affect the economy through their actions.

Because I almost exclusively teach microeconomics, my students wouldn't necessarily recognise their work. However, the macroeconomics topics in ECONS101 include consumption functions as well as the role of expectations, both of which are key features of New Keynesian economics.

This year's award appears to have been very positively received, although as I noted in this post from last week, the prediction markets had different favourites in mind (as did I). Marginal Revolution has a series of posts (see here and here from Tyler Cowen, and here and here from Alex Tabarrok), and there is a good writeup in the Financial Times, especially noting the disconnect between Hall and Woodford's work and the practice of the current Fed Chair, Kevin Warsh.

As always, the Nobel Committee provided some excellent commentary in the form of a popular science summary of Hall and Woodford's work. That is well worth a read if you are not familiar with them.

Book review: Not in Narrow Seas

It's been a while since my last book review, back in August. And that's because I read a couple of novels, and after that I've been working my way through Brian Easton's Not in Narrow Seas. Regular readers of this blog will know that I have a soft spot for economic history, and Easton sets out to provide an in-depth account of the economic history of New Zealand. He justifies his approach as follows:

Many historians have told the Aotearoa New Zealand story before; but an economist uses a particular lens and, I would argue, helps us to see our nation's history in a new way.

Easton won't get any disagreement from me on that point. However, despite the prominence that Easton accords the economic lens, I found that the book fell short on that point in two ways. First, the book is heavily focused on political history, rather than economic history. Now, I will grant that it can be difficult to disentangle the political from the economic. However, a great number of words in this book are devoted to cataloguing national electoral results and the shifting political climate, which makes the economics seem somewhat of secondary importance. I would have preferred more emphasis on explaining the economic mechanisms underlying New Zealand's history and development, rather than recounting the political decisions and events that accompanied it. The core economic history context is actually available in a concise form in the Appendix, which I thought was actually one of the highlights of the book!

Second, when economic theory or technical detail does make an appearance, as opposed to the purely descriptive economic statistics that tend to suffuse books on economic history, that theory or technical detail is sometimes insufficiently explained to the reader. Consider this, on measuring the political position of government after the introduction of MMP:

The first coalition was about 0.8 standard deviations to the right of the New Zealand centre on a left-right scale and the second about the same distance to the left. That means that only 15 percent of the population were on the right of the first coalition or on the left of the second. On the social conservative-progressive scale the first coalition was also about 0.85SDs on the conservative side, while the second was a slightly closer 0.65SDs on the progressive side, with about 25 percent more progressive. The sharp change in the politics of the government did not reflect a sharp change in the politics of the voters.

None of that passage is really explained to the reader, and we are left trying to figure it out for ourselves. The book is not footnoted, but there are references at the end of each chapter. Except for the chapter that quoted passage comes from, where there are no sources provided. Easton is usually much more careful than this, so I was surprised. And it wasn't the only example of sometimes quite technical aspects of economic or political theory that were not really explained. Perhaps in a book that is already over 600 pages, there isn't room for detailed explanations? I was disappointed.

Having said that, there is a lot to learn from the book. I have to admit being under-informed about the political history of New Zealand prior to Muldoon, so despite my gripe above about the political content, it actually turned out that I appreciated that aspect of the book. There is also good coverage of early New Zealand, including pre-European times and the early encounters between Pākehā and Māori. I also learned some surprising facts, such as that New Zealand's first liquor 'laws' date to an 1833 agreement between grog sellers, brokered by British Resident James Busby. Or that New Zealand's first Labour Day celebrations, in 1890, marked the first anniversary of the formation of the Maritime Council, a union that would go on to suffer a crushing defeat in the Great Maritime Strike later that year. And this:

Wellington, which in 1843 was so much of the doldrums that one visitor described it as 'rotten to the core', was saved by sheep.

I won't forget that sentence in a hurry. So. Many. Jokes.

It takes a lot of commitment to write a 600-plus page book on economic history, and to do so in such a way that you keep the reader's attention to the end. Easton is to be commended on that, and I found Easton's writing considerably more engaging than Gary Hawke's The Making of New Zealand (which I have never reviewed, as I read it before I started blogging). I'm not sure that this book is "The" economic history of New Zealand, as the subtitle claims. But it is certainly "A" history of New Zealand, with an economic flavour, and worth reading if you have an interest in that area.

Job fatality and injury risk, and compensating differentials

Last week, my ECONS102 class covered health economics. Part of that topic involves discussion of approaches to measuring the 'value of a preventable fatality' (sometimes also called the 'value of a statistical life', or VSL). One of those approaches is the willingness-to-pay (WTP) approach, which involves estimating what people would be willing to pay in order to avoid small increases in risk. One way of applying the WTP approach in practice is to look at revealed preferences, based on people's actual risk avoidance behaviour. And one example that I use in class to illustrate the revealed preference method is to measure compensating differentials which, rather than measuring willingness-to-pay to avoid risk, measure workers' willingness-to-accept additional risk.

Compensating differentials recognise that jobs that differ in non-monetary characteristics may offer different wages. For example, consider two jobs that are similar in terms of skill requirements. If one of the jobs has positive non-monetary characteristics (e.g. it is safe, clean, and engaging), then more people will be willing to do that job. This leads to a higher supply of labour for that job, which leads to lower equilibrium wages. In contrast, if the other job has negative non-monetary characteristics (e.g. it is dangerous, dirty, and boring), then fewer people will be willing to do that job. This leads to a lower supply of labour for that job, which leads to higher equilibrium wages. The difference in wages between the attractive job that lots of people want to do and the unattractive job that fewer people want to do is called a compensating differential. The compensating differential essentially compensates workers for working in jobs with negative non-monetary characteristics compared with working in jobs with positive non-monetary characteristics.

The risk of being injured on the job is a substantial negative non-monetary characteristic for some jobs. Based on the idea of compensating differentials, workers should only be willing to take on a job with a higher risk of death or injury if that job comes with a higher wage. Measuring that additional wage would provide a measure of the compensating differential, which could then be used to estimate the value of a preventable fatality.

As an example, consider this 2022 article by Zac Reynolds, Daehoon Nahm and Craig MacMillan (all Macquarie University), published in the journal Economic Record (open access). They estimate compensating differentials for job fatality and injury risk in Australia, using income data from the Household, Income and Labour Dynamics in Australia (HILDA) survey from 2010-2018, and risk data from Safe Work Australia. Their regression model estimates the compensating wage differential (CWD) controlling for industry and occupation, as well as for a range of demographic and labour force characteristics of the workers in the sample. In their preferred set of results:

...the CWD and corresponding VSL for fatal risk are 0.13% and $9.7 million respectively.

Interestingly, union members receive significantly higher compensation for fatality risk. In their preferred model, the CWD for union members is 0.44 percent for an increase in fatality risk of one in 100,000 workers, significantly higher than for non-members. If you need a good reason to join a union, this might provide it.

Surprisingly, Reynolds et al. find no evidence of a CWD for non-fatal risk overall. In fact, their preferred results suggest that higher non-fatal injury risk is associated with lower wages, rather than higher wages, for non-union workers. However, alongside that there is a significant CWD for union members, where:

...union members receive a CWD of 0.09% for a unit increase in injury risk from the mean level which corresponds to a VSI of $65,300.

VSI is the 'value of a statistical injury', which is the non-fatal equivalent of the VSL. One limitation of Reynolds et al.'s analysis is the potential for omitted variable bias. Their models include either fatal risk or non-fatal risk, but never both together. Since these risks are positively correlated, omitting one may bias upwards the estimated effect of the other. However, Reynolds et al. explain that including both risks would create multicollinearity problems, meaning that the results would potentially be less biased, but also less precise.

Finally, Reynolds et al. compare their results with the 'official' VSL applied by the Australian government, noting that:

For Australian workers who belong to a union, this estimate implies a VSL of $34.9 million. Ignoring union coverage results in a VSL of $9.7 million. These estimates are both well above the Australian government VSL estimate of $5.1 million...

For comparison, the New Zealand government currently applies a VSL of $12.5 million (in 2021 dollars, which is comparable with the 2021 VSL that Reynolds et al. report for Australia) for measuring the benefits of road safety improvements. This is equivalent to about AU$10 million at the current exchange rate.

The 'value of a preventable fatality' (or value of a statistical life, if you prefer) isn't a price tag on anyone's life, but an estimate based on how people trade off money against small changes in the risk of death. As Reynolds et al. demonstrate, those estimates can vary substantially, which is worth remembering when governments use a single number to represent the benefits of increased safety and in other applications.

Read more:

• The likely consequences of New Zealand's new value of a statistical life

This week in research #147

Here's what caught my eye in research over the past week (a very quiet week, it seems):

• Silva, Sá, and Biscaia (open access) find that a one-standard-deviation increase in high school GPA adds 5.9 first-year ECTS (European Credit Transfer System) at university, versus 1.8 ECTS for a one-standard-deviation increase in standardised national exam score, with the difference larger in the most selective university programs

The economic impact of the 2010 FIFA World Cup

Economic impact studies undertaken in advance of large sporting events typically anticipate large positive economic effects. However, studies undertaken after the event almost universally fail to demonstrate the anticipated effects. To that list (and see the links at the end of this post for specific examples), you can now add the 2010 FIFA World Cup, hosted by South Africa.

This 2025 article by Busani Moyo (University of South Africa) and Tendai Gwatidzo (University of the Witwatersrand), published in the Journal of Sports Economics (open access), provides the details. They apply the synthetic control method, which involves comparing the time series of outcome variables for South Africa with a counterfactual made up of a weighted average of other countries, chosen because that weighted average best replicates the trajectory of the outcome variable from before the treatment date.

In their main analysis, they use 2010 as the treatment date, in order to demonstrate any effect of the staging of the 2010 FIFA World Cup itself. The synthetic control they employ is made up of 31.7 percent Brazil, 21.3 percent Morocco, 18.1 percent Algeria, 15.5 percent Botswana, and small weightings of Kazakhstan, Malaysia, Mauritania, and Argentina. The results are shown in Figure 1 from the paper:

The solid blue line is the actual GDP per capita for South Africa, while the dashed red line is the counterfactual 'synthetic South Africa', which is supposed to show what would have happened to GDP per capita if South Africa hadn't hosted the FIFA World Cup. The results clearly demonstrate that actual South Africa performed worse than synthetic South Africa, suggesting that hosting the FIFA World Cup may have made South Africa worse off (at least, using GDP per capita as a measure).

But wait, not so fast. A lot of economic activity occurs in the lead-up to the hosting of the event. Moyo and Gwatidzo therefore change the treatment date to 2004, when South Africa was announced as the host, and check whether there was any impact then. The synthetic control changes (since they are using a different treatment date), now being made up of 31.3 percent Morocco, 24.7 percent Brazil, 17.9 percent Algeria, 14.3 percent Malaysia, and small weightings of Kazakhstan, Costa Rica, and Mexico. The results from this analysis are shown in Figure 3 from the paper:

Moyo and Gwatidzo interpret this as showing a positive effect on GDP per capita in the lead-up to the event, but I struggle to see evidence of any effect in Figure 3 (at least, before 2010 where, as in Figure 1, actual South Africa performs worse than synthetic South Africa). Actual GDP per capita tracks synthetic GDP per capita remarkably closely between 2004 and 2010, with little indication of any sustained positive effect. Moyo and Gwatidzo don't provide a clear quantitative estimate of the effect, which makes it difficult to assess whether the apparent positive effect is economically meaningful or statistically significant. Both seem unlikely to me.

Moyo and Gwatidzo then turn to looking at tourism arrivals, where the results are mixed. Using 2010 as the treatment date, actual tourism arrivals exceed those for synthetic South Africa for several years after the World Cup. However, using 2004 as the treatment date, synthetic South Africa clearly outperforms actual South Africa.

Moyo and Gwatidzo then turn to looking at tourism arrivals. In this case, the effect is modestly positive, as shown in Figure 10 from the paper:

Notice that in this case, there is a small positive difference in international tourism arrivals between actual South Africa and synthetic South Africa, at least up to 2013 or 2014. So, at least there is some good news for South Africa. Nevertheless, given the apparent negative effects on GDP per capita after the event, I am not convinced by their conclusion that:

These results, at least for South Africa, show that it is not the actual hosting of the World Cup that has a positive effect on GDP but the period of preparation for the tournament. The actual hosting of the tournament is only beneficial in the form of international tourism inflows.

Like many other studies, I think the results suggest that South Africa had a good party, which attracted extra visitors from outside the country, but failed to deliver the large positive economic benefits that were anticipated. I suggest that you file this study along with the multitude of others that cast doubt on the large positive economic benefits that are routinely predicted for major sporting events.

Read more:

• Tim Hazeldine on economic impact studies
• Beware economists bearing impact studies of the America's Cup
• Even less reason to believe in the economic impact of the America's Cup
• The overstated impact of the Super Bowl
• The economic impact of the 2010 World Cup in South Africa
• The economic impact of the 2000 Sydney Olympics
• Maybe hosting the Olympics just shuffles income around a country, rather than increasing it

Modelling the Nobel Prize in Economics

The Nobel Prize in Economics (technically, the Sveriges Riksbank Prize in Economic Sciences in Memory of Alfred Nobel) will be announced next Monday. Who will win? It is a closely guarded secret, but of course there are prediction markets - at the time of writing, Kalshi has Ariel Pakes as the favourite, with Susan Athey a close second, and Richard Blundell a distant third.

In a discipline where modelling is ubiquitous, it is worth asking whether it is possible to model who will win the Nobel Prize. This recent working paper by Peter Dolton and Richard Tol (both University of Sussex) makes a good attempt. They build a dataset of past winners and candidates (based on research performance as well as winners of other awards), and then estimate a model that shows the factors correlated with winning the Nobel Prize.

Dolton and Tol start with a simple assumption, which is that the Nobel Prize Committee first chooses a field to award the prize to, and then they choose the best candidate within that field. They justify this assumption by showing that there is some regularity in the way that the award goes from field to field over time (based on their categorisation of 14 fields of economics). They use this to construct a 'transition matrix', that records how often the Nobel Prize goes from one field in one year to another the next year. 

Which fields get Nobel Prizes? Dolton and Tol show that larger fields (those with more candidates), those that have waited longer since their last prize win, and the transition matrix, are all statistically significantly correlated with which field wins the prize.

Who wins the Nobel Prize within the winning field? Dolton and Tol find that winning candidates are older (although the relationship with age is non-linear, and peaks at age 70-71 before declining), and those who have had a student already win a Nobel Prize are more likely to win.

Some important things come out of this paper. First, who will win the Nobel Prize in 2026? The 2025 winners were Joel Mokyr, Philippe Aghion, and Peter Howitt. Their research field, according to Dolton and Tol's categorisation, is 'Growth'. According to the transition matrix in Table E.8 of the paper, the most likely field to follow 'Growth', is 'Equilibrium and Welfare' (although 'Development and Economic History' and 'Macro' are also possible). 

I got ChatGPT to comb through Dolton and Tol's candidate list and identify the top candidates in 'Equilibrium and Welfare'. ChatGPT suggested Andreu Mas-Colell or Partha Dasgupta. I have heard Dasgupta's name in people's shortlists before, but not Mas-Colell. Of course, if the Committee considered they had already ticked the 'Economic History' box with Joel Mokyr last year, then next according to the transition matrix would be 'Games and Market Structure', 'Information', or 'Macroeconomics'. Across those fields, ChatGPT suggested many names, but noted that David Kreps is the strongest overall, to which I would add Ariel Pakes (note the consistency with the Kalshi market prediction). ChatGPT rated Oliver Blanchard the top candidate in Macroeconomics. Anyway, we will see next week!

Second, I love the list in Table 4 of the paper, which shows the economists who, according to Dolton and Tol's model, have had the greatest chance of winning the Nobel Prize but have not done so. Top of this list is Michal Kalecki (died in 1970), followed by Lionel Robbins (1984), Jacob Marschak (1977), Arthur Burns (1987), and Frank Hahn (2013). Other notable names in that list (at least, according to me) are Bill Phillips (died in 1975), Harold Hotelling (1983), Bill Baumol (2017), and Henri Theil (2000). There are a few on the list who are still alive, including Tim Besley, Robert Barro, Guido Tabellini, George Loewenstein, and Torsten Persson.

One reason why so many outstanding economists died without receiving the prize is that it was only first awarded in 1969, by which time there was a long backlog of worthy candidates. The third important thing to come out of this paper is a counterfactual exercise, predicting who would have won the Nobel Prize each year if the prize had been first awarded in 1901 (along with the other Nobel Prizes). Table F.15 in the paper has the results (up to 1976, after which it is assumed that the awards would continue as they have been). A number of important names appear in this list, including the first three winners being Leon Walras, Carl Menger, and Francis Edgeworth. My students would no doubt recognise Alfred Marshall (1910), Corrado Gini (1924), Joseph Schumpeter (1930), Arthur Pigou (1935), John Maynard Keynes (1938), and Joan Robinson (1958), among others. Interestingly, in this counterfactual exercise, the first woman to win the prize would have been Beatrice Webb in 1924 (jointly with Sidney Webb and Corrado Gini), some 85 years before Elinor Ostrom.

The Nobel Prize announcement is one of the highlights of my year. Now that I've considered the results from the model, with some assistance from my special adviser ChatGPT, I'm prepared to make my prediction: Kreps and Pakes. We'll find out next week!

[HT: Marginal Revolution]

The beauty premium doesn't seem to extend to billionaires

There is a broad literature supporting the existence of a 'beauty premium' in labour markets - more attractive workers tend to earn more. There are a number of channels proposed to account for this, including confidence, social or communication skills being higher among more attractive workers, attractive workers sorting into occupations that reward attractiveness, and discrimination by employers (or by customers). How does the beauty premium extend to wealth, and in particular to extreme wealth?

In this 2022 article published in the journal Labour Economics (ungated earlier version here), Daniel Hamermesh (University of Texas at Austin) and Andrew Leigh (Australian Member of Parliament) investigate the relationship between attractiveness and wealth among billionaires on the 2008 Forbes billionaires list. The attractiveness ratings for each billionaire are derived from 0-10 scores given by 16 students at Australian National University, based on photos published with the list. Hamermesh and Leigh control for age (and age-squared), education (whether the billionaire was a college graduate), and whether they were male, from a Western country, and inherited their wealth.

In their main analysis, Hamermesh and Leigh find that wealth is not statistically significantly correlated with attractiveness in their billionaires sample. They find the same for education - it is not statistically significantly correlated with wealth in their sample either. They try various alternative specifications, none of which show a significant correlation. Hamermesh and Leigh conclude that their results:

...show that a sample that is taken from the extreme tail of the dependent variable no longer exhibits the systematic patterns seen across the distribution.

They illustrate this further using data from the American Community Survey. While they don't have attractiveness in that data source, they do have education and labour market experience variables. They find that, for the whole sample, both college education and labour market experience are positively and statistically significantly associated with income. However, when they restrict the sample to the 0.1 percent highest earners in the sample, there is no significant correlation between education and income (or between labour market experience and income).

Hamermesh and Leigh use their results to caution us against drawing conclusions from analyses conducted on extreme tails of a distribution, noting that:

Empirical regularities that are common in the general population may not hold up among atypical subsets of the population, whether drawn from the elite (e.g., Olympic athletes, Fortune 500 CEOs) or the most disadvantaged (e.g., prisoners, the homeless). Within these groups, outcomes are more likely to be due to unobservables and pure chance than observable traits.

Coming back to the billionaires, why is there no correlation between attractiveness and wealth? Think about the main mechanisms that are proposed to underlie the beauty premium. More attractive billionaires likely aren't benefiting from employer discrimination, and customer discrimination seems less obviously important. Sorting into different occupations that reward attractiveness is unlikely to explain much of the differences in wealth at the extreme (and if we treat 'billionaire' as an occupation, then arguably they all have the same occupation). That leaves differences in confidence, social or communication skills as more plausible mechanisms. There might be differences between billionaires in those traits, but would they really explain the difference in wealth between Warren Buffett (US$62 billion in 2008) and Larry Ellison (US$25 billion)? Perhaps some further research might uncover the relationship between confidence and wealth among billionaires, but I think it more likely that a lot of it is luck. Buffett even famously said as much.

This research is a useful reminder to pay attention to how a sample has been selected. Relationships that are strong across the population may disappear when we look only at people drawn from the extreme tail of the outcome distribution. That doesn't necessarily mean that attractiveness doesn't matter at all for billionaires, only that we cannot expect the same relationship to hold in such an unusual sample. As F. Scott Fitzgerald wrote about the very rich: "They are different from you and me".

Read more:

• The beauty premium in undergraduate study is small, and more attractive women major in economics
• New results questioning the beauty premium should be treated with caution
• The beauty premium in the LPGA
• The beauty and height premiums in the labour market
• Blonds have more fun (or rather, they get paid more)
• The beauty premium for economists
• More evidence on the blond wage premium
• The beauty premium at the intersection of race and gender
• Narrowing down on the source of the beauty premium
• The beauty premium and student grades for in-person and online education
• Is there a premium for very unattractive workers?

How Alaska is like South Waikato

I've written a couple of times about the challenges that South Waikato faces in attracting workers (see here and here). This problem is not unique to South Waikato - a lot of rural and remote areas face the same challenges. In just the latest example I've seen, Jordan McGillis reports in City Journal on the situation in Alaska:

My analysis of Census data finds that more than 40 percent of civilian, prime-age, noncollege Alaska men employed in blue-collar occupations earn at least $75,000. More strikingly, 9.4 percent of Alaska’s civilian, prime-age, noncollege men work a blue-collar job and earn at least $100,000, compared with 3.6 percent nationally, ranking Alaska first in the country. Among those in such occupations more than one in four earns six figures...

That’s no surprise to [Ray Weber, Dean of technical and vocational education at the University of Alaska Anchorage]. “Can [wages in Alaska] be higher [than in the rest of the country]? Yes, especially Slope jobs or jobs that suck. Like, we have electrical linemen that go across the state. They’re going to Unalakleet, the only way to get there is by airplane. And there’s one pizza joint. One. No other restaurant. You generally end up sleeping either in a bunkhouse, if you’re lucky. Or you’re sleeping in the school auditorium. . . . I like Alaska, but we’re asking the wrong questions if you’re saying, ‘high-paying, lucrative jobs.’ What do the younger generation consider important? The answer is: Things that are not in Alaska.”

Economists call this a compensating wage differential: the premium required to induce workers to accept jobs with undesirable nonpecuniary characteristics. Work in Alaska is colder, darker, lonelier, and often more dangerous. The roughly $120,000 premium earned by North Slope oil-and-gas workers over their counterparts elsewhere, and the roughly $40,000 premium earned by electricians deploying to isolated communities such as Unalakleet, are partly the price employers must pay to fill jobs few want.

Workers with the same skills can often accept lower wages in Texas or Louisiana in exchange for a more attractive lifestyle.

Wages differ for the same job in different firms or locations. Consider the same job in two different locations. If the job in the first location has positive non-monetary characteristics (e.g. it is in an area that has high amenity value, where people like to live), then more people will be willing to do that job. This leads to a higher supply of labour for that job, which leads to lower equilibrium wages. In contrast, if the job in the second location (Alaska, for example) has negative non-monetary characteristics (e.g. it is in an area with lower amenity value, where fewer people like to live, where it is cold, dark, lonely, and more dangerous), then fewer people will be willing to do that job. This leads to a lower supply of labour for that job, which leads to higher equilibrium wages. As McGillis notes, the difference in wages between the attractive job that lots of people want to do and the unattractive job that fewer people want to do is called a compensating differential. The compensating differential essentially compensates workers for working in jobs with negative non-monetary characteristics compared with working in jobs with positive non-monetary characteristics.

And so, jobs that pay less in the rest of the US must pay much more in Alaska in order to attract workers.

Read more:

• Reason to be wary if a job in Taumarunui offers an Auckland salary
• The compensating differential for rural GPs must be enormous
• Compensating differentials are alive and well in Tokoroa
• Rural Australia sets a new high bar for compensating differentials
• The $300,000 job that no one wants
• The South Waikato compensating differential strikes again

This week in research #146

Here's what caught my eye in research over the past week:

• James and Kjorstad find that spelling and grammar errors significantly reduced perceived research quality within a sample of US economists, while manuscripts prepared in LaTeX were generally evaluated more favourably than those prepared in Microsoft Word within an international sample of economists
• Fornwagner (with ungated earlier version here) finds no effect of menopause on women’s risk preferences, using an online sample of over 1700 UK women
• Wang, Wang, and Zhou (with ungated earlier version here) use global air traffic data as instrumental variable to estimate the effect of in-person interactions on bilateral trade, finding that online interactions are an imperfect substitute for in-person contact
• Garthwaite et al. (with ungated earlier version here) find that the rent-sharing between college basketball and football programmes and other sports transfers spending away from students who are more likely to be Black and come from poor neighbourhoods toward students more likely to be White and come from higher-income neighbourhoods

The declining returns to higher education in Australia

In yesterday's post, I discussed a recent meta-analysis on the returns to compulsory schooling. That research was necessarily focused on secondary education. What about the returns to higher education? Two recent articles tell an interesting story about the change over time in returns to higher education in Australia.

The first article is this one by Elisa Birch and Alison Preston (both University of Western Australia), published in the journal Economic Record (open access). Using data from the Household, Income and Labour Dynamics in Australia (HILDA) survey, they document a striking decrease in the returns to higher education over the period from 2001 to 2023. They apply a fairly standard Mincer wage equation, comparing hourly earnings between workers aged 28 to 38 years with no post-secondary qualification to those with one of seven categories of education: PhD; Master’s; Graduate Diploma; Graduate Certificate; Honours; Bachelor’s; and Diploma/Certificate. Strictly speaking, these estimates are wage premiums associated with different qualifications, rather than necessarily causal estimates of the returns to obtaining those qualifications. Nevertheless, they do show some interesting trends, and their headline results are summarised in Figure 2 from the paper:

Notice that, in all cases (except, arguably, Graduate Certificates), the general trend in the returns to higher education has been downward. The wage premium between higher education and no post-secondary education is lower in 2019-2023 than it was in 2001-2005. Birch and Preston also look in more detail at the difference in returns between the early period in their data (from 2001-2011) and the later period (from 2011-2023), and the differences by gender, and find that:

...over the study period, the wage premium for a Master’s degree declined by 13.2 per cent for males. For females, the decline in the Master’s wage premium was even more pronounced, falling by 16.6 per cent... The reduction in the return to a Bachelor’s degree was similar for both males and females, with a decrease of approximately 13 per cent in both groups. In Period 2, the wage premium on a Master’s degree was equal to 35 per cent among men and 26 per cent among women (a gender gap of 9 percentage points). Among Bachelor’s degree holders the wage premium in 2012–23 was equal to 27 per cent among men and 22 per cent among women (a gender gap of 5 percentage points).

So, there isn't even anything positive to say about the changes in relation to the gender wage gap. The higher wage premium for male graduates remained the same, or increased, compared to the wage premium for female graduates. Even worse, when looking across the whole wage distribution, Birch and Preston find that:

While the male wage structure appears to have shifted downward and flattened, the female wage structure shows both a downward shift and a notable decline in returns at higher wage levels.

So, the returns to higher education have been declining in Australia (a result that other studies have also found), particularly at the upper end of the wage distribution for women. Birch and Preston largely leave open the question of why. That is where this new article by Michael Coelli and Jeff Borland (both University of Melbourne), published in the journal Australian Economic Review (open access), comes in.

Coelli and Borland first note that the decline in the returns to higher education is concentrated in the period after 2001, using five-yearly Australian Census data from 1981 to 2021, and using income as a measure (rather than hourly earnings, which isn't available in the Census). Then, they explore different explanations for why the returns to higher education changed, focusing on explanations that might explain a decrease starting around 2001. They find little support for a slowdown in skill-biased technical change relative to increases in the supply of skilled workers. Instead, they find that the decrease is associated with a relative increase in the wages of workers with no post-secondary education since 2001, and link that to two main probable causes.

First, the Federal Minimum Wage began increasing in real terms from 1996. On top of that:

The method of setting wages for employees covered by awards during 1993 to 2010 may also have raised the relative wages of low‐skill employees. During that time, the annual increases to rates of pay for employees covered by awards were made in flat dollar amounts, resulting in higher percentage growth in wages for lower wage earners. This may have underpinned stronger wage growth in low‐skill occupations... Award rate increases were above inflation at the lower end of the wage distribution (especially at the minimum wage), but below inflation further up the distribution. Since 2011, a common percentage increase has been set for employees covered by all awards.

So, the process of wages being determined by awards, alongside an increasing real minimum wage, may have compressed the wage distribution, especially before 2011. This would have the effect of raising wages for workers at the bottom of the wage distribution, including many workers without post-secondary education, relative to workers at the top of the wage distribution, including many workers with higher education.

Second, Coelli and Borland note that:

Starting around 2001, Australia experienced a large and extended boom in mining, coinciding with the rise of China in global trade after its accession to the WTO. This affected employment in mining and in construction. Its impact was felt more strongly among low‐skilled workers than among the higher‐skilled.

Since workers with no postsecondary education are more likely to be working in mining and construction than workers with higher education, the former workers likely benefited disproportionately from the mining boom.

Alongside those two explanations, Coelli and Borland also note a phenomenon that they call 'occupational downgrading':

Growth in the supply of workers with a bachelor's degree generally exceeded demand growth throughout the 1981 to 2021 period, leading to them moving down the jobs ladder.

These three explanations (minimum wages and wage awards; the mining boom; and occupational downgrading) are Coelli and Borland's explanations for the decrease in the returns to higher education. They do leave some questions from Birch and Preston unanswered though - why has the decrease in the returns to higher education been concentrated among the top of the wage distribution for female workers? The top of the wage distribution is least likely to be affected by minimum wages and wage awards. However, that leaves the other explanations for the decrease. Is it because female workers are more likely to 'occupationally downgrade'? Or, did female workers miss out on the mining boom? Clearly, there is more research to be done here.

Read more:

• Internal rates of return for college majors in the US
• The returns to each additional year of compulsory schooling

The returns to each additional year of compulsory schooling

This week, my ECONS102 class is covering the economics of education. Part of that topic considers the private education decision - the decision each individual makes about the amount of education they receive. This depends on the private costs and benefits of education. Arguably the primary benefit of education is incremental income - the additional lifetime earnings that education brings.

How big are those gains? Previous studies I have seen (such as a study reported in Paul Oyer's book Everything I Ever Needed to Know about Economics I Learned from Online Dating (which I reviewed here)) have estimated effects in the order of 10 percent per additional year of education. However, that was a single study, which could easily provide a biased view if considered in isolation.

What are the effects more generally? This recent article by Gregory Clark and Christian Nielsen (both University of Southern Denmark), published in the journal Kyklos (open access), provides an answer based on a meta-analysis of 79 causal estimates of the effect of an additional year or years of compulsory schooling on earnings, drawn from 53 other papers that apply methods consistent with causal inference. Meta-analysis provides a method of combining the estimates across many studies into one overall estimate. Importantly, Clark and Nielsen apply various methods to correct for the effects of publication bias (which is the tendency for statistically significant results to be more likely to be published, while statistically insignificant results tend to be missing from the research record).

Clark and Nielsen report baseline results that are somewhat lower than I expected:

With these procedures the average percentage gain in earnings from an additional year of schooling was 8.2% for 79 independent estimates, from 53 papers... If these estimates are weighed by their precision in a random effects estimation, the gains are reduced to 6.0%, and similarly 6.2% for a fixed effects weighting.

Given that the true effect is likely to vary across populations and study settings, I would consider the random-effects estimate of 6.0 percent for each additional year of schooling as the more plausible of the three. However, Clark and Nielsen also employ meta-regression, using PET (Precision Effect Test) and PEESE (Precision Effect Estimate with Standard Error) regression models, which are designed to address imprecision and publication bias in the overall sample of results. After removing some outliers, these approaches suggest:

...a true effect of 6.1%–6.4% to an extra year of education and an insignificant effect of publication selection...

So again, it seems like the effect may be in the order of six percent per additional year of schooling. Interestingly, Clark and Nielsen themselves note that the PET-PEESE evidence for publication bias is highly model-dependent, and that these results could instead reflect genuine heterogeneity in the true effect. However, Clark and Nielsen go a little further, looking in more detail at the distribution of effects across the 79 estimates. They note that if the estimates of the earnings return were normally distributed, then there are a lot of 'missing' studies with negative returns. They also consider whether the underlying returns might be log-normally distributed. Once you consider sampling error, the log-normal distribution can still generate negative estimates, and Clark and Nielsen argue that there are fewer negative estimates in the published literature than the model would predict. Consider Figure 10 from the paper:

In the figure, the yellow bars show a histogram of the 79 estimates in the sample. The blue line shows a log-normal distribution based on the estimates. Notice that, compared with the log-normal distribution, there are more studies than expected with estimates in the 0-4 percent range, and too few in the negative range. Clark and Nielsen argue that, if the 'missing' studies were included, a better estimate of the returns to each additional year of schooling may be in the range of 0-3 percent.

However, I’m not convinced that Clark and Nielsen’s distributional assumptions are justified here. Figure 10 (like Figure 8 in the paper) shows that the observed distribution of estimates doesn’t match their chosen distribution. But they haven’t established that their chosen distribution is what we should expect in the absence of publication bias. That distribution would depend on differences in estimation methods, the precision of the estimates, and genuine differences in returns across the populations studied. A mismatch between their chosen distribution and the observed study estimates therefore does not, by itself, demonstrate that studies with negative estimates are missing. So, while publication bias remains a possible explanation, I would treat Clark and Nielsen’s proposed 0-3 percent range for the average return very cautiously.

Nevertheless, overall it does seem clear from their other results that the returns to an additional year of schooling may be somewhat lower than the 10 percent found in other studies. What would be interesting to know next is whether the estimated returns have been changing over time (which my next blog post will look at, for Australia), and what study features lead to different estimates. Clark and Nielsen do note that instrumental variable regression leads to higher estimates than difference-in-differences or regression discontinuity estimates. However, there are other contextual variables that may matter as well. Hopefully, someone else can pick up on this work and take it to the next step. That might help to distinguish publication bias from the more mundane possibility that there simply isn't a single 'true' return to an additional year of schooling.

Read more:

• Internal rates of return for college majors in the US

The costs of traffic noise

Traffic noise is an example of a negative externality - the impact of an action on a third party (a 'bystander') who has not consented to or played any role in the carrying out of that action. In this case, traffic noise caused by drivers on the road negatively affects people who live nearby, and those people haven't played any role in creating the noise.

To the extent that traffic noise negatively affects people living nearby, it should be reflected in property prices. Under hedonic demand theory, when someone buys a property they are really buying a bundle of characteristics of the property, one of which is the presence of traffic noise. Since traffic noise is a negative characteristic, we would expect properties that are exposed to more traffic noise to have lower prices, ceteris paribus (holding all else constant).

That means that the cost of traffic noise can be evaluated by carefully looking at the relationship between property values and traffic noise. That is what this 2025 NBER Working Paper (ungated here) by Enrico Moretti (University of California, Berkeley) and Harrison Wheeler (University of Toronto), sets out to do. They start by looking at the effect of roadside noise barriers on property prices, noting that these barriers provide an exogenous source of variation in traffic noise. They compare houses that are close to the barrier (within 500 metres directly away from the barrier) with those that are further away (between 500 and 1500 metres), using a difference-in-differences (DiD) strategy. That involves comparing the change in property prices between before and after the barrier was erected, between properties close to and those further away.

Moretti and Wheeler primarily use data from Florida, which provides details about the noise barriers that were completed, but also about barriers that were proposed but not completed. That also allows them to use a 'triple-differences' strategy, by matching areas that had a barrier erected, with those that didn't (but where one was proposed). Now, it turns out that the results from the triple-differences model are similar to the more standard DiD, but that should provide some further comfort with the robustness of the DiD results. The data on house prices and other characteristics comes from CoreLogic and covers the period from 1990 to 2022 (house prices) or 2006 to 2022 (property characteristics). Their final dataset includes nearly 600,000 home sales within 1500m of a noise barrier (and on the same side of the road as the barrier).

Focusing on their results that include parcel fixed-effects (so that time-invariant property characteristics are controlled for), Moretti and Wheeler find that:

For houses within 100 m of the barrier, the estimated effect increases to 8.59%. For houses 100–200 and 200–300 m from the barrier, the estimated effects increase to 5.79% and 4.41%, respectively. The effect on properties 300–400 m from the barrier is marginally statistically significant.

So, reducing traffic noise increases property values, and the effect is largest for properties closest to the road generating the noise. Beyond about 300 metres, the effect is statistically indistinguishable from zero, but within 300 metres, the construction of a noise barrier increases property values by between 4.41 and 8.59 percent. Moretti and Wheeler then note that:

Since our data report the construction cost of each barrier, we can compute the marginal value of public funds (MVPF), defined as the property value appreciation over costs... The average MVPF for barriers that were built amounts to 1.7, while the MVPF for barriers proposed but not built is 1.4. This is to be considered as a back-of-the-envelope calculation that ignores property taxes. Property taxes would reduce both the social benefits (since some of the home value increase gets taxed), and the social costs (since property taxes end up in local government coffers).

So, on this measure at least, noise barriers appear to be a good use of public funds, since the increases in property values exceed the costs of erecting the barriers (although noting that there may be other uses of public funds with even higher MVPF values).

Next, Moretti and Wheeler change their model in order to allow the change in price to vary based on the expected decibel reduction. They use a model where the effect is non-linear in noise reduction, and find that:

The effect plateaus at 10 dB of reduction, which represents the 96th percentile in our sample. The effect is estimated to be zero when noise reductions are 4.9 dB...

The average barrier in our sample reduces noise by 7.15 decibels. At this level of noise reduction, our estimates imply that the average price of a decibel is 0.94%...

That is the result that Moretti and Wheeler use later in their paper to estimate the economic cost of the externality. However, first they need to rule out some competing explanations for their effects. They show that air quality is somewhat better near to the barriers, but the effect is small and not statistically significant. They show that the results do not change when accounting for tree canopy or the presence of other buildings, meaning that noise barriers blocking views of the road is unlikely to explain the results. And, they show that the construction of new homes (with higher unobserved quality) does not explain the results.

Moretti and Wheeler next use their results to estimate the economic cost of the externality for each census tract in the US. From that, they find that there is:

...a negative correlation between the cost of the externality and median family incomes. The slope is -0.10 (0.01), indicating that a 10% lower income is associated with a 1% higher per capita cost. The correlations with the share of residents who are Black and the poverty rate are positive. The slopes are 0.08 (0.01) and 0.63 (0.05), respectively, indicating that a 10 percentage point higher share of Blacks or a 1 percentage point higher poverty rate are associated with 0.8% and 0.6% higher per-capita costs...

That means that the externality cost is regressive. That is, the cost of the externality is a larger proportion of income for low-income families than for higher-income families (taking into account the location of low-income and higher-income families and the property values where they live and the traffic noise they face).

In total, Moretti and Wheeler estimate the cost of traffic noise to be US$7.0 billion in Florida, and $109.75 billion for the US as a whole. Looking across cities, they find that per-capita traffic noise costs increase with urban share of the population and population density, which they suggest is because cities with greater urban share or those that are denser have both higher noise exposure, and higher property values.

Finally, Moretti and Wheeler estimate that a one-off Pigovian tax equivalent to US$974 per car would be equal to the marginal external cost of the traffic noise externality (noting that the optimal Pigovian tax is one that is equal to the marginal external cost). They also estimate the potential benefit of a move to 100 percent electric vehicles (which have lower engine noise) at US$5.39 billion for Florida, and $77.28 billion for the US as a whole.

This research tells us that traffic noise is a costly negative externality. Those costs are capitalised into property values and are borne disproportionately by lower-income households. Fortunately, the research also suggests that there are worthwhile ways of reducing those costs, including erecting noise barriers and rolling out more electric vehicles, and the benefits of those solutions may be substantial.

[HT: Marginal Revolution, last year]