This week in research #64

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

• Kawada, Nakamura, and Okamoto propose a new measure of population ageing based on the ‘‘thickness" of the working-age population (I'm disappointed that they didn't reference my measure of structural ageing)
• Meyer et al. (open access) review the economics literature on the impacts of the COVID-19 pandemic on income inequality

And the latest paper from my own research (or, more accurately, these are the second and third papers from the thesis research of my PhD student Jayani Wijesinghe, on which I am a co-author along with Susan Olivia and Les Oxley):

• This new working paper investigates the impact of homicide on state-level life expectancy and lifespan inequality in the US, as well as the factors associated with life expectancy loss and lifespan inequality change due to homicide, and finds that, among other things, corrections and judicial spending influence both life expectancy and lifespan inequality, while police and health spending mitigate lifespan inequality, and welfare expenditures correlate with higher lifespan inequality
• This new working paper compares the impacts of three pandemics on life expectancy and lifespan inequality - (1) the 1918 influenza pandemic; (2) the HIV/AIDS pandemic; and (3) the COVID-19 pandemic - and explores the similarities and differences in their impacts

Also new from the Waikato working papers series:

• Dorner et al. test consumers' willingness to engage with a Digital Product Passport (DPP), a way of collecting and sharing a product’s information throughout its lifecycle, which is part of the EU's  new circular economy action plan, and find that the DPP may not lead to further shifts in environmental orientation and behaviour

When John Maynard Keynes was the examiner for New Zealand economics students

One of my former colleagues (now retired), Brian Silverstone, used to share stories of the old days, when examination papers were shipped off from New Zealand to England for marking at the end of the year. The examiner would be some moderately known academic in England, who few in New Zealand would know, and who would likely never have set foot in New Zealand before.

Thankfully, such colonial practices have long since ceased. However, they have lasting effects. For example, this historical examination process explains why most New Zealand universities have graduation ceremonies in April, for students who completed their studies at the end of the previous year. That was necessary in times when the examination scripts would face an eight-week journey by boat in each direction, but it makes little sense these days, when final grades are known within days or weeks of the end of the trimester.

Anyway, not all of the examiners were relatively unknown. As Conrad Blyth (University of Auckland) noted in this 2007 article, published in the journal History of Economics Review (sorry, I don't see an ungated version online), John Maynard Keynes was the economics examiner for the University of New Zealand (as it was in those days) in 1919. Interestingly, his father John Neville Keynes had been the examiner for 1899-1903.

Blyth's article includes full copies of the six Honours and MA examination papers that the younger Keynes set for students. They make for an interesting read. It is surprising how many of the questions bear close resemblance to questions I might ask my students now, such as:

Why are lawyers better paid than coal-miners? Should they be? The answer should contain a careful analysis of all the important underlying causes...

What sort of undertakings can be suitably run by municipalities? Should they aim at making a profit? Would they ever be justified in making a loss?...

Discuss the justification and the utility of Governmental intervention to fix prices. What are the chief dangers of such action?...

What is the incidence of taxes on (a) rent, (b) profits, (c) bread, (d) alcoholic drinks (e) domestic servants?

Of course, Keynes would have been looking for a much greater depth of insight and analysis from these Honours and MA students than what I expect in my ECONS102 class. Nevertheless, the persistence of the sorts of questions that economics academics ask of students is striking, even a century later. In another sign that the more things change, the more they stay the same, this essay topic appears in Paper (f):

The Policy of New Zealand towards Immigration now and in the future. 

I don't ask students to write essays, having dispensed with them after my first year of teaching. However, if I were setting an essay question today, that seems to me to be a topic that would be attractive right now.

My students generally do a good job of answering these questions (to the standard I am expecting). However, Keynes was not impressed by the quality of the students that he examined. He made the following comments in his examiner's report:

The candidates as a whole showed that they had been reasonably industrious and conscientious in their work. If, however, I may make a criticism, it would be that they are unduly tied down to their textbooks, often imperfectly understood, and show an insufficient independence of mind. There was hardly any single instance in any of the papers of an individual or characteristic remark showing the personality of the writer. There was hardly even a lively or impudent passage amongst the lot. The general impression produced on the examiner’s mind is therefore very dull...

I'm extra glad that I don't have dull students!

[HT: Thomas Scrimgeour, via Frank Scrimgeour]

More notes on generative AI and teaching (in higher education)

Over the last six months, I have become increasingly disappointed by the response of higher education to generative AI. Don't get me wrong. There are lots of individual academics doing great things with generative AI, in teaching and research. However, the sector as a whole is, in my view, not fully grappling with it. At least, not in a way that is helpful to most academics who are on the frontline and worried about what they should be doing.

I've written a couple of posts on this topic (see here and here), plus a number of other posts related to various aspects of generative AI. The paid versions of ChatGPT and Elicit have increasingly become part of my regular workflows (although I'm almost certainly not making optimal use of either tool, and I'm resistant to having them take over too many aspects of my work, because I enjoy it).

I'm not an expert on AI and higher education, but what I have seen coming out of those that are being held up as experts, is pretty underwhelming. What I'm seeing is a lot of frameworks and white papers that are incredibly light on solutions - what should we, as teachers, be doing in order to prepare our students for a future labour market where working with generative AI is the norm. Early work like this TEQSA paper from late 2023 can probably be forgiven for a narrow focus on generative AI's impact on assessment. That was where my thinking was initially drawn to (see here). However, we should be doing better by now.

I was very disappointed in reading APRU's recent white paper which was big on framework, and light on detail. It is stacked full of banalities like:

...an existential threat is felt by higher education researchers and educators who may see their functions or parts of their roles being diminished or replaced by AI, may not know how to adapt from more traditional approaches, and are already under significant workload pressures... Early student perspectives suggest, however, that despite students being open to receiving assistance from AI, they still value the human elements of teacher-student relationships...

And this bit demonstrates that thinking hasn't really moved on from TEQSA's paper over one year ago, despite the rapid advances in generative AI that have occurred since then:

With generative AI increasingly able to perform well in assessments... unsupervised assessments are no longer able to assure attainment of learning outcomes. This does not mean that every assessment must now be supervised; rather, it means that assessment redesign is needed so that there is a pedagogically beneficial mixture of ‘secured’ assessment of learning and ‘open’ assessment for learning.

I want to pick up on two bits from the APRU white paper though. First:

...recent reports suggest that universities are not providing the necessary familiarity-building activities that students need...

And second:

...universities need to prepare learners for an AI-driven world and shift from a focus on knowledge to values and skills...

The problem is that the white paper is light on solutions to those problems. So, I'm going to do so here, at the risk of providing yet another framework. However, this approach that I will outline is grounded in a belief that higher education should be preparing students to work with AI, and should do so in an intentional way, scaffolding students through their studies to ensure that they are well prepared. My comments are focused on education across three years within a single discipline (a major within a degree), but can easily be extended to consider a whole degree programme.

In the first year, developing core knowledge and basic principles remains important. Even though generative AI can do a better job of answering questions about basic knowledge (and more complex knowledge) in any discipline than humans can, learning basic principles is about more than knowledge. It socialises ways of thinking that have developed within the discipline. In economics for example, we talk about the 'economic way of thinking'. If, in educating future students, we try to leapfrog this key step, in my view we generate graduates who lack a clear framework of understanding and interpreting the world. Even worse, they lack the foundation for interpreting generative AI model outputs, and this will hamper them from working effectively with generative AI.

Previously, I have pointed to the problems of models hallucinating (see here), and how knowledge of basic principles could help students in recognising hallucinations. However, the latest models, and especially those that can employ complex reasoning, are much less prone to hallucinations. However, I don't think this means that students can forego learning the basics.

This approach to the first year of a programme has consequences for how it can be taught or assessed. While generative AI can be used in a tutoring role (as I am doing in my ECONS101 class), that comes with risk in terms of assessment. Students should not be able to outsource their assessment activities to a generative AI that can easily answer questions about basic knowledge and principles, and to a high standard. That means that in the first year, more invigilated assessment may be required. Indeed, in my ECONS101 class we have moved from 84 percent invigilated in-person tests to 88 percent. Much of the rest of the assessment is based on completion of tutorial tasks, which is linked to learning, but supported in class by human tutors. We have built in a lot of learning opportunities for students in that class, in order to ensure that the basic principles are learned well.

In the second year, students can build on their basic knowledge, and begin working with AI. Hopefully, students will have taken the opportunities to be exposed to generative AI tools in their first year, although not having used them in high-stakes assessment, this means that students need to develop more specific skills in task-oriented prompting of generative AI. One of my colleagues, Pedram Nourani, demonstrated some excellent examples of students working with AI, step-by-step, through a very tightly structured assessment task. The assessment of the task in that example was based on their final output (of an Excel spreadsheet that analysed a bespoke dataset for each student), but it could easily be based on a combination of final output and transcripts of the conversations with the AI. Invigilated assessment, therefore, potentially has less of a role to play. Having students work directly with AI, in a tightly structured format, allows them to build their prompting skills in a task where the teacher can be fairly sure about the outcome.

In the third year then, students can progress to more unstructured tasks, working with one or more generative AI tools on fairly open-ended assessment tasks that more closely mimic the types of tasks that students will encounter after graduating. In these tasks, students might be given a research question to answer or an objective to meet, and decide for themselves, using the generative AI tools that are available, how best to answer the research question or meet the objective. Students might even be encouraged to consult with multiple generative AI tools for suggestions on how to complete the assessment, then execute the assessment in collaboration with the generative AI. Students' performance can be judged based on the final output, as well as transcripts of their various interactions with generative AI tools.

As with my proposal for the second year, invigilated assessment may not be required. However, it is worth noting that at both second and third years, ensuring that it is the students themselves that are interacting with the generative AI tools, and not some proxy (such as another human, or another generative AI!) remains a challenge. An intermediate approach then may be to have students complete the assessment in a supervised manner within a computer lab setting on campus.

As you can hopefully see, despite some remaining challenges, this overall approach to embedding generative AI into teaching across a programme (a major, or perhaps a degree) scaffolds the student through their learning journey, giving them the core disciplinary knowledge and skills necessary to work in their chosen field, as well as the transferable skills of working effectively with generative AI. All that's left is for me to convince my colleagues that this is a sensible way forward, and that it will ultimately lead to more employable and more successful graduates.

[HT: Karyn Rastrick for the APRU white paper]

AI tutoring vs. active learning in physics

In yesterday's post, I signalled that one of the factors that convinced me to develop Harriet, our ECONS101 AI tutor, was this 2024 paper by Gregory Kestin (Harvard University) and co-authors. The created an AI tutor that would walk students through some lessons in the Harvard physics class Physical Sciences 2 (PS2), and compared the learning gains with those for students who experienced the traditional (active learning) approach. As Kestin et al. explain:

Through content-rich prompt engineering, we developed an online tutor that uses GAI and best practices from pedagogy and educational psychology to promote learning in undergraduate science education. We conducted a randomized controlled experiment in a large undergraduate physics course (N = 194) at Harvard University to measure the difference between 1) how much students learn and 2) students’ perceptions of the learning experience when identical material is presented through an AI tutor compared with an active learning classroom.

So, this wasn't a case of comparing an AI tutor with nothing, or even comparing an AI tutor with a traditional static learning experience, but comparing an AI tutor with a teaching approach that has demonstrated high efficacy (active learning). Kestin et al. separated the class into two groups, and conducted their experiment over two weeks (two lessons):

The first week, group 1 engaged with an AI-supported lesson at home while group 2 participated in an instructor-guided active learning lecture. The conditions were reversed the following week. To establish baseline knowledge, students from both groups completed a pre-test prior to each lesson—focusing on surface tension in the first week and fluid flow in the second. Following the lessons, students completed post-tests to measure content mastery and answered four questions aimed at gauging their learning experience, including engagement, enjoyment, motivation, and growth mindset.

After controlling for the pre-test score, midterm exam score, and a measure of prior proficiency in physics, Kestin et al. found that:

...controlling for all these factors, the students in the AI group performed substantially better on the post-test compared with those in the active lecture group. We show this to be a highly significant... result with a large effect size.

The effect size was a 0.63 standard deviation greater improvement in knowledge for the AI tutored group, compared with the active learning group. That is a substantial difference! And on top of that, students reported feeling more engaged in the lesson and more motivated to learn.

All in all, this was quite a convincing endorsement of AI tutoring. However, Kestin et al. then conclude that:

As in a “flipped classroom” approach, an AI tutor should not replace in-person teaching—rather, it should be used to bring all students up to a level where they can achieve the maximum benefit from their time in class.

If I had one gripe about the paper, it is that Kestin et al. didn't look at differences in effect size between the top students and the bottom students. Having said that, this was a Harvard physics class, so it's not clear what an analysis of heterogeneity might tell us (because even the average Harvard students are much better than most). My worry is that, like blended learning and a lot of other initiatives that we could put in place as teachers, an AI tutor has the potential to increase the divide between the most engaged students and the least engaged students. And that's part of the reason why Harriet, our ECONS101 tutor, is being used as a complement to other ways that my ECONS101 students can improve their learning. I don't see the AI tutor replacing the human tutor entirely just yet.

[HT: Marginal Revolution]

Introducing Harriet, our ECONS101 AI tutor

I've spent a substantial proportion of the last couple of weeks on a new project, testing out and training a new AI tutor for my ECONS101 class. And now she's ready to debut. I released her for the human ECONS101 tutors to try out over the weekend, and to my ECONS101 class in their first lecture today.

You can find the AI tutor here. Her name is Harriet, and currently she is trained on material from Topics 1 to 3 from ECONS101. That covers some of the basic economic concepts like opportunity cost, incentives, and economic rent, a simple production model (with iso-cost lines, but not isoquants), constrained optimisation models (with budget constraints and indifference curves), and game theory. There will be separate instances of Harriet for later groupings of three topics.

The idea for having an ECONS101 AI tutor has been percolating for a while. However, the impetus for development was set off by two seminars that I briefly mentioned in this post last month - this seminar by Justin Wolfers, and this seminar by Kevin Bryan (and especially his description of the AI teaching assistants at alldayta.com), as well as this 2024 paper describing the effectiveness of an AI tutor in Harvard engineering papers (which I will blog about later this week). I was convinced that we could develop something that would work for our students.

I should also acknowledge the parallel work on AI tutors in ECONS101 being undertaken by Michael Ryan at our Tauranga campus. Michael started development in advance of me, so I was able to free ride a little bit on his learnings (although we have slightly different use cases for our AI tutors). Michael was able to point out some important blind spots that AI tutors in economics seem to suffer from. I hope that he has learned something from me along the way as well.

In testing, it has become clear that Harriet is very good at some things, and less good at others. With basic economic concepts, Harriet is excellent, albeit a bit more verbose than I would like. I'd rather that she prompts students to ask more questions, but it seems like she prefers to give a more fulsome answer herself to any question she is presented with. I know that we could train her out of this, but it seems that the only effective way to do so is to try and anticipate exactly (or nearly exactly) the questions that students may ask her. In the Harvard paper I referred to above, they turned this into a key feature of their AI tutor, which was trained to walk students through a particular tutorial, step-by-step. Instead, I wanted students to be able to ask any question they like, and get a helpful answer.

Harriet is great at interpreting economics diagrams, and solving game theory problems, provided the diagram or problem is uploaded as a screenshot when the student asks the question. She is rubbish at drawing her own diagrams, and at creating her own game theory examples and solving them. Clearly, there is still substantial scope for human tutors in ECONS101, at least until we can train an AI to draw more than a very simple supply and demand diagram (and even then, Harriet has trouble lining up the price and quantity with the equilibrium point).

Why name our AI tutor Harriet? Harriet Martineau was a 19th Century social theorist, author, and populariser of economics ideas. She wrote a famous book in 1832, titled Illustrations of Political Economy, which drew on the work of Adam Smith, and applied a tutorial style to help readers understand Smith's ideas. Her later books included similar treatments of economic ideas by David Ricardo, James Mill, and Jeremy Bentham. Harriet Martineau is a perfect model of what we wanted our ECONS101 AI tutor to be - someone who could explain economic concepts in an accessible way. So, we named our AI tutor after Martineau.

I can see from the server logs on Moodle that Harriet has already been accessed dozens of times by students, and we're only one day into the new trimester. No doubt most of those students are simply curious about what the AI tutor is. Hopefully, they will find her useful as we move through the trimester. And, if they identify explanations that Harriet gets wrong, we are even offering students a bounty (a small amount of extra credit) for letting us know, so that we can improve her training.

I'm looking forward to seeing how things progress. Harriet is one of several new initiatives we are trying in ECONS101 this year, to get students more engaged and improve their grades and their learning of business economics. My ECONS102 class can expect to have their own separate AI tutor in the B Trimester later this year, based on similar principles.

Japan releases some of its emergency rice stockpile

The New Zealand Herald reported last week:

The Japanese Government said on Friday it will release a fifth of its emergency rice stockpile after hot weather, poor harvests and panic-buying over a “megaquake” warning nearly doubled prices over a year...

Rice prices had already began to change consumption patterns for some like Tokyo resident Eriko Kato.

“I still do buy rice occasionally, but since it’s so expensive I sometimes give up on buying it once I see the price,” Kato, 41, told AFP...

Tokyo resident Kato says she “sometimes just switches to noodles like udon or soba instead” because rice is more expensive.

The changes in the market for rice in Japan are illustrated in the diagram below. Before the poor rice harvest and panic-buying, the market was in equilibrium with a price of P0, and Q0 rice being traded. The poor rice harvest decreases the supply of rice from S0 to S1, while panic-buying increases the demand for rice from D0 to D1. The combined effect of these two changes is an increase in the price of rice (from P0 to P1). The change in the equilibrium quantity of rice traded is ambiguous - it depends on the relative size of the shifts in supply and demand. On the diagram below, the equilibrium quantity remains at Q0. However, if the increase in demand had been larger than the decrease in supply, then the equilibrium quantity traded would have increased. And if the increase in demand had been smaller than the decrease in supply, then the equilibrium quantity traded would have decreased. So, the model only tells us that we can be sure that the price of rice will go up.

If the Japanese government releases some of its emergency rice stockpile, then that will increase the supply of rice in the market. Let's assume that the increase in supply shifts the supply curve back from S1 to S0. The result is a decrease in the equilibrium price of rice (to P2), and an increase the equilibrium quantity of rice traded (to Q2). Notice though that the price of rice hasn't fallen all the way back to the initial equilibrium price. The Japanese government would have to release a lot of its emergency rice supply for that to happen. Or, the panic-buying of rice would have to subside (which is probably what the government is hoping will happen when it releases the emergency rice).

In the meantime though, the changes in the rice market will affect other markets. Consider the market for noodles, shown below. Rice and noodles are substitutes. When the price of rice increases, some consumers will switch to buying noodles (as described in the quote from the Herald article at the start of this post). The noodle market was previously in equilibrium with an equilibrium price of P0, and an equilibrium quantity of Q0 noodles traded. Consumers switching from rice to noodles increases the demand for noodles (from D0 to D1). This increases the equilibrium price of noodles (from P0 to P1), and increases the equilibrium quantity of noodles traded (from Q0 to Q1).

So, higher rice prices affect Japanese consumers, regardless of whether they are buying rice or noodles! Fortunately, the release of the emergency rice, lowering the equilibrium price of rice, will likely also lessen the demand for noodles and lower the price of noodles as well.

This week in research #63

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

• Fudenberg (open access) summarises the work of last year's John Bates Clark medal winner, Philipp Strack
• Borusyak, Hull, and Jaravel (open access) provide a practical guide to shift-share instruments (also referred to as Bartik instruments)
• Thrane (open access) finds evidence of own-country bias in experts' player ratings in football (soccer)
• Henrekson, Jonung, and Lundahl (open access) argue that the tyranny of the top five economics journals is increasingly extending from the US to European universities
• Carpenter et al. (open access) find large employment and earnings penalties for transgender and gender diverse people in New Zealand, using data from driver licence records linked to income in Stats NZ Integrated Data Infrastructure (I wonder if this can be corroborated with data from the 2023 Census, which now collects more detailed data on gender identity?)
• Solga, Rusconi, and Hofmeister (open access) look at whether gender bias in assistant professor recruitment exists and differs across disciplines (mathematics/physics, economics/sociology/political science, and German studies) in Germany, and find that in all disciplines female applicants receive higher ratings than male applicants, both for perceived qualification for an assistant professorship and for being invited for an interview
• Cao et al. find that firms with accountant CEOs are associated with lower innovation output and a weaker propensity to pursue explorative innovations but greater efficiency in generating innovation output per unit of resource input
• Aigner, Greenspon, and Rodrik (open access) look at the universe of economics and business journal articles published since 1980, and find that while Western and Northern European authors have made substantial gains, the representation of authors based in low-income countries remains extremely low, and that articles by developing country authors are far less likely to be published in top journals even when holding constant article quality

Avian flu and US egg prices

The New Zealand Herald reported earlier this month:

A resurgence of avian flu, which first struck the United States in 2022, is hitting chicken farms hard, sending egg prices soaring and rattling consumers accustomed to buying this dietary staple for only a few dollars.

In Washington and its suburbs, supermarket egg shelves are now often empty, or sparsely stocked. Some stores limit the number of cartons each client may buy. And everywhere, consumers are shocked by the high prices.

“They’re getting expensive,” 26-year-old student Samantha Lopez told AFP as she shopped in a supermarket in the US capital. “It’s kind of difficult ... My budget for food is already very tight.”...

More than 21 million egg-laying hens have been euthanised this year because of the disease, according to data published Friday by the US Agriculture Department. Most of them were in the states of Ohio, North Carolina and Missouri... 

The department reported the “depopulation” of a further 13.2 million in December.

Higher prices were the inevitable result, experts say.

“If there’s no birds to lay eggs ... then we have a supply shortage, and that leads to higher prices because of supply and demand dynamics,” said Jada Thompson, a poultry specialist at the University of Arkansas.

Let's consider the market for eggs, and how the shortage arose and why that means higher egg prices. This is all illustrated in the diagram below. Before avian flu, the market for eggs was in equilibrium, with a price of P0 and a quantity of eggs traded of Q0. The culling of chickens due to avian flu means that fewer eggs are produced. This is a decrease in the supply of eggs, shown by the supply curve shifting up and to the left, from S0 to S1. If egg prices were to remain at the original equilibrium price (P0), then the quantity of eggs demanded (Q0) would exceed the quantity of eggs supplied (QS) at that price, because egg producers are only willing to produce QS eggs at the price of P0, after the supply curve shifts. There would be a shortage of eggs, which is one of the things that we observe in the US egg market.

However, when there is a shortage, the market will tend to adjust. In this case, the market will adjust through the price of eggs increasing. How does that happen? Some buyers, who are willing to pay the market price (P0), are missing out on eggs. Some of them will find a willing seller, and offer the seller a little bit more, in order to avoid missing out. In other words, buyers bid up the price. The result is that the price increases, until the price is restored to equilibrium, at the new (higher) equilibrium price of P1. At the new equilibrium price of P1, the quantity of eggs demanded is exactly equal to the quantity of eggs supplied (both are equal to Q1). We can say that the market clears. There is no longer a shortage.

So, this model of supply and demand tells us that, because of avian flu, we should expect to see shortages of eggs (at least initially), and overall higher egg prices. Which is what we are observing.

Effects of the minimum wage on the nonprofit sector

After a few days of 'rest' (by which I really mean some intensely long work days), I'm going to pick up again on my recent series of posts about the minimum wage (see here for the most recent post), but returning to more familiar ground - the disemployment effects of the minimum wage. The story we tell using basic supply and demand is that a minimum wage that is above the equilibrium wage in a labour market will reduce the number of jobs (reduce the quantity of labour demanded by employers). 

Considering this post from last week (along with others), we should now be recognising that the simple story is incomplete, because there are other margins that employers may adjust along. They might not reduce jobs, but they might change some of the non-monetary characteristics of jobs, for example. Employers may also absorb some of a higher minimum wage in the form of higher costs, and reduced profits.

Minimum wages don't only affect for-profit firms though. They also affect nonprofit firms. And non-profit firms have one margin that they cannot adjust - profits. A non-profit firm cannot sustainably absorb higher costs by accepting lower profits. So, we might expect to see larger disemployment (and other) effects on firms operating in the nonprofit sector.

That is essentially what this 2023 article by Jonathan Meer (Texas A&M University) and Hedieh Tajali (University of Edinburgh), published in the journal Oxford Economic Papers (ungated earlier version here), looks at. Specifically, Meer and Tajali use data from electronic charity filings to the US Internal Revenue Service (IRS) from 2011 to 2017, as well as data for the same years from the Quarterly Census of Employment and Wages, collected by the Bureau of Labour Statistics (BLS). The two different data sources paint a generally similar picture, but the IRS data offers somewhat more detail for the analyses.

Meer and Tajali then look at the difference in impacts of higher state-level minimum wages, differentiating (similar to what Clemens and Strain did in the paper I discussed here) between states that had large minimum wage rate changes (more than US$2), small minimum wage rate changes (less than US$2), and indexed rates (that change annually in step with inflation). Notice though that the threshold between large and small changes is $2, rather than the $1 that Clemens and Strain used. I'm unsure if that is material, but they don't present any alternative analyses based on other thresholds.

Meer and Tajali find that, using the IRS data:

Large statutory changers have a statistically significant 7.1% (s.e. = 2.3%) decrease in employment relative to nonchangers. But states with smaller minimum wage increases see little impact on employment... States with inflation-indexed minimum wages also see a negative effect despite relatively small increases.

The effect for small (less than $2) minimum wage increases is a statistically insignificant decrease in employment of 1.6 percent, while for indexers a minimum wage increase is associated with a 2.3 percent decrease in employment. Those are quite substantial effects (particularly for large minimum wage increases). The BLS data shows that:

...states with large statutory increases see a 2.7% (s.e. = 1.1%) decrease in employment relative to states that did not increase their minimum wage. Small statutory increasers see a negative but imprecise effect, while there is no meaningful impact on indexers.

So the effects are smaller using the BLS data. However, the BLS data:

...only includes organizations with an employee covered by unemployment insurance, it does not include nonprofits without paid workers...

You might think that shouldn't make much difference, but a higher minimum wage will also affect employers who don't pay any of their workers (because those workers are volunteers), because for some of those volunteers a higher minimum wage represents a better 'outside option'. So, with higher minimum wages, nonprofit firms might lose some workers to for-profit firms (or to other nonprofit firms) that are paying the now-higher minimum wage. That would mean that there would likely be larger disemployment effects showing up in the IRS data than in the BLS data. However, I doubt it explains a large proportion of the difference, and in any case, both datasets show statistically significant disemployment for nonprofit firms when minimum wage increases are large.

Moving on to other aspects of nonprofit firms, Meer and Tajali find (using the more-detailed IRS data) that large minimum wage changes are associated with lower grant receipts, less fundraising expenditures (which might partially explain the lower grant receipts), and lower total expenses. Looking at different sizes of nonprofit firms, they find that:

The smallest nonprofits, with three or fewer employees (including those that are entirely volunteer-run), are the most affected. Aggregate employment in this size bin is 25.0% (s.e. = 15.6%) lower in states with large statutory changes relative to nonchanging states. Estimates for other size categories are negative but not statistically significant.

Altogether, it is apparent that employment in nonprofit firms is negatively affected by large increases in the minimum wage, which is consistent with my impression of the overall literature on employment generally.

[HT: Marginal Revolution, back in 2023]

Read more:

• A bunch of research findings on minimum wages in New Zealand
• The disemployment effects of Canadian minimum wages
• Jeffrey Clemens on the disemployment effects of the minimum wage
• The impact of an online minimum wage
• Seattle's minimum wage, revisited
• The latest evidence supports negative employment effects of the minimum wage
• Latest research suggests the minimum wage DOES reduce employment
• More empirical support for the disemployment effects of the minimum wage
• The minimum wage and job vacancies
• The minimum wage and hiring standards
• The employment effects of the minimum wage, from American Community Survey data
• The minimum wage and homelessness
• New York restaurants find a new way to respond to the minimum wage
• The minimum wage and workplace safety
• How employers respond to minimum wage increases
• Minimum wage increases and employer noncompliance
• Higher minimum wages and poverty revisited
• Minimum wages and health

Minimum wages and health

Picking up again on the theme of last week's posts about recent research on the minimum wage, this 2024 article by David Neumark (University of California-Irvine), published in the journal Labour (open access), reviews the literature on the impacts of minimum wages on health and health behaviours. It's somewhat of a systematic review, although it doesn't closely follow the PRISMA reporting guidelines. Nevertheless, it is a helpful summary of the literature relating minimum wages to health, which is important in light of statements such as this one from the American Public Health Association, claiming unambiguously that higher minimum wages would improve health.

As you might expect the reality is somewhat more nuanced. Neumark starts by pointing out why the effect of higher minimum wages on health is theoretically ambiguous:

The potential for higher minimum wages to improve health is clear, as a higher minimum wage unambiguously raises incomes for some workers (and their families). On the other hand, job loss can reduce income among other workers and their families... it is entirely possible that health benefits from income gains for some workers outweigh adverse health effects for others who lose their jobs, perhaps because there are almost certainly more income gainers than job losers. This net gain might be more likely if there was clear evidence that minimum wages raise incomes in lower income families (rather than for low-wage workers). However, the evidence on family income is ambiguous, in part because many minimum wage workers are not in poor or low-income families, and many low-income families have no workers...

That latter point relates to my most recent post on the effect of minimum wages on poverty, covering research by Burkhauser et al. that demonstrated (as has been shown before) that only a minority of minimum wage workers live in poor families. Neumark's review covers 63 published and peer-reviewed articles, mostly using US data, and mostly published in the last decade. He separated his review into sections on adult and teen health, infant and child health, diet and obesity, mental health, suicide, family structure and children, risky behaviour, crime (which seems a little out of place, but many studies that consider risky behaviour also consider crime), and mechanisms that can affect health (like access to health insurance). Neumark briefly summarises each paper, notes some of the positives and negatives of the methods employed, and draws a conclusion about how convincing (or otherwise) each study is (generally on the basis of the methods employed).

There is a lot to unpack in the review, and I'm not going to try to summarise it all here. Instead, here's what Neumark says in the concluding section:

...the evidence, even focusing on the more-compelling studies (which I do), is decidedly mixed. The evidence on overall physical health points in conflicting directions, and may lean toward adverse effects—possibly a reflection, in part, of the conflicting influences of minimum wages on factors that can affect health (related to how higher income is spent). In particular, research on the effects of minimum wages on diet and obesity sometimes points to beneficial effects, whereas other evidence indicates that higher minimum wages increase smoking and drinking and reduce exercise (and possibly hygiene). In contrast, there is rather strong evidence that higher minimum wages reduce suicides, perhaps partly consistent with the evidence on effects on other measures of mental health/depression being either positive or mixed.

Going a little farther afield, research on minimum wage effects on family structure and children indicates that mothers spend more time with children, provides no clear indication of changes in treatment of children, but point to declines in children's test scores—clearly a mixed picture. There are many good studies of the effects of minimum wages on crime, but the conclusions are mixed. Turning to channels of influence on health (most notably, health insurance), the stronger evidence points to declines in employer-provided health insurance, and other adverse effects on potential influences on health, but there is no clear evidence of effects on unmet medical needs.

When Neumark narrows his focus only to those studies where the evidence is most convincing, he concludes that:

...the mixed conclusions on how minimum wages affect health and related behaviors undermine the evidence base for concluding that the minimum wage is an effective means of improving health.

However, one thing that this review highlights is the comparative lack of research on the effect of minimum wages on health, particularly in comparison to, say, studies on the effect of minimum wages on labour market outcomes (of which there are many). It also highlights that few studies, even relatively recent studies, perform even basic supplementary analysis such as placebo checks on the effects of minimum wages on groups unlikely to be affected by higher minimum wages (such as those with high education), and many studies over-control by including unemployment, income, or poverty in their analyses. Clearly, there is substantial scope for additional research in this space. Indeed, in a footnote to the paper, Neumark notes that:

Effects of minimum wages on drug use, perhaps particularly opioids, could impact health and suicides (as well as other outcomes). This would be a natural question to consider. However, I have not found any such evidence.

So, not only is there scope to improve on the extant studies, there is also scope for studies on areas of health that have not been considered to date. Clearly, there will be more research to come on this theme.

Read more:

• A bunch of research findings on minimum wages in New Zealand
• The disemployment effects of Canadian minimum wages
• Jeffrey Clemens on the disemployment effects of the minimum wage
• The impact of an online minimum wage
• Seattle's minimum wage, revisited
• The latest evidence supports negative employment effects of the minimum wage
• Latest research suggests the minimum wage DOES reduce employment
• More empirical support for the disemployment effects of the minimum wage
• The minimum wage and job vacancies
• The minimum wage and hiring standards
• The employment effects of the minimum wage, from American Community Survey data
• The minimum wage and homelessness
• New York restaurants find a new way to respond to the minimum wage
• The minimum wage and workplace safety
• How employers respond to minimum wage increases
• Minimum wage increases and employer noncompliance
• Higher minimum wages and poverty revisited

This week in research #62

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

• Bai and Veall (open access) find no evidence that minimum wages increase drinking overall, using data from Canada (and if I have time to read it, I'll blog about it as part of my current series of posts on minimum wage research)
• Randerson et al. (open access) look at eight case studies of alcohol licensing decisions in New Zealand, and find that structural barriers, including racism, restricted the influence of under-resourced communities and Māori in licensing decisions and weakened risk assessment
• Norström, Nilsson, and Svensson (open access) find that a 10% increase in bar density in Sweden would result in a 0.46% increase in nighttime police calls, and that in high-socioeconomic areas the effect was not significant, while the effect was twice as strong in the low-socioeconomic areas as in mid-socioeconomic areas

Finally, I spend today and yesterday at the New Zealand Economics Forum. I wasn't one of the speakers this year, so I could enjoy the proceedings from the floor. You can watch recordings of the sessions now. The Day One video is here, with the video from the breakout session on gene technology here. The Day Two video is here, with the video from the breakout session on Kiwi schools here. Enjoy!

Higher minimum wages and poverty revisited

One of the key purposes of a minimum wage is to decrease poverty. However, there is no certainty that poverty would be reduced by a higher minimum wage. In the simplest sense, we can think about several effects of higher minimum wages on poverty. Income goes up for those earning the minimum wage, and therefore poverty may decrease. However, if there is any disemployment (workers losing their jobs) resulting from the minimum wage, poverty may increase. Poverty may also increase if there is significant pass-through of higher minimum wages into prices (as noted in this post), increasing the cost of living for all households. So, whether higher minimum wages increase or decrease poverty, or don't affect poverty at all, is essentially an empirical question (and one that I have written about before).

Until relatively recently though, there was some general consensus among economists that the minimum wage is ineffective at reducing poverty. Aside from the offsetting effects I noted above, the minimum wage isn't really well targeted at the poor (consider, for example, the number of teens from relatively high-income families who work in jobs paying the minimum wage).

So, the lack of an effect of minimum wages on poverty was generally agreed on by economists. Until this 2019 article by Arindrajit Dube, published in the American Economic Journal: Applied Economics (open access), which claimed to show that there was a large effect of higher minimum wages on poverty in the US between 1983 and 2012. That Dube article got a lot of press at the time, and encouraged support for a much higher federal minimum wage, as encapsulated in the US Raise the Wage Act of 2021 (which died in Committee).

Dube's article also shook the consensus among economists, and thus it also attracted attention from other researchers. And unsurprisingly, some have looked closely at the analyses. This 2023 NBER Working Paper by Richard Burkhauser (Cornell University), Drew McNichols (Amazon), and Joseph Sabia (San Diego State University) is one such effort. As they explain:

This study revisits the relationship between minimum wage increases and poverty. We highlight four key results. First, we replicate and reassess the findings of Dube (2019), based on poverty data from the March 1984 to March 2013 CPS (corresponding to calendar years 1983-2012). After precisely replicating his estimates, we show that his results are driven by two specification choices: (1) the inclusion of macroeconomic controls (the state unemployment rate and per capita state Gross Domestic Product) that may also capture a mechanism through which the minimum wage affects poverty: its employment and hours effects, and (2) restricting treatment states’ counterfactuals to states within the same census division (“close controls”), even when geographically proximate states are rejected by a data-driven synthetic control approach to generate counterfactuals. When we (1) use the state house price index and the unemployment and average wage rate among more highly educated individuals to control for state macroeconomic conditions that are less likely to capture pathways through which minimum wages affect poverty in a difference-in-differences framework, or (2) allow states outside a treatment state’s census division to serve as potential donors in a synthetic control framework, we find no evidence of poverty-reducing effects of the minimum wage over the 1983-2012 period... The 95 percent confidence intervals around our preferred estimates rule out poverty elasticities with respect to the minimum wage of less than -0.138, which include central estimates reported by Dube (2019).

In other words, Burkhauser et al. show that Dube's results are not robust to several modelling choices that Dube made. When the models are run with different control variables, or with a different selection of control states, there are no negative effects of higher minimum wages on poverty. Also:

...when we explore the most recent decade of CPS data, which captures the years following the Great Recession (2010-2019), the contemporaneous and longer-run poverty findings reported by Dube (2019) are largely absent, including in models that use Dube’s preferred macroeconomic controls or controls for spatial heterogeneity. Specifically, we find no evidence that post-Great Recession minimum wage increases had a statistically significant or economically important effect on poverty.

So, Burkhauser et al. show that Dube's results are sensitive to the choice of the time period that the dataset covers. And then:

...when we combine the two data windows discussed above and amass our “full panel” from 1983-2019, we find little support for the hypothesis that minimum wage increases reduce poverty over this 37-year period. Estimated elasticities below -0.131 for non-elderly individuals (and below -0.129 for all persons) lie outside of our 95% confidence interval, which would rule out the central long-run estimate reported by Dube (2019). Our preferred estimate shows that a 10 percent increase in the minimum wage is associated with a (statistically insignificant) 0.17 percent increase in the probability of poverty among all persons.

So, using the broader dataset from 1983 to 2019, the effect of higher minimum wages on poverty is small and statistically insignificant. Finally, Burkhauser et al. reiterate earlier findings in the literature, by showing that:

...less than 10 percent of those whose hourly wage rate would be directly impacted by a $15 minimum wage live in poor families. Approximately two-thirds live in families with incomes over two times the poverty line and nearly half live in families with incomes over three times the poverty line.

Burkhauser et al. conclude that:

In summary, our findings provide little compelling evidence that raising the minimum wage will be an effective or target efficient policy tool for alleviating poverty.

Burkhauser et al.'s paper is a comprehensive and systematic take-down of Dube's earlier work. And, it reestablishes the earlier consensus - higher minimum wages do not reduce poverty (at least, in the US - the paper I discussed in this earlier post showed some short run, but not long run, effects on poverty in Brazil). As with all research, it pays not to overcorrect greatly on the basis of a single new research paper. Policy makers would do well to remember that.

[HT: Marginal Revolution, back in 2023]

Read more:

• A bunch of research findings on minimum wages in New Zealand
• The disemployment effects of Canadian minimum wages
• Jeffrey Clemens on the disemployment effects of the minimum wage
• The impact of an online minimum wage
• Seattle's minimum wage, revisited
• The latest evidence supports negative employment effects of the minimum wage
• Latest research suggests the minimum wage DOES reduce employment
• More empirical support for the disemployment effects of the minimum wage
• The minimum wage and job vacancies
• The minimum wage and hiring standards
• The employment effects of the minimum wage, from American Community Survey data
• The minimum wage and homelessness
• New York restaurants find a new way to respond to the minimum wage
• The minimum wage and workplace safety
• How employers respond to minimum wage increases
• Minimum wage increases and employer noncompliance

Minimum wage increases and employer noncompliance

As noted in yesterday's post, one of the ways that employers may respond to an increase in the minimum wage is to increase noncompliance (by paying a wage that is below the new, higher, minimum wage). Is there evidence for employers changing their behaviour on that margin? 

This 2022 article by Jeffrey Clemens (University of California at San Diego) and Michael Strain (American Enterprise Institute), published in the journal Labour Economics (open access), provides some evidence. They look at wage data from the Current Population Survey, where:

The first key piece of information is an indicator for whether respondents are paid on an hourly basis. When they are, respondents are asked for their hourly wage rates. When they are not, hourly wage rates can be inferred by dividing an individual’s usual weekly earnings by his or her usual weekly hours.

Using this survey data, Clemens and Strain are able to determine whether the hourly wage rate for each person in the survey is below the minimum wage in their state, suggesting that their employer was noncompliant with the minimum wage. To avoid overstating the amount of noncompliance, they only consider noncompliance in the case where the wage is more than 25 cents per hour below the minimum. 

Now, because of the potential for measurement error and other biases, the absolute level of noncompliance is not instructive. Instead, Clemens and Strain look at the changes in noncompliance when the minimum wage rate changes. They distinguish between large minimum wage rate changes (more than US$1), small minimum wage rate changes (less than US$1), and indexed rates (that change annually in step with inflation). Focusing their main analyses on workers aged 16 to 25 (who are most likely to be affected by changes in the minimum wage), they find that:

...a one-dollar increase in the minimum wage predicts, on average, a wage gain of 29 cents and a 3.6 cent increase in underpayment... On average across the country, we estimate that each dollar of minimum wage increase would, if applied nationally, have generated an increase in subminimum wage payment of roughly $1.16 billion and an increase in realized wage gains, among the employed, of roughly $6.86 billion. Our results thus suggest that compliance with the minimum wage is the norm, but that avoidance and evasion are nontrivial.

Comparing the different types of minimum wage rate change, Clemens and Strain find that there are:

...far larger increases in subminimum payment following increases that were enacted through new legislation than following increases that came about due to pre-existing laws that call for annual, inflation-indexed updates to states’ minimum wages.

In other words, when minimum wage rate changes are small and expected and can be planned for (as indexed minimum wage rate changes are), then noncompliance is low. However, when minimum wage rate changes are unexpected, and especially if they are unexpected and large, there are significant increases in noncompliance.

Finally, Clemens and Strain look at the effect of enforcement regimes, and find that:

...increases in the minimum wage predict larger increases in the prevalence of subminimum wage payment under strong enforcement regimes than under weak enforcement regimes.

This is quite a counter-intuitive result, which they explain in relation to models of enforcement, where:

...strong enforcement regimes encourage reporting and compliance when the minimum wage is low, such that enforcement is in the worker’s interest. If the minimum wage rises beyond the level an employer would be willing or able to pay, however, enforcement may cease to be in the worker’s interest. As the minimum wage crosses this threshold, reductions in compliance will tend to be larger under strong enforcement regimes because these were precisely the regimes in which workers had an incentive to report noncompliance at baseline.

So, states get higher noncompliance in strong enforcement states because higher minimum wages turn those states into weak enforcement states. In contrast, noncompliance is already higher in weak enforcement states and enforcement there cannot get weaker. I'm not sure that I find this argument convincing, as the way that the enforcement regimes (strong or weak) are defined in the data is not related to workers' actions, but to the penalties that minimum wage violators face, and the authority of enforcement agencies. These aspects would not necessarily be affected by the change in the minimum wage rate.

Aside from that, Clemens and Strain are very careful in presenting various robustness checks based on different samples of workers, and different types of models (as you would expect, if you read the article by Clemens that I referred to in yesterday's post). They are also quite worried by the potential for measurement error in wages (arising from people misreporting their wage, or their hours worked), but their additional analyses suggest that measurement error is not likely to be driving their results.

What we can take away from this article is that, again, employment is not the only margin along which employers adjust to a higher minimum wage. Some (not all) employers may respond by not complying with the higher minimum wage, leaving their workers working for a lower wage (but still working, at least). If the increase in noncompliance is large, then that would reduce any negative employment impacts of a higher minimum wage.

Read more:

• A bunch of research findings on minimum wages in New Zealand
• The disemployment effects of Canadian minimum wages
• Jeffrey Clemens on the disemployment effects of the minimum wage
• The impact of an online minimum wage
• Seattle's minimum wage, revisited
• The latest evidence supports negative employment effects of the minimum wage
• Latest research suggests the minimum wage DOES reduce employment
• More empirical support for the disemployment effects of the minimum wage
• The minimum wage and job vacancies
• The minimum wage and hiring standards
• The employment effects of the minimum wage, from American Community Survey data
• The minimum wage and homelessness
• New York restaurants find a new way to respond to the minimum wage
• The minimum wage and workplace safety
• How employers respond to minimum wage increases

How employers respond to minimum wage increases

In yesterday's post, I made reference to this 2021 article by Jeffrey Clemens (University of California at San Diego), published in the Journal of Economic Perspectives (open access). Clemens puts forward an interesting perspective on the debate about the observed employment impacts of the minimum wage (or lack thereof):

...I contend that controversies over the economics of minimum wages stem, to a surprising degree, from a common but under-considered assumption. The assumption of interest is that when studying labor markets, output prices and nonwage aspects of jobs (which include benefits and working conditions) can be taken as fixed. In standard diagrams of the labor market, this assumption implicitly underlies each supply or demand curve. When these curves are held fixed, output prices and nonwage aspects of jobs have also, whether implicitly or explicitly, been held fixed...

...textbook models of both perfectly and imperfectly competitive labor markets sweep many factors under the rug. Benefits, including employer-provided health insurance, account for around one-third of compensation costs... Working conditions, including safety measures and flexible schedules, can also have value to workers while generating costs to firms... In the jargon of undergraduate instruction, the ceteris paribus assumption (that is, “other things held constant”) that professors invoke when we draw labor supply and demand curves is unlikely to describe the real world.

Clemens then goes on to show how changes in the non-wage aspects of jobs affect employment, using the standard supply and demand model. I'm going to reproduce that here, but with a couple of modifications. First, I'll show an increase in the minimum wage in each diagram (Clemens shows changes without changing the minimum wage). Second, I'll explain the changes using the language I use in my ECONS101 and ECONS102 classes, to make it more familiar for my students (past and future - my first ECONS101 class for 2025 starts two weeks from today).

The standard demonstration of the effect of an increase in the minimum wage is shown in the diagram below. The equilibrium wage, W0, is the wage where the quantity of labour supplied (Q0) is exactly equal to the quantity of labour demanded (Q0). Economists say that the market clears. Every worker who wants to work for the wage W0 is able to find a job that pays W0. Now consider the effect of a binding minimum wage, WMIN1, which is higher than the equilibrium wage. The quantity of labour supplied increases to QS1, because more workers want to work (or because already-employed workers want to work even more) at the higher wage. The quantity of labour demanded decreases to QD1, because employers want to employ fewer workers as they are now more expensive. The difference between QS1 and QD1 is structural unemployment arising from the minimum wage. The decrease in employment as a result of the minimum wage is the difference between Q0 and QD1. If the government increases the minimum wage from WMIN1 to WMIN2, then the quantity of labour supplied increases even more (to QS2), and the quantity of labour demanded decreases even more (to QD2). Structural unemployment increases (to the difference between QS2 and QD2), and the employment impact of the minimum wage increases as well (to the difference between Q0 and QD2). The employment impact of the higher minimum wage (compared to the lower minimum wage) is the difference between QD1 and QD2.

Now consider alternative firm responses to an increase in the minimum wage. The first example that Clemens uses is pass-through of the increased costs onto consumers:

If demand for a firm’s output is not perfectly elastic, it can raise prices while losing some, but not all, of its customers. A price increase in response to a minimum wage increase is often called pass-through: that is, the minimum wage’s cost passes through the firm to its consumers. An increase in output prices implies an outward shift of the labor demand curve...

The demand curve shifts out to the right because, if the firm increases the price that it sells its output for, each worker now produces more value for the firm, and so the firm will want to hire more workers. This is shown in the diagram below. At the same time that the minimum wage increases from WMIN1 to WMIN2, the demand for workers increases from D0 to D3. Now, although the quantity of labour supplied still increases to QS2, the quantity of labour demanded only decreases to QD3. Structural unemployment increases, but by less (to the difference between QS2 and QD3), and the employment impact of the higher minimum wage (compared to the lower minimum wage) decreases to the difference between QD1 and QD3. 

Notice that firms' ability to pass-through the higher minimum wage onto consumers in the form of higher prices will make it more difficult to detect any employment impact. Not all firms can pass-through higher costs equally. Clemens notes that the price elasticity of demand for the firm's output matters:

Firms that produce widely traded goods or services may face large demand elasticities and thus have little capacity to raise prices. By contrast, firms that produce “nontradable” goods and services may face smaller demand elasticities and have more substantial scope for passing cost increases to consumers... Standard examples of non-tradables include beauty services, meals at restaurants, and home construction, which are more or less constrained to be provided where they are consumed. Pass-through may also depend on whether the minimum wage is increased at the city, state, or federal level. When a minimum wage increase is localized, there is greater scope for importing products from unaffected firms.

The second example that Clemens uses is 'non-cash compensation', which is essentially the additional benefits that firms provide to workers. These may include health insurance, contributions to retirement savings, travel allowances, a company vehicle, and so on. If firms respond to higher minimum wages by reducing the value of non-cash compensation, then this:

...can shift both the supply curve and the demand curve...

From the perspective of firms, lower noncash compensation implies a higher labor demand curve because it increases revenues net of nonwage costs. From the perspective of workers, lower noncash compensation implies a higher labor supply curve, since a higher wage is required to make employment attractive when nonwage benefits are lower.

These combined effects are shown in the diagram below (and we are now ignoring any pass-through of the higher minimum wage to consumers). The demand curve increases to D4, and the supply curve decreases to S4. The overall effect of the higher minimum wage, compared with the lower minimum wage, is no effect on either the quantity of labour supplied (which remains QS1) or the quantity of labour demanded (which remains QD1), and so there is no effect on structural unemployment and there is no effect of the higher minimum wage on unemployment (compared with the lower minimum wage). However, this outcome is purely an artefact of how I have chosen to draw the diagram (as was the case in Clemens' article). If the effect of lower non-cash compensation was larger for workers than shown in the diagram, then the shift in supply will be larger, and structural unemployment will increase (although the employment impact of the minimum wage would still be unaffected). The reverse is true for a smaller effect for workers. And, if the effect of lower non-cash compensation was larger for employers than shown in the diagram, then the shift in demand will be larger, and structural unemployment will increase (and the employment impact of the minimum wage would be negative). The reverse is true for a smaller effect for employers (and in that case the employment impact of the minimum wage would be positive). [*]

Finally, Clemens discusses employer changes in 'productive disamenities' and 'unproductive amenities':

Conceptually, a firm facing minimum wage increases might seek to offset some of the increase in costs by raising productive disamenities (like effort requirements) and reducing unproductive amenities (like the quality of office furniture). As with changes in noncash compensation, these changes will shift both the supply curve and the demand curve...

Increases in productive disamenities... imply upward shifts in the labor demand curve (due to an increase in marginal product) and upward shifts in the labor supply curve (due to an increase in disutility from work effort).

I'm not going to draw a new diagram for this situation, as the third diagram above already shows the effect of increasing the demand for labour (which would happen if firms increased productive disamenities) and decreasing the supply of labour (which would happen if firms decreased unproductive amenities). Clemens notes that there is little empirical evidence 

...little, if any, empirical evidence on the minimum wage’s effects on scheduling, workplace safety, workplace comfort, and other related margins.

However, Clemens' article was published in 2021, and the literature has moved on. These firm responses relate directly to the new empirical evidence I discussed in yesterday's post about workplace safety. The Davies' et al. paper showed an increase in workplace injuries, and suggested that it arose because of firms demanding increased effort from workers (an increase in a productive disamenity). The Liu et al. article showed an increase in workplace injuries, and instead suggested that it arose because firms were spending less on workplace safety (a decrease in an unproductive amenity).

Finally, Clemens turns to other adjustments that firms might make, including changes in profits, firms shutting down, changes in job design and production technologies, and firms' compliance with the minimum wage (the latter of these will be the subject of my post tomorrow). It is clear that there are many ways that firms can respond to a higher minimum wage, and that changes in employment are not the only relevant margin that researchers should be considering. In fact, changes on these other margins may cause researchers to erroneously conclude that there are no employment effects of a higher minimum wage, when they solely focus on the number of jobs.

[HT: Marginal Revolution, which reminded me that I hadn't read the Clemens article]

*****

[*] I haven't drawn all of these alternative scenarios, but I am confident that you can imagine smaller or larger shifts in the demand or supply curves, and their effects. Otherwise, this would turn into an even longer post than it already is!

Read more:

• A bunch of research findings on minimum wages in New Zealand
• The disemployment effects of Canadian minimum wages
• Jeffrey Clemens on the disemployment effects of the minimum wage
• The impact of an online minimum wage
• Seattle's minimum wage, revisited
• The latest evidence supports negative employment effects of the minimum wage
• Latest research suggests the minimum wage DOES reduce employment
• More empirical support for the disemployment effects of the minimum wage
• The minimum wage and job vacancies
• The minimum wage and hiring standards
• The employment effects of the minimum wage, from American Community Survey data
• The minimum wage and homelessness
• New York restaurants find a new way to respond to the minimum wage
• The minimum wage and workplace safety

The minimum wage and workplace safety

This week, I'm going to write several posts about the minimum wage, as I've recently read a bunch of research related to it. It's a topic I've written on many times before (see the links at the end of this post). It's also a research topic where the evidence is far from settled. At least in terms of the employment effects, there is still active debate, although I conclude (as does Jeffrey Clemens) that the totality of the evidence points towards a reduction in employment for those most directly affected.

However, reducing the number of jobs is not the only way that employers may respond to a higher minimum wage. This 2021 article by Clemens outlines a number of margins where employers may change behaviour in response to a higher minimum wage. I'll write about that Clemens article tomorrow, because today I want to focus in on one particular margin where employers may adjust - workplace safety.

How might a change in the minimum wage affect workplace safety? As this new paper by Michael Davies (MIT), Jisung Park (University of Pennsylvania), and Anna Stansbury (MIT) notes:

Whether minimum wages help or harm workplace safety is theoretically unclear. Higher minimum wages might induce employers to cut costs, reducing spending on safety measures, or to push for productivity gains by intensifying the pace of work. Either of these would lead to more injuries. Alternatively, higher minimum wages could lead to safer workplaces by reducing turnover in low-wage labor markets, through efficiency wage-type effects, or by incentivizing capital intensification (which often improves workplace safety).

Davies et al. use data on all workplace worker's compensation insurance claims from California over the period from 2000-2019 (over 13 million claims), and how those claims are related to changes in state and local minimum wages. More specifically, they:

...regress the log annual injury rate of a metro-occupation labor market on the interaction between the minimum wage change (real year-on-year growth in the metro minimum wage) and the metro-occupation’s exposure to that minimum wage change (the estimated share of the wage distribution that falls below 1.3x the local minimum)...

That allows them to focus in on the change in workplace injuries for workers working at or close to the minimum wage. Davies et al. find that:

...minimum wage increases lead to significantly higher injury rates. Our headline estimate is that a 10% increase in the local minimum wage increases the injury rate by 11% in an occupation which is fully exposed to the minimum wage change (i.e. where all workers earn less than 1.3x the minimum).

How meaningful is that change? Davies et al. report that:

The average injury rate for low-wage workers in our data is 4.4% per year; applying our coefficient estimate to these workers, we estimate that a 10% increase in the minimum wage leads to a 0.3 percentage point increase in the injury rate, or on average an additional 3 injuries per 1,000 low-wage workers per year.

If they focus on particular occupations, there are (as seems obvious) some occupations where the effects on workplace injuries are even larger. Davies et al. then go on to explore the mechanisms that may drive the increase in workplace injuries. Is it because employers reduce spending on safety measures, or because they push their workers to work harder? Davies et al. conclude that it is the latter, based on looking at specific injuries that are related to both physical exertion and repeated stress (such as carpal tunnel syndrome or repetitive strain injury). Specifically, they find that:

...the effect of minimum wages is nearly twice as large for these cumulative physical injuries: a 10% minimum wage increase leads to a 21% increase in cumulative physical injuries per worker.

Davies et al. aren't able to adequately test for the reduction in spending on safety mechanism, so it could be that both are contributing to the effect of minimum wages on workplace safety.

However, Davies et al. aren't the only researchers looking at this question. Another new article, by Qing Liu (Renmin University), Ruosi Lu (University of International Business and Economics in Beijing), Stephen Teng Sun (City University of Hong Kong), and Meng Zhang (Capital University of Economics and Business in Beijing), published in the Journal of Public Economics (sorry, I don't see an ungated version online), also looks at minimum wages and workplace safety, but using different data and methods. Specifically, Liu et al. use data from the Occupational Safety and Health Administration (OSHA) over the period from 2002 to 2011:

Each year OSHA surveys approximately 80,000 private-sector establishments, which are required to maintain records of any work-related injury or illness that results in death, loss of consciousness, days away from work, restricted work activity or job transfer, or medical treatment beyond first aid.

Liu et al. match those OSHA employer records with data from CompuStat to identify the parent firms of each establishment, and so their analysis is essentially looking at variation in workplace injuries across establishments within the same parent firm. They focus on 'large' minimum wage changes (of $1 or more), and apply a two-way fixed effects (TWFE) estimator, and separately apply a stacked difference-in-differences research design. Using TWFE, they find that:

...a large increase in state minimum wages raises workplace injury rates by 4.6 percent relative to the sample mean. With an average establishment size of 222 employees, our point estimate implies an additional 0.8 injury cases per year, or 3.2 cases over the first four years post-treatment.

Two-way fixed effects have some well-known problems. When they apply the stacked difference-in-differences approach, Liu et al. find very similar results. Turning to mechanisms, they find that:

...the adverse effect is more pronounced for establishments affiliated with financially constrained parent firms or in industries highly dependent on external finance.

Liu et al. argue that this suggests that financial constraints are the dominant channel driving the increase in workplace injuries, consistent with employers responding to higher minimum wages by decreasing spending on workplace safety. They argue that there is little evidence for changes in work effort, based on an analysis that shows that illnesses are not affected by the minimum wage. Davies et al., however, criticise that analysis because:

...the illnesses these authors cite as motivating examples — “stress, depression, heart diseases, and strokes”... —are conditions for which employers hold considerable discretion in deciding whether to report them to OSHA ((CFR 29 C.F.R. § 1904.5(b)(2)(ix)); and many illnesses for which OSHA mandates reporting are not those which one would expect to be responsive to work intensification (e.g. respiratory disease, hearing loss).

So, what we can take away from these two studies is that higher minimum wages are associated with higher rates of workplace injury, which is an important point that I will return to tomorrow.

[HT: Marginal Revolution, for the Davies et al. paper]

Read more:

• A bunch of research findings on minimum wages in New Zealand
• The disemployment effects of Canadian minimum wages
• Jeffrey Clemens on the disemployment effects of the minimum wage
• The impact of an online minimum wage
• Seattle's minimum wage, revisited
• The latest evidence supports negative employment effects of the minimum wage
• Latest research suggests the minimum wage DOES reduce employment
• More empirical support for the disemployment effects of the minimum wage
• The minimum wage and job vacancies
• The minimum wage and hiring standards
• The employment effects of the minimum wage, from American Community Survey data
• The minimum wage and homelessness
• New York restaurants find a new way to respond to the minimum wage