The supply of, and demand for, child brides in India

One of the targets embedded within Sustainable Development Goal #5 is to eliminate child marriage. However, the practice of child marriage remains surprisingly high in some countries. UNICEF noted in 2018 that:

...approximately 650 million girls and women alive today were married before their 18th birthday. While the global reduction in child marriage is to be celebrated, no region is on track to meet the Sustainable Development Goal target of eliminating this harmful practice by 2030.

So, what explains the practice of child marriage? This 2017 article by Peter Leeson (George Mason University) and Pablo Suarez (New York University), published in the Journal of Economic Behavior & Organization (ungated version here), looks at child brides in India. Leeson and Suarez adopt a framework grounded in supply and demand. Specifically:

Our theory is grounded in son preference: parental taste for sons over daughters, common in developing countries. In trying to produce sons, son-preferring couples sometimes produce daughters. To afford the sons they want, some of these couples must dispose of their unwanted daughters, one way of which is to marry them off prematurely, creating a supply of prepubescent brides.

Son-preferring couples invest fewer resources in the care of their young daughters than their young sons, so more males survive to traditional marriage age than females. To find brides in the face of this sex ratio imbalance, some traditional marriage-aged men must reach into younger female cohorts, requiring them where that imbalance is severe to reach into prepubescent cohorts, creating a demand for prepubescent brides.

So, a preference for sons rather than daughters leads to both a supply of child brides as well as a demand for child brides. Leeson and Suarez use data from India to test their theory. India is a useful case study because India is "one of the most son-preferring and child-bride populous nations in the world". There is also substantial variation in son preference and child brides across states in India:

Prepubescent brides are most common in the northern and central regions of the country and less common in the south and northeast. In the northern state of Rajasthan, for example, in 1993, more than 17% of women who had ever married did so before puberty; in the southern state of Tamil Nadu, less than 0.5% did so...

Also like child brides, son preference in India is most pronounced in the north and least pronounced in the south. In the northern state of Rajasthan, for example, in 1993, the average ever-married woman’s ideal son/daughter ratio was nearly 1.7; in the southern state of Tamil Nadu, it was just 1.15. 

The key facts in the quote above already point to a correlation between son preference and the incidence of child brides, since both are higher in northern states, and lower in southern states. However, Leeson and Suarez put the theory to the test more rigorously, using data from the 1992-1993 National Family Health Survey (NFHS). While the NFHS data are cross-sectional, they construct a pseudo-panel dataset made up of different five-year age cohorts of women. That allows Leeson and Suarez to control for:

...unobserved cultural differences relating to son preference or marriage practices between people in different states and between older people and younger ones.

Controlling for underlying time-invariant differences between states, and common time trends across states, is important. Based on this setup, Leeson and Suarez find that:

...stronger son preference is associated with the birth of more unwanted daughters, younger postpubescent-female age at marriage, and a higher incidence of prepubescent brides. Moreover, son preference has a stronger positive association with prepubescent brides where poverty is more extreme...

This supports their theory that son preference drives an increase in supply of, and demand for, child brides. That highlights the challenge of achieving the target in SDG #5. Son preference is a cultural characteristic that may take years, if not generations, to alter. And India has one of the world's most imbalanced sex ratios, which will be a further contributing factor. Unfortunately, the Leeson and Suarez paper is silent on how countries might address the problem of child marriage. Prohibition isn't working, and I'd hate to think that we simply need to wait for generational change. Perhaps achieving some of the other targets in SDG #5, which focuses on the empowerment of women and girls, will lead to spillover reductions in child marriage? We can only hope.

How increased wages over the last century reduced the incentive to work

In my ECONS101 class this week, I had to rush through the last bit of the lecture. So, I thought it might be handy to outline the last example we were working on in a bit more detail. It relates to the stylised facts outlined in the figure below, which comes from Unit 3 of The Economy (the textbook we use in ECONS101), but the original data comes from Robert Fogel's book The Fourth Great Awakening and the Future of Egalitarianism.

The figure shows the estimated lifetime hours of discretionary time (24 hours per day, minus the time spent sleeping, eating, and on personal hygiene) in blue, and breaks discretionary hours down into lifetime work hours (in red) and lifetime leisure hours (in green), for 1880, 1995, and a projection for 2040. The interesting thing here (aside from the increase in discretionary hours over time), is the big increase in leisure hours between 1880 and 1995, and reduction in work hours. That change in leisure and work happened in spite of a substantial increase in real wages over the same period.

These facts should come as a little bit of a surprise. That's because we expect that, when wages are higher, people should be working more, not less. In other words, we expect the supply of labour to be upward sloping. The reason we expect people to work more is because, as wages rise, the opportunity cost of leisure increases - every hour spent in leisure means that the worker gives up more income than before. When the opportunity cost of doing something increases, we expect people to do less of it. But in this case, as the opportunity cost of leisure has increased, workers are consuming more leisure, not less.

To understand and explain this puzzle, we need a model. In this case, it is a model of the work decision of workers. In the work decision, the worker is asked to trade off between two goods: leisure, and consumption (or income). The worker wants more leisure, but for every hour of leisure, they give up an hour of working, which means their income (and consumption) will be lower. If the worker wants higher income, they have to give up an hour of leisure.

Our model of the worker's decision is outlined in the diagram below. The worker's decision is constrained by the amount of discretionary time available to them. Let's call this their time endowment, E. If they spent every hour of discretionary time on leisure, they would have E hours of leisure, but zero income. That is one end point of the worker's budget constraint, on the x-axis. The x-axis measures leisure time from left to right, but that means that it also measures work time (from right to left, because each one hour less leisure means one hour more of work). The difference between E and the number of leisure hours is the number of work hours. Next, if the worker spent every hour working, they would have zero leisure, but would have an income equal to W0*E (the wage, W0, multiplied by the whole time endowment, E). That is the other end point of the worker's budget constraint, on the y-axis. The worker's budget constraint joins up those two points, and has a slope that is equal to the wage (more correctly, it is equal to -W0, and it is negative because the budget constraint is downward sloping). The slope of the budget constraint represents the opportunity cost of leisure. Every hour the worker spends on leisure, they give up the wage of W0. Now, we represent the worker's preferences over leisure and consumption by indifference curves. The worker is trying to maximise their utility, which means that they are trying to get to the highest possible indifference curve that they can, while remaining within their budget constraint. The highest indifference curve they can reach on our diagram is I0. The worker's optimum is the bundle of leisure and consumption where their highest indifference curve meets the budget constrain. This is the bundle A, which contains leisure of L0 (and work hours equal to [E-L0]), and consumption of C0.

Now, consider what happens when the wage increases. Real wages (adjusted for inflation) increased a lot between 1880 and 1995. This is shown in the diagram below. For simplicity, let's assume that the worker's time endowment remained the same [*]. As the wage increases, the worker's budget constraint pivots outwards and becomes steeper. It has the same intercept on the y-axis x-axis (equal to the time endowment E), but if the worker spent every discretionary hour working, they would now have consumption (and income) equal to W1*E, where W1 is the new (higher) wage. Notice that the new budget constraint is steeper. This is because the slope of the budget constraint is equal to the wage, and the wage is now higher (and a higher value for the slope means a steeper line). The worker can now reach a higher indifference curve (I1). Their new optimal bundle of leisure and consumption is B, which contains leisure of L1 (and work hours equal to [E-L1]), and consumption of C1. Notice that the worker now consumes more leisure and more consumption as well. Because leisure has increased, that means that the number of work hours has decreased.

Coming back to the key facts we started this post with, the model does clearly show that, when wages go up, the worker will consume more leisure (and work less). Where did our original intuition go wrong?

Our original explanation, that as the wage increases the opportunity cost of leisure increases, and the worker will choose to work less, is the substitution effect of the change in the wage. However, when the relative price of two goods changes, the substitution effect is not the only thing that affects constrained decisions. There is also an income effect. In this case, the income effect says that, when wages go up, the worker can afford more consumption, but can also afford more leisure. Since both leisure and consumption are normal goods (we demand more as our income increases), the worker will want more leisure (and will want to work less). Over the period from 1880 to 1995, the income effect must have been larger than the substitution effect, and has driven the decrease in work hours (and increase in leisure hours).

Does this mean that increasing wages will always induce workers to work less? Of course not. The increase in real wages over the period from 1880 to 1995 was huge. And, no individual worker working in 1880 was still working in 1995, so no worker actually experienced the change we represented in our model. Most wage changes are much more modest, and it does appear that labour supply curves are generally upward sloping at the level of wages we observe over the short run. What that means is that the income effect is likely to be much smaller than what our diagram shows (and that would mean that L1 would be slightly to the left of L0. I'll leave drawing that situation as an exercise for you.

*****

[*] We know this isn't true, but it helps to make this assumption because we are more interested in the effect of the change in wages, and not the effect of the change in time endowment.

Economic illiteracy vs. Turkish inflation

The Turkish economy is in real trouble. As news.com.au reported yesterday:

While the rest of the world has tightened monetary policy to deal with the global surge in inflation, Turkey’s central bank has kept its cash rate stubbornly unchanged since January.

In fact it actually cut the rate significantly in the final third of 2021, lowering it from 19 per cent to the 14 per cent at which it now remains.

The central bank has adopted this approach under sustained pressure from Turkey’s President, Recep Tayyip Erdogan, who believes higher interest rates actually fuel inflation.

The conventional economic wisdom is the other way around. It holds that high interest rates restrain inflation, and looser monetary policy inflames it.

Speaking in May, the President defended his gamble and branded those who were concerned about Turkey’s monetary policy “illiterates”.

“Those who try to impose on us a link between the benchmark rate and inflation are either illiterates or traitors,” he said.

Unfortunately, it is the Turkish leader who is the economic illiterate. Low interest rates fuel inflation. That's because when interest rates are low, borrowers need to spend less on interest payments, so that leaves them with more money to spend on other goods and services. Also, businesses can borrow more cheaply to spend on investment goods. Being able to spend more sounds like a good thing, but increasing investment spending, and increasing consumption spending, both increase demand for goods and services. Without a corresponding increase in production, there is essentially more money chasing the same number of goods, and prices start to go up. In other words, inflation increases when interest rates are low. And so, you end up with this:

A year ago, Turkey’s inflation rate was an already troubling 19 per cent. By the end of last month, it had risen to a staggering 79 per cent, the highest it has been in 24 years, and 16 times the central bank’s inflation target of 5 per cent.

Turkey isn't alone in facing high inflation. But it is a substantial outlier:

Incidentally, those viewing the world from London are currently experiencing an inflation rate of 9.4 per cent. Over in New York it is 9.1 per cent.

That’s far more inflation than either country would like – hence, the US Federal Reserve is poised to announce a fresh interest rate hike this week – but it’s preferable to 80 per cent...

According to data released by the Bureau of Statistics this morning, Australia’s consumer price index (CPI) rose by 1.8 per cent in the June quarter, with our annual inflation rate increasing to 6.1 per cent...

For comparison, New Zealand's inflation rate for the year to June 2022 was 7.3 percent, which was the highest level since 1990. That has both economists and lay people worried, but nowhere near as much as inflation of 80 percent would! Those worries arise because there are a bunch of costs associated with high inflation, including:

• Shoe leather costs - the costs to consumers of holding less cash so that they are not exposed to their cash losing value, as well as the cost of time spent searching for the lowest current prices (they are called shoe leather costs because consumers would make frequent trips to the bank to get money, and spend a lot of time searching around for prices, thereby wearing out their shoes);
• Menu costs - the costs to firms of having to constantly adjust prices (they are called menu costs because if your firm is a restaurant, you have to print all new menus when you change prices);
• Arbitrary redistributions of wealth - such as from lenders (who are being paid back in money that is worth less than before) to borrowers;
• Tax distortions - as taxpayers get pushed into higher tax brackets due to their increasing nominal wages; and
• Confusion and inconvenience - since it makes it difficult for anyone to figure out what the current prices are.

Those costs all tend to reduce economic activity, and those costs are going to be far higher in Turkey now than they were before. And all because Turkey's President followed his own (incorrect) ideas about how interest rates affect inflation. I guess we can see who the real economic illiterate is.

There is one small benefit to the current high inflation. Economics teachers will now have students who have actually seen some appreciable inflation, which hasn't been the case for many years. We'll now be able to point to recent experience in our classroom examples.

Bar opening hours, alcohol consumption, and workplace accidents

Does alcohol increase workplace accidents? It seems pretty obvious that it must. If workers are drunk on the job, they are more likely to make mistakes and injure themselves or others. But what about drinking the night before work? What effect does that have on workplace accidents? And, if we are worried about workplace accidents arising from late-night drinking, would closing bars earlier help the situation?

That is the research question addressed in this 2018 article by Nicolau Martin Bassols (Universitat Pompeu Fabra) and Judit Vall Castello (Institut D’economia de Barcelona), published in the journal Labour Economics (sorry, I don't see an ungated version online). They use data on Spain, where:

Spain is divided into 17 regions (Autonomous Communities). Each of these regions implemented the reduction in bar opening hours at some point between 1994 and 2011. Before the reform, bars in Spain were allowed to open until 6am. This was reduced to 2am–3.30am, depending on the region.

There was quite some variation in the timing of the change in bar opening hours, and it is the difference in timing that Bassols and Vall Castello exploit to estimate the effect of the change in opening hours on workplace accidents. They apply a difference-in-differences analysis, which involves comparing two differences: (1) the difference between states before the policy change; and (2) the difference between states after the policy change. Of course, it isn't quite that simple as the states adopted the policy at different times (a point I will return to later in the post).

The main data that Bassols and Vall Castello use (on bar opening hours and workplace accidents) covers the period from 1990 to 2011, which allows them to have at least four years before the policy was introduced, and at least four years after the policy was introduced, for every state except one (Catalunya, which didn't adopt the policy until the end of 2011). They also use some survey data from the Spanish Family Expenditure Survey and the Spanish Health Survey, each of which covers a shorter time period.

Looking at the workplace accident data, Bassols and Vall Castello find that:

...the coefficients that capture the impact of the policy on workplace accidents for employed individuals are all negative. Furthermore, the effect is significant for both men and women as well as individuals working in the service sector. When we focus on the total population, we can see that the reduction in bar closing hours decreased the number of working accidents in 2.62 per 100.000 employed individuals. When we compare the size of the effect with the mean of working accidents in the sample (which is 28.03 for the total population) we can see that the policy caused a reduction in workplace accidents by 9%. For woman, the policy reduced working accidents by 15% and for man by 7%. For the services sector the reduction in working accidents amounted to 11%.

However, once they control for region-specific time trends, the effects almost all go away, and:

...all the results show the similar sign but smaller magnitudes. However, only the coefficient for the construction sector is significant. Not only the coefficient is significant but the size of the effect for this sector of the economy is really big: we estimate that the reduction in bar closing hours decreased working accidents in 6.60 accidents per 100.000 workers, which implies a reduction by 18.3%.

In other words, the only coefficient of interest that was statistically significant without the region-specific time trends included in the model, is the only coefficient that is statistically significant with those trends. That doesn't strike me as particularly strong evidence for an effect of bar opening hours on workplace accidents. Bassols and Vall Castello haven't adjusted their results for multiple comparisons, so the statistical significance they report could have come up purely by chance. Also, there is a further problem with the analysis. The 'two-way fixed effects' approach that they have adopted has recently attracted a lot of criticism (which is nicely outlined in two posts on the Development Impact blog, here and here, as well as this post). The short version is that the two-way fixed effects approach is likely to lead to biased estimates of the treatment effect - in this case, it would lead to a biased estimate of the effect of bar opening hours on workplace accidents. So, even putting aside that the results are statistically insignificant, they are likely to be biased as well.

Nevertheless, Bassols and Vall Castello press on and show with their other data that:

...expenditure in bars was reduced by 13% as a result of the advancement in bar closing hours. As expected, none of the coefficients capturing the effects of the reform on other durable goods is significant...

This comparison strikes me as somewhat odd. Ok, the effect on expenditure in bars is statistically significant and negative, but why compare it with expenditure on durable goods? Why not compare it with spending on other forms of entertainment (which would be subject to similar underlying consumption trends, only differentiated by the change in bar opening hours)? While at face value the results suggest that people spent less in bars after bars started closing earlier, perhaps they were also spending less on other entertainment as well. We don't know, because we never see the comparisons. Looking at self-reported consumption of alcohol, they find that:

...the coefficients for daily and weekly consumption of wine are significant and the effects are stronger for men... The regressions for the consumption of whisky, liquor, aperitifs and mixed drinks as well as the probability of smoking are also negative but non-significant.

Again, those results are underwhelming, and could easily have arisen by chance, given that it is only wine consumption, and only for men (and not for women or overall) that is statistically significant. Finally, looking at the health data (specifically the rate of hospitalisations caused by excessive alcohol consumption), Bassols and Vall Castello find that:

...all the coefficients of the diff-diff variable are negative but not significant and that the strongest reduction in alcohol-related hospitalizations is reported again for men.

When they limit the data to weekends only, they find:

 ...a significant negative coefficient for men pointing towards important reductions in alcohol-related hospitalizations during the weekends as a result of the reduction in bar opening hours. More precisely, the reduction in hospitalizations for men is by 0.28 per 100.000 individuals which represents a reduction by 16.5% with respect to the mean...

Overall, I'd say that this paper provides only very weak evidence for a relationship between bar opening hours and workplace accidents. The only result that is even mildly convincing is that weekend hospitalisations fall for men, by 16.5%. Note that isn't hospitalisations resulting from workplace accidents though - it is hospitalisations overall. Also, with the bias introduced as a result of the two-way fixed effect approach, even that result should be treated with due caution. What we should take away from this is that we would need a lot more convincing evidence before we concluded that bars should close earlier in order to reduce workplace accidents.

Causality vs. correlation in the relationship between alcohol and suicide

Last week in my ECONS101 class, we covered the difference between causation and correlation. It is important to recognise that just because two variables appear to be related, that does not mean that a change in one of the variables will cause a change in the other variable. Even if we can tell a convincing story that a change in Variable A causes a change in Variable B, the relationship we observe in the data might not arise instead because:

• a change in Variable B actually causes the change in Variable A (reverse causation);
• a change in some other variable (Variable C) causes the changes in both Variable A and Variable B (confounding, or a common cause); or
• the two variables are completed unrelated and the relationship showed up by chance (spurious correlation).

Understanding the difference between correlation and causation is important, because if you create some policy based on the false conclusion that changing Variable A will change Variable B, but the relationship is not actually causal, then you won't get the change you expected or wanted.

Now, although it is a simple maxim to say "correlation is not causation", in practice it is actually quite difficult to avoid attributing causality to correlational data. A lot of research uses language that implies causality, like "the impacts of X on Y" or "the determinants of Z", even when the research design is really only showing a correlation (I know I've done this myself). So, we need to take extra care in interpreting results to avoid people drawing the wrong conclusions. And the media, in particular, are fond of over-stating conclusions. Take this article in Stuff from earlier this month:

One in four New Zealanders who die by suicide do so with excess blood alcohol – with the figures even worse for Māori and Pasifika, new analysis of coronial data has found.

The findings raise major red flags and should prompt urgent changes to Aotearoa's suicide prevention strategy and the Sale and Supply of Alcohol Act, according to the authors, from the University of Otago.

New Zealand’s “alcohol-saturated culture" meant the numbers were disheartening, but not surprising, study author and lecturer Dr Rose Crossin said.

The original research paper, by Rose Crossin (University of Otago at Christchurch) and co-authors, published in the New Zealand Medical Journal, is here (gated). It finds that:

...around one quarter (26.6%) of all suicides over the study period involved acute alcohol use.

We also found significant ethnicity differences, with Māori and Pacific peoples more likely to die by suicide involving acute alcohol use than European and Asian ethnicities.

Now, even though a good story can be told about how alcohol use reduces inhibitions and may lead to increased risk of suicide, there is nothing in this study that demonstrates that link. And, no matter how much we might abhor the statistics, they still don't (by themselves) show that alcohol use causes suicide.

Why is this study not demonstrating causality? For one thing, there is a potentially confounding third variable that is not accounted for - mental health state. People in poor mental health are more likely to use alcohol excessively. They are also more likely to commit suicide. So, high alcohol use may be incidental to suicide, because both are caused by a third variable (mental health). Now, that doesn't mean that alcohol is completely off the hook as a cause, only that there may be other explanations that haven't been excluded in this study. In fact, in the coronial data that they use, the coroners themselves indicate that alcohol was "a contributory cause of death" in about one-third of those that involved acute alcohol use, or about 9 percent of all suicides.

Notice also from the Stuff article quote above that the authors are advocating for policy change, based on the correlations observed in their study. There is good reason to believe that we need a change in the way that alcohol sale and supply is controlled, and my past (and possibly, future) research has contributed to this. However, this study on its own does not contribute much to the case for reform. Even if we assumed that changes in alcohol regulation eliminated suicides with alcohol as a contributory cause, that would reduce suicides by only 9 percent. That sounds like a worthwhile goal. Indeed, any reduction in the suicide rate would be a great victory. However, before we adopt the policy change on the basis that it would reduce suicide risk, we'd want to know how much (if any) of the correlation arises from a causal relationship. And that's not something we know right now.

Social norms, defaults, and tipping taxi drivers

When you come from a culture (like New Zealand) where tipping is not a social norm, understanding when and how much to tip in a tipping culture (like the US) can be a bit of a mystery. I admit that my tipping behaviour is strongly driven by the menu options that are available on screen when I pay by credit card. If the screen suggests 15%, 20%, and 25%, I'll often choose one of those. That way, I don't have to try and work out how much the tip should be. Choosing one of the default menu options lowers the cognitive cost of the tipping decision. Besides, if it wasn't a social norm to tip those percentages, why would they display them on-screen?

Obviously, they are important for me, but how important are the on-screen menu options in general? That is essentially the question addressed in this 2021 working paper (with a non-technical summary here) by Kwabena Donkor (Stanford University). Donkor has data on over 250 million Yellow Taxi trips in New York City between 2010 and 2018, but only uses a randomly-selected subset of that data (a 20% sample of the data from 2014, and a further 10% sample from the full dataset) to make the task more manageable. To understand how the default menu options affect tipping behaviour, Donkor looks at:

...variation in tipping behavior a year before and a year after the CMT [Creative Mobile Technologies] default tip menu changed on February 8, 2011. The menu showed default tips of 15%–20%–25% then changed to 20%–25%–30%.

If taxi customers were purely rational, then the default options should not affect tipping behaviour at all. They would automatically choose the size of the tip that was 'best' for them. However, as we discussed in my ECONS102 class this week, people are not purely rational, and are affected by (among many other things), how a decision is framed. If you change the default tipping options, then you are changing the context of the decision. In turn, this may affect tipping behaviour. Indeed, Donkor finds that taxi customers are not purely rational. That is:

...the average tip rate increased from 17.45% of the taxi fare to 18.84% (an 8% increase), but the share of default tips decreased from 58.39% to 47.13% (a 19.3% decrease). However, the share of passengers who do not tip stayed [the] same (no extensive margin adjustments).

So, changing the default increased the amount of tipping on average, but as the defaults increased in value, fewer people chose the default. Interestingly, this result supports both rational and quasi-rational behaviour. It was quasi-rational because changing the default menu options affected the amount of tipping. It was rational because, as choosing the default costs more, people do a little bit less of it.

Donkor then looks at what happens when you shift from three default options to five default options (as happened in 2017). Comparing 2016 and 2018 data, he finds that:

...default tips increased by 11.5% (up from 59% to 66%).

Perhaps that means that more people were able to see a social norm tip that they could agree with, when there were more options available? This reduces the cognitive cost of calculating an alternative tip, and so more people choose one of the (five) default options.

There were also some other interesting results that Donkor notes in the paper:

The norm tip increases by 24% on New Year’s, 6% on Thanksgiving, and 18% on Christmas. Norm conformity increases by 89% on New Year’s Day, by 42% on Christmas day, but not significantly on Thanksgiving. The norm tip increases by 5% and 1%, respectively, when it snows or rains. However, norm conformity does not change much during bad weather.

Traveling with co-riders reduces the norm tip by 3.5% and norm conformity by 12.3%. This finding aligns with the bystander effect: when traveling in a group, no one person feels directly responsible for the guilt of not tipping or paying a low tip.

The takeaway message overall is that people tend to follow social norms because it is less costly to do so than the alternative. The costs of not following a social norm are not monetary. They fit into the category of 'social incentives' (which Levitt and Dubner note in their book Freakonomics are the incentives that arise from other people thinking that something is right or wrong). However, as the costs of following the social norm rise, more and more people will choose not to follow the norm. So, fortunately there is a limit to how high taxi companies can push the default tips in the menu before they induce too many customers to opt out.

[HT: Marginal Revolution, earlier this year]

Only a minority of real people may actually be loss averse

I've written a couple of posts this week about loss aversion (see here and here). However, loss aversion is not uncontested in the research literature. In fact, the research by Gal and Rucker that I discussed in this 2018 post argued that there was "little evidence to support loss aversion as a general principle". One way of thinking about this is that Gal and Rucker are arguing that not everyone is loss averse. And that is likely true, in the same way that not everyone is risk averse, and not everyone is averse to pineapple on pizza.

A new working paper by Jonathan Chapman (University of Bologna), Erik Snowberg (University of Utah), Stephanie Wang (University of Pittsburgh), and Colin Camerer (Caltech) provides some more evidence for this. In fact, they don't just show that some people are not loss averse. They show that about half of people may in fact be 'loss tolerant'.

Chapman et al.'s main results are based on a sample of 1000 people who completed a survey with the survey panel provider YouGov in 2020. The specific method that they used is quite detailed, but essentially involved 20 different 'gambles', with each gamble using information from the earlier gambles to provide a nuanced understanding of each research participant's attitudes towards risk and towards loss. This Dynamically Optimized Sequential Experimentation (DOSE) method provides estimates for both risk aversion and loss aversion for each research participant.

Importantly, the sample of research participants in the YouGov survey is representative of the underlying US population. Chapman et al. contrast the results from the representative sample with those from a smaller sample of 437 students from the University of Pittsburgh. This comparison is important, because most experimental economics samples are based on student populations (and it has been shown before that student samples are meaningfully different to representative population samples in economics experiments - for example, see here).

For the Chapman et al. paper, the key results are demonstrated in their Figure 3:

Looking at the blue distribution, there are some people in the general population sample who are loss averse (λ>1), but also a lot of people who are loss tolerant (λ<1), as well as some people in the middle. In terms of raw numbers, 57% of the general population sample is loss tolerant. For the student sample, again there is a distribution where some are loss tolerant, but a far higher proportion are loss averse. In the student sample, just 32% are loss tolerant.

So, what is it that makes the student population so much more loss averse than the general population? Chapman et al. show that:

...more educated and more cognitively-able individuals - both characteristics of student samples... - tend to be more loss averse and also less risk averse.

So, university students may be more loss averse because they have higher cognitive ability than the general population and are more educated. That may be good reason to think carefully about whether student samples are necessarily always the best choice to economics experiments.

However, Chapman et al. don't stop there. They then look at why it is that less cognitively-able and less educated people are more loss tolerant, hypothesising that:

...the groups that tend to be more loss tolerant - the less educated, lower income, and less cognitively able - are also those that we might expect to have encountered more losses in life. This raises the intriguing possibility that loss tolerance is shaped by everyday experiences.

And that is what they find:

...loss-tolerant individuals appear more likely to gamble, commit a greater portion of their assets to equities, experience financial shocks, and have lower overall wealth...

So, what should we take away from this research? First, not everyone is loss averse. In fact, a majority of people may be loss tolerant. That doesn't mean that loss aversion is irrelevant for understanding individual decision-making. It just means that we should not assume that everyone is loss averse. Second, we need to take care in extrapolating from student samples in economics experiments to the general population. This is not a new finding (as I noted above), but it is important that we don't lose sight of it. Third, and probably most important, when people routinely experience losses as part of their everyday life, they become more loss tolerant (and less loss averse). That may or may not be a good thing. After all, we talk about loss aversion as a deviation from purely rational decision-making. Being less loss averse may not be a bad thing. On the other hand, if loss tolerance leads to greater losses in the future, that may require a policy response. On this last point, we really need more research.

[HT: Ranil Dissanayake]

Loss aversion and the endowment effect in health-seeking behaviour

When I teach loss aversion in my ECONS102 class, I raise one of the consequences of loss aversion as the endowment effect. The explanation is fairly simple. When people are loss averse, they value losses much greater than otherwise equivalent gains. Giving something up therefore makes people very unhappy, and so people prefer to hold onto the things that they have. That means that, when a person owns something, they have to be given much more to compensate them for giving it up than what they would have been willing to pay to get it in the first place.

However, it turns out that loss aversion may not be the best (or only) explanation for the endowment effect. In a new NBER Working Paper (ungated version here, with a non-technical summary here), Emily Beam (University of Vermont), Yusufcan Masatlioglu (University of Maryland), Tara Watson (Williams College), and Dean Yang (University of Michigan) look at how people respond to a $50 incentive to attend a health service provider, when it is framed as a loss versus when it is framed as a gain. More specifically:

In this study, we implement a randomized field experiment that compares loss versus gain framing to promote preventive health care utilization. We offer individuals in and near Dearborn, Michigan, an incentive to visit a health clinic run by our partner organization, the Arab Community Center for Economic and Social Services (ACCESS). In the “Visa gift card” (loss framing) treatment, participants are given a Visa gift card of either $50 or $10 that can be activated by visiting the clinic; they will effectively lose the value of the card if they choose not to visit a clinic. In the “reminder card” (gain framing) treatment, participants are given a physically similar generic reminder card with the promise that it will be exchanged for a gift card if they visit the health clinic, but they are not given the gift card up front. In both cases, any individual who went to the health clinic would receive an active Visa gift card, and any funds remaining after the visit could be spent elsewhere. Because of random assignment to the treatments, differences in responsiveness to the incentives are attributable to the differences in the frames.

Research participants who were given a Visa gift card essentially face a loss if they choose not to activate it. That's because, once they finish the initial survey, they have the card in hand and choosing not to activate it is like losing $50. The other participants only receive a reminder card, which can be converted into a gift card. So, if they don't go to the clinic, they aren't losing in the same way. So, if there is an endowment effect, we'd expect those who received the gift card to be more likely to visit the clinic.

However, that isn't really what Beam et al. are looking at. They are investigating why there is an endowment effect. So, in the initial survey, they ask questions that are designed to provide an estimate of how loss averse people are. If the endowment effect is related to loss aversion, then the effect should be larger for people who are more loss averse. However, the endowment effect might also arise because of trust. As Beam et al. explain:

A second possible explanation for the effectiveness of loss framing is that giving an incentive up front induces an individual to have more confidence that the incentive will be provided as promised. The perceived probability of receiving a reward is likely higher for someone who has a tangible reward in hand relative to someone hearing about a promised reward. This trust‐related response is likely more relevant in field contexts outside the lab, and it is expected to be most relevant when individuals do not initially trust the person or institution offering the incentive.

If the endowment effect arises because of trust issues, then it should be larger for people who report less trust in the health care organisation, ACCESS. So, armed with measures of trust and loss aversion for around 1500 people (whose gift card was worth $50, and ignoring a smaller group whose gift card offer was only $10), and knowing which of the research participants visited the clinic (to receive or activate their Visa gift card), Beam et al. then find that:

The overall average difference in take‐up between those who receive the $50 Visa gift card (loss frame) and $50 reminder card (gain frame) is about 2.2 percent... The differences between Visa gift card and reminder card redemption rates are 4 to 5 percentage points for more loss‐loving participants and 1 to 2 percentage points for more loss‐averse participants. These results are not statistically significant... There is no clear pattern linking loss aversion to take‐up rates, nor to the gap in take‐up rates between gift card and reminder treatments.

So, there is a small endowment effect, but it isn't related to loss aversion. What about trust? Beam et al. find that:

Participants without trust of the organization at baseline are much more responsive to the gift card treatment (loss frame); the impact of the gift card is 7.2 percentage points for this group... The statistically significant interaction term... suggests that there is no comparable effect for those who do trust ACCESS at baseline.

In other words, there is an endowment effect for research participants who do not trust ACCESS, but no endowment effect for research participants who do trust ACCESS. When I first read those results, I was a little concerned that they arose only because Beam et al. combined people who reported low trust with people who had no opinion because they hadn't heard of ACCESS before. However, when those categories are separated, the results remain similar.

So, it really does seem that it is trust, and not loss aversion, that likely explains the endowment effect in this context. That doesn't necessarily mean that loss aversion is never a source of the endowment effect though, so I think I am safe (for the moment) in continuing to teach it as closely related to loss aversion.

The Beam et al. paper is also interesting in noting some of the real-world difficulties in research, especially this bit:

During our first survey wave, we encountered several safety issues: some interviewers were harassed by residents; on another day, interviewers witnessed gunfire a few blocks away. After these experiences, we excluded tracts that reported relatively high recent crime levels, and we contacted the Dearborn police department to exclude any additional tracts that they considered to be unsafe...

 Yikes! And slightly more mundane, this bit on how they had to adapt the measurement of loss aversion:

Although these questions are typically worded as a gamble, we adjusted the wording to be an “opportunity” after pilot testing revealed many subjects would reject all gambles because of religious objections to gambling.

Real-world field research is often not as straightforward as we hope it would be.

[HT: Ranil Dissanayake]

Loss aversion may affect how students answer multiple choice questions

In my ECONS102 class this week, among other things we covered a bunch of concepts in behavioural economics. One such concept was loss aversion - the idea that people value losses much more than equivalent gains (in other words, they like to avoid losses much more than they like to capture otherwise equivalent gains). Loss aversion seems to explain a lot of quasi-rational behaviour (however, as a concept, loss aversion is contested - more on that in another post later this week). People do seem to adjust their behaviour to try and avoid losses.

One example is the answering of multiple choice questions in tests and exams, but only where a wrong answer results in a penalty (subtracting marks from the overall test or exam score). I never grade multiple choice in that way, but many academics do. The argument they put forward is that, when incorrect answers are penalised, it creates a disincentive to students guessing. However, because the penalty creates a loss, it may be that loss averse students avoid answering when they are a little bit unsure, even if their not-quite-sure guess would have been the correct answer.

So, how much does loss aversion affect students' multiple choice answering behaviour? That is the question that this recent article by Heiko Karle (Frankfurt School of Finance and Management), Dirk Engelmann (Humboldt-Universität zu Berlin) and Martin Peitz (University of Mannheim), published in the Scandinavian Journal of Economics (ungated earlier version here), sets out to answer. Karle et al. use data from 646 students, combining an experimental-based measure of loss aversion (that the students completed early in the semester) with the results of a 30-question multiple-choice examination (held some three months later), where:

There are four possible answers to each question: a correct answer gives three points, no answer one point, and an incorrect answer gives zero points, as in the exam in our data set.

Notice that, on the face of it, there is no penalty for an incorrect answer, since an incorrect answer receives zero points. However, choosing not to answer yields one point, so considering not answering as the status quo reference point, choosing an incorrect answer makes a student worse off by one point (a loss, which they would try to avoid). Karle et al. hypothesise that students who are more loss averse will answer fewer questions, and will get more of the questions that they do answer correct. Both hypotheses are obvious from loss aversion - students try to avoid the loss, so are more likely not to answer questions when they are unsure, so more loss averse students will answer fewer questions. That means that the questions that more loss averse students do answer are those that they are surer about, and so they are more likely to get those answers correct. Karle also test a related hypothesis, that students who are more loss averse get fewer questions correct overall, which depends on how many the students answer (fewer, when the students are more loss averse) and how many of the ones they do answer are answered correctly (more, when the students are more loss averse). What to expect for this overall hypothesis is unclear.

In addition to loss aversion, Karle et al. measure students' self-confidence, which is based on the difference between students' estimates of the percentage of their own correct answers to a set of general interest questions and the average percentage of other students’ correct answers. Students who are more self-confident can be expected to be more likely to answer questions in the multiple choice exam.

Now, using their data, Karle find that:

...loss aversion and confidence have a negative and positive effect, respectively, on the number of answered questions... This effect is statistically significant at the 1 percent level... Our estimates suggest that loss neutral students answer approximately two questions more than otherwise identical students in the highest category of loss aversion (and 5/3 more than those in the middle category).

So, more loss averse students answer fewer questions than less loss averse students, as expected. Next:

We do not find a statistically significant effect (at the 5 percent level) of loss aversion or confidence on the ratio of correct answers per questions answered. However, the coefficient of loss aversion (but not strong loss aversion) turns positively significant at the 10 percent level when considering loss aversion and confidence together...

This is very weak support (or, rather, no support) for the hypothesis that more loss averse students get more of the questions that they do answer correct. Combining the two hypotheses, I'm sure you can guess for the overall hypothesis, that Karle et al. find that:

...loss aversion and confidence have a negative and positive effect, respectively, on the dependent variable [the number of correct questions overall]... Our estimate suggests that, ceteris paribus, students in the highest category of loss aversion give approximately 1.5 fewer correct answers than otherwise identical students who are loss neutral.

Karle et al. then spend a bit of effort trying to determine whether the effect of loss aversion on question answering behaviour is causal or not, and their results suggest that it is causal for some, but not all, students. I don't find those results as convincing as the overall takeaway that students' loss aversion is related to how they answer questions.

Nevertheless, these results are interesting even putting aside the question of causality. Loss aversion isn't something that is easy to change, so even a correlation between loss aversion and question answering behaviour is potentially important. And it may be even more important given that Karle et al. show that female students in their sample are more loss averse than male students. So, the effect of this style of grading multiple choice questions is a bias against female students, decreasing female grades relative to male grades. That's the last thing we need in economics. That may have contributed to this:

In our setting, according to a university directive, the differential treatment of wrong responses and no responses was no longer allowed after the academic year 2013/2014, which is the exam year we used in this paper.

Coming back to my earlier point, many academics like the style of grading that doesn't award 'free' marks to student guessing. However, there is a trade-off. If students are penalised for guessing, they are likely also penalised based on how loss averse they are. So, if multiple choice questions come with a penalty, the exam score will be more precise for each question (in terms of telling the grader whether students actually were confident in the answer) but also more biased overall (because more loss averse students will get fewer questions correct than less loss averse students). The trade-off seems untenable to me. I'll continue to use multiple choice marking that implicitly includes a reward for guessing.

Queensland weather and New Zealand vegetable prices

With both of my first-year economics classes due to start next week, it is timely to have some economics news that provides a simple application of the supply and demand model that is covered in both classes. As the New Zealand Herald reported yesterday:

Extremely wet weather across the Tasman means some fruit and veggies will be off the menu for customers here because importers are unable to get their hands on some types of produce.

Queensland fruit and vegetable growers are warning there would be reduced quality and increased prices over the next fortnight as weather events decimate crops.

ABC rural news reports Cross Family Farms, one of Australia's largest fruit and vegetable growers, was recording major losses after widespread rain across the state...

Back here, the head of United Fresh, Jerry Prendergast, said the knock-on effects of that meant some foods are off the menu for this season.

"It really has knocked product, like beans, courgettes, the melons out of Queensland, we normally expect to get strawberries at this time of year out of there too and it's been almost impossible," he said. 

Consider the market for beans, as shown in the diagram below. The market was initially in equilibrium, where demand D0 and supply S0 intersect, leading to an equilibrium price of P0 and an equilibrium quantity of beans traded of Q0. The bad weather in Queensland reduces the supply of beans to S1 (a decrease in supply is a shift of the supply curve up and to the left). The market now moves to a new equilibrium, where the demand curve D0 intersects the new supply curve S1. This leads to a higher price (P1) and a lower quantity of beans traded (Q1).

There will be flow-on consequences for the price of other vegetables too. From the article:

Prendergast said anyone with a hankering for beans should try Brussels sprouts instead, as there was plenty of them around.

If consumers switch to Brussels sprouts, then we can expect the price of Brussels sprouts to increase as well. This is shown in the diagram below. Before the increase in the price of beans, the Brussels sprouts market is operating in equilibrium, where demand D0 and supply S0 intersect, leading to an equilibrium price of P0 and an equilibrium quantity of Brussels sprouts traded of Q0. When beans become less available and more expensive, some consumers switch to buying Brussels sprouts instead, because beans and Brussels sprouts are substitutes. This increases the demand for Brussels sprouts to D1 (an increase in demand is a shift of the demand curve up and to the right). The market now moves to a new equilibrium, where the new demand curve D1 intersects the supply curve S0. This leads to a higher price (P1), and a higher quantity of Brussels sprouts traded (Q1).

So, even if only certain vegetables (beans) are directly affected by the bad weather in Queensland, it is likely to lead to higher vegetable prices across the range (including Brussels sprouts).

Full employment, bargaining power, and wages for low-income workers

I had an article published in The Conversation this morning, with the headline "NZ has reached ‘full employment’ – but not all workers will benefit from a tighter labour market". The article lacks some of the theoretical background behind some of the points that I make, and that theory relates to things that I will discuss with my ECONS101 class later this trimester, so I thought it would be worthwhile to outline them in a bit more detail here. In particular, this bit:

When there is full employment, it starts to become more difficult for employers to find workers to fill their vacancies. We are seeing this already, with job listings hitting record levels.

A tight labour market, where there are relatively more jobs than available workers, increases the bargaining power of workers.

But that doesn’t mean workers have all of the power and can demand substantially higher wages, only that workers can push for somewhat better pay and conditions, and employers are more likely to agree.

This shift in bargaining power is why some employers are now willing to offer significant signing bonuses or better work conditions and benefits, including flexible hours or free insurance.

This is based on the underlying theory of a search model of the labour market. A search model recognises that each matching of a worker to a job creates a surplus that is shared between the worker and the employer. This surplus is the difference between the additional value that the worker will create for the employer, and the search cost (essentially, the cost of finding and hiring the worker). Because job matching creates a surplus, and the employer wants that surplus, the worker has a small amount of bargaining power, with the employer. This is because, if the worker rejects the job offer, the employer has to start over in looking for someone else to fill the vacancy and will face additional search costs.

How is the surplus split between the worker and the employer? It depends on their relative bargaining power. The worker will have relatively more bargaining power (and will capture more of the surplus) if search costs are relatively higher for the employer than for the worker - for example, if the employer would find it harder to find an alternative worker, and/or the worker could more easily find an alternative job. This happens when unemployment is low, or when the skill requirements of the job are high (so that few other workers could do the job). The worker will also have more bargaining power if the costs of remaining unemployed are low (so the worker doesn’t so much mind saying no to a job offer). This happens when unemployment benefits are generous, or when the stigma or negative psychological costs of being unemployed are low.

Now, coming back to the situation of full employment, the unemployment rate is low (in fact, as I note in the article, the unemployment rate is currently the lowest it has been since 1986). Employers are finding it difficult to fill vacancies, so search costs are high. Both of these are factors that give workers more bargaining power, as I noted above. Workers can leverage that bargaining power to get higher wages and/or better work conditions.

Does that mean that all workers will be better off? No, and this is the key point I make in The Conversation:

Many low-income workers are in jobs that are part-time, fixed-term or precarious. Low-wage workers are not benefiting from the tight labour market to the same extent as more highly qualified workers.

Nevertheless, a period of full employment may allow some low-wage workers to move into higher paying jobs, or jobs that are less precarious and/or offer better work conditions. That relies on the workers having the appropriate skills and experience for higher-paying jobs, or for increasingly desperate employers to adjust their employment standards to meet those of the available job applicants.

Low-income workers tend to also be working in jobs that do not have high skill requirements. That means that there are more other potential workers who could fill those roles. In other words, it is a bit easier for an employer to find a worker to fill a low-skilled vacancy than to fill a high-skilled vacancy. That means that, even in times of low unemployment, low-income workers do not necessarily benefit from greater bargaining power (as noted above, workers have more bargaining power when the skill requirements of the job are higher). Or, at the least, low-income workers do not benefit to the same extent that higher-income workers do.

And it gets worse. In the article, I didn't make the point that when the cost of living is high (as it is now), low-income workers may feel like they have little choice but to accept low wage offers, because the alternative (no job and a low job-seeking benefit) is so much worse. That isn't a consequence of full employment, but rather that even though benefit rates are indexed so that they increase as the cost of living increases, that adjustment is always playing catchup.

 So, as I conclude in my article:

Although a full employment economy seems like a net positive, not everyone benefits equally, and we shouldn’t ignore that some low-wage workers remain vulnerable.

The gender gap in finance is (slightly) worse than in economics

The gender gap in economics is large and persistent. I've written a number of posts on this point, and on some of the initiatives that the profession is enacting to try and address the gap (see this post, and the links at the end of it). There is also a large gender gap in STEM (science, technology, engineering, and mathematics) subjects as well. However, one discipline where I haven't seen much discussion is finance.

That is addressed in this new working paper (with a non-technical summary here) by Renee Adams (University of Oxford) and Jing Xu (University of Technology Sydney). They use data from this paper by Ioannidis et al., which provides a database of standardised citations and other measures for over 160,000 scientists across 175 fields of science, in 20 disciplines. Interestingly, the list of fields includes economics (actually, several fields relate to economics) and finance. However, there is a bit of an issue with this data, as it does not include the gender of each scientist. Adams and Xu code the gender of those in the finance field manually, and use genderize.io to automatically assign gender to the others. This is a little problematic, as genderize.io doesn't deal well with some names (particularly Asian names). Adams and Xu drop those that cannot be gender classified with at least 90% certainty, leaving them with data on 126,403 scientists (including finance and economics).

With their dataset in hand, Adams and Xu then focus on the top two percent of researchers in each discipline, and compare metrics by gender. They find that:

Finance ranks 132nd out of 175 fields in terms of the representation of women among its top scientists. The percentage of women in Finance is lower than the percentage of women in economics...

Economics doesn't fare a whole lot better than finance, ranking 125th out of 175, with 11.2% of top scientists being women (compared with 10.3% for finance). Adams and Xu then look at this using a statistical model, finding that after controlling for career span:

The coefficient on finance is -0.016, which suggests a scientist in finance is 1.6% less likely to be a woman than a scientist in Economics4... we compare women’s representation in finance and STEM9 fields and find a scientist in finance is 1.8% less likely to be a woman than a scientist in STEM9 fields... we compare women’s representation in finance to their representation in all other fields. The results shows that a scientist in finance is 7.5% less likely to be female than a scientist in other fields.

So, top scientists are much less likely to be female in finance, than in economics (incidentally, 'economics4' is made up of the four economics fields: agricultural economics and policy, economics, econometrics, and economic theory), in STEM, and across all fields. However, it is their next results that are most surprising, and set finance apart. When looking at academic rank, the number of papers, citations, and other metrics of research quality, Adams and Xu find that:

On aggregate, women’s ranks are 7.7% lower than the ranks of their male peers in the same field. The gap increases to 9.3% in STEM9 fields. When we examine the gender gap in ranks in Economic4, we find the gap is much smaller (2.1%) and statistically insignificant. We also find that women in Economics4, STEM9 fields and all fields have fewer cited papers and lower total citations...

Female finance academics have 13.9% fewer cited papers but on average each of their papers has 18.6% more citations than papers by male finance academics. 

So, while in most fields female academics have fewer papers and fewer citations, in finance they have fewer papers but more citations. This suggests that female finance academics are more influential within their discipline than female academics in other disciplines, and despite this, the overall representation of women in the top two percent of finance academics is lower. What explains this anomaly? 

Adams and Xu don't really answer that question, but they do implicate 'ability belief' as a driver of the low representation of women in finance. They define 'ability belief' as:

...individuals’ beliefs about the importance of innate talent in their fields.

It is measured based on responses (taken from this study) to:

...the following questions:(1) Being a top scholar of [discipline] requires a special aptitude that just can’t be taught; (2) If you want to succeed in [discipline], hard work alone just won’t cut it; you need to have an innate gift or talent; (3) With the right amount of effort and dedication, anyone can become a top scholar in [discipline]; (4) When it comes to [discipline], the most important factors for success are motivation and sustained effort; raw ability is secondary. Fields in which respondents placed more weight on (1) and (2) are considered fields with higher field specific ability beliefs.

Using this measure as a predictor of women's representative in a field, Adams and Xu find that:

The coefficient on Ability belief is -0.169 and significant at the 5% level. A one standard deviation increase in Ability belief is associated with a 0.585 standard deviation decrease in women’s representation.

When women's and men's beliefs are included separately in the model, only men's beliefs are statistically significant. Adams and Xu conclude that:

These results suggest that women face greater barriers to entry into finance than men do.

Yeah, I don't understand that conclusion either. Adams and Xu don't do a good job of explaining why men's beliefs about innate ability reduce women's representation at the top of the finance discipline. I'll need that explained to me in sentences composed of very short words, I think. Nevertheless, the descriptive results are pretty damning for finance (and they don't look particularly good for economics either, to be fair). Clearly, there is more to do. Unfortunately, this paper is short of solutions.

[HT: The Conversation]

Working while studying is mostly bad for students

I've written a number of posts on the impact on students of working while studying (most recently here). The literature is much broader than I have the capacity to review on the blog. Fortunately, I don't have to do the donkey work there. This 2019 article by Brecht Neyt and Eddy Omey (both Ghent University), Dieter Verhaest (KU Leuven), and Stijn Baert (Ghent University), published in the Journal of Economic Surveys (ungated earlier version here) reviewed the literature up to that point.

In fact, Neyt et al. do such a thorough job, they look at the literature on the effects of working for both secondary school students and tertiary students. They first outline the theory:

On the one hand, according to Human Capital Theory (Becker, 1964), student employment can be a complement to education due to the additional skills and knowledge obtained while working. There are several reasons why student work may lead to such an increase in human capital. First, student employment enables the acquisition of new general and transferable skills such as work values, communication skills and a sense of time management... Second, combining study and work may offer students the opportunity to apply in practice what they have learned in school... Third, student employment may change students’ intertemporal preferences and increase their future-orientedness, thereby motivating them to work harder in school in order to achieve a certain career goal...

On the other hand, building on the Theory of the Allocation of Time (Becker, 1965), Zero-Sum Theory suggests that student employment and education are substitutes. More formally, this theory states that students have fixed time resources and that student employment strongly constrains students’ use of these resources. As a consequence, time resources used to work cannot be used for activities that enhance academic performance (e.g. studying, doing homework and attending classes... As the reduced time spent on these activities subsequently worsens academic performance... student employment may have a detrimental effect on educational outcomes.

It appears that the great Gary Becker has a bob each way on whether working affects student outcomes. Anyway, those are essentially the theoretical positions that I already outlined in this recent post. Neyt et al. also outline an additional theory from sociology:

Another theory that supports a negative association between student work and educational success is Primary Orientation Theory... often cited in the field of sociology. This theory suggests that the worse academic performance of working students compared to non-working students is related to their primary orientation being toward work rather than toward school. In other words, it reflects a disengagement from school that existed before the decision to work was made, rather than a negative effect due to student employment itself.

Having outlined the theory, Neyt et al. then discuss the problems with trying to find an effect, free of selection bias and endogeneity. As they note, much of the literature is based on correlations, rather than causal estimates. In fact, of the 50 studies they include in their review, they only consider nine to be sufficiently convincing. The results overall suggest negative impacts of working:

Of these nine studies we perceive as most convincing, three (i.e. 33.33%) find evidence of a negative effect of student employment on educational outcomes, although one study reports the effect is rather small. Additionally, four (i.e. 44.44%) report both negative and neutral effects, depending on the type of educational outcome (infra, Subsection 4.3.3), type of student job... or type of student... considered. Finally, two studies report neutral effects.

Note that none of those studies are reporting overall positive effects of working on students. Overall, it appears that Zero Sum Theory might be the best characterisation of the effect of student work on student outcomes. However, as Neyt et al. note, there are differences, particularly:

This pattern of a more negative effect for students in tertiary education is also found when only considering the most convincing studies... Indeed, the two more ambitious studies that find neutral effects of student work examined student work in secondary education. In contrast, the convincing studies in tertiary education always find at least some negative relationship.

That finding might be a bit surprising, given that tertiary education students have a bit more flexibility in their schedules. On that point, Neyt et al. write that:

We believe, however, that this finding is sensible for two reasons. First, due to the more challenging nature of studies at college or university (compared to those in secondary school), combining study and work during tertiary education may be less feasible. Second, combining study and work in tertiary education may change students’ attitudes toward school and intertemporal preferences. This may cause their present discounted value of continuing school to decrease and, as a consequence, the probability to quit tertiary education earlier than they anticipated before experiencing the student work to increase... This reasoning is less valid for students in secondary education, as they are to a lesser extent confronted with choosing between continuing school and joining the labour market.

There are some complex interactions between student engagement, student performance, and preferences for work or study at play. Tertiary students are less constrained than secondary students in terms of how they enact their preferences for attendance at the least, which may be why the impacts are more negative for tertiary students. This may also relate to the Primary Orientation Theory that Neyt et al. refer to earlier in their article.

So, the takeaway from this review is that student work is mostly negative. However, not always negative. As I noted in this post, working in the same subject area in which the student is studying may have positive medium-term benefits. These results are important, and students should take note. Policymakers should also take note. We should be supporting students better, so that they don't have to compromise their studying and future economic outcomes by working.

Read more:

• Students should work less, but not stop working entirely
• Working and academic performance at university
• Working while studying may not always be bad for students

Lockdowns, alcohol, and why all natural experiments are not created equal

A natural experiment occurs when there is some unexpected policy or other shock that affects some, but usually not all, of the units of observation (often people, but could be firms, states, or countries). Because by definition the shock is not anticipated (that's why it's called a 'shock'!), it is often as good as randomly assigned, and can be used to extract the causal impact of one variable on another. So, for example, if a school classroom is destroyed in a fire, and the students are distributed to other classes, that might be able tell us something about the effect of class sizes on student outcomes.

However, perhaps more than other experiments, a natural experiment relies crucially on how you define the counterfactual - what would have happened in the absence of the shock. Sometimes, you can assume that there are other similar people (or firms, states, or countries) that were not affected by the shock. In the case of the torched classroom, students in another nearby (and sufficiently similar) school might provide a good counterfactual for how students would have performed if their classroom hadn't burned down. However, that doesn't work so well if everyone was affected by the shock. In that case, you might assume that whatever was happening before the shock would continue afterwards. That's sometimes a heroic assumption that simply fails on the face of it.

The coronavirus pandemic is a good example of this. While it might be attractive to think of the pandemic (or lockdowns) as a natural experiment, it is very difficult to tell what would have happened if there hadn't been a pandemic. Comparing outcomes before and after the pandemic might be ok, but comparing the period before the pandemic with the period during the pandemic is fraught, because the fundamental relationships may be different.

However, that hasn't stopped Christopher Snowdon, in this discussion paper on the effects of lockdowns on alcohol-related mortality in the UK. Snowdon uses the lockdown to take aim at the 'single distribution theory' and its implications, which he outlines as:

Starting in the 1960s, a view began to take hold in the alcohol research community that the amount of heavy drinking and alcohol-related harm... in a society were directly linked to per capita alcohol consumption... It was argued that this was not because heavy drinkers consumed a large quantity of alcohol (thereby raising the mean), but because the entire distribution of consumption was determined by the mean...

This is known as the single distribution theory and it was given support by the influential theories of the British epidemiologist Geoffrey Rose in the early 1990s...

Adhering to the belief that reducing per capita alcohol consumption is necessary and sufficient to reduce alcohol-related harm, proponents of the whole population approach endorse population-wide, supply-side interventions in the market - targeting ‘the Three A’s’: affordability, availability and advertising. Policies include alcohol duty rises, minimum pricing, tougher licensing laws and advertising bans.

Single distribution theory is mathematically dubious. We don't need research to tell us this. A simple example will suffice. If greater availability of alcohol increases alcohol consumption among heavy drinkers, regardless of any effect on the mean alcohol consumption of the population, you would expect greater harm (among those heavy drinkers). So, Snowdon has created a bit of a straw man argument in order to support a call for reductions in alcohol control (ok, maybe some public health researchers use single distribution theory to support restrictions on alcohol availability, but I had never heard of it before reading this discussion paper, so it can't be that widely employed).

Anyway, back to the evidence that Snowdon provides. He first assets that alcohol availability was curtailed during the pandemic:

The number of places in which alcohol could be bought collapsed on 20 March when pubs, clubs, bars and restaurants were closed by law... Hotels were closed for leisure and tourism, although some remained open for limited business purposes. Nightclubs were closed from 20 March 2020 until 19 July 2021 in England and until January 2022 in Scotland and Wales. At a stroke, most premises with a commercial alcohol licence were shut...

The number of places selling alcohol therefore fell by approximately two-thirds. Its availability, as defined in modern public health, was greatly reduced.

That seems sensible at first glance (although see my comments below). Snowdon then presents some evidence on the how alcohol consumption changed, finishing with

In summary, per capita alcohol consumption fell in the UK but most people did not change the amount they drank and those who did went in opposite directions, with heavy drinkers tending to drink more. There was no single distribution.

Finally, Snowdon looks at the effect on alcohol-specific mortality (noting that "‘alcohol-specific’ deaths are wholly due to alcohol use, as distinct from the broader category of ‘alcohol-related’ deaths which include diseases in which alcohol is one risk factor, such as cancer"). He shows that:

...there is clear evidence of a dramatic increase in alcohol-specific mortality in the UK with the death rate rising by 18.7 per cent in 2020...

So, to summarise, according to Snowdon: the availability of alcohol declined dramatically; alcohol consumption didn't change on average (but there was a change in distribution, with more drinking among heavy drinkers, and less drinking among occasional drinkers); and alcohol-specific mortality increased. Snowdon then concludes that:

A more reasonable conclusion to draw, which was amply illustrated by the extraordinary natural experiment of 2020, is that harmful drinking is not driven primarily, if at all, by ‘commercial determinants’ but by personal circumstances, hardship and stress. From this we might conclude that tackling harmful drinking requires focusing on harmful drinkers rather than on the whole population.

I disagree. This natural experiment offers minimal guidance on how alcohol availability affects consumption or harm. First, in spite of the facts that bars and hospitality venues were forced to close, it's not clear that alcohol availability even reduced. People could still buy alcohol, as Snowdon himself notes:

During the lockdowns, off-licences - including home delivery companies and most supermarkets - were almost the only places from which alcohol could be bought.

So, people could still access alcohol. They probably bought more alcohol using home delivery, and less from bars. Snowdon doesn't provide any evidence on this. Alcohol consumption on average didn't change, so clearly people were getting it from somewhere. The relationship between alcohol outlets and alcohol availability fundamentally changed during the lockdowns. Comparing the time before lockdown with the time during lockdown isn't going to tell us anything meaningful about how availability and consumption are related. The counterfactual is not appropriate.

All we can really conclude from this paper is that heavy drinkers drank more, and alcohol-specific mortality increased. However, even that (seemingly obvious) relationship is tainted by the lack of a valid counterfactual. Medical services were overwhelmed during the pandemic, and so mortality might be higher from all causes included alcohol-specific mortality, but not because of increased drinking among heavy drinkers.

In both cases, it would be better to compare the time before lockdowns with the time after lockdowns, ignoring any data during the lockdown period. It is likely that some alcohol outlets closed down during lockdown and did not reopen. The lockdown therefore provides an exogenous change in the number of outlets, and that comparison may be a valid natural experiment, with a sensible counterfactual. Perhaps we'll see some research along those lines in the future. Instead, what Snowdon is done is largely unhelpful.

This is the second paper of Snowdon's that I've taken issue with this year (see here), although admittedly the earlier one was written in 2018. However, I had been planning to read his 2010 book The Spirit Level Delusion, which offers a critique of Wilkinson and Pickett's famous book The Spirit Level (which I discussed here, with some additional critique of my own here). If the standard of evidence in Snowdon's book is the same as in these two papers, I don't think I will bother.

[HT: Eric Crampton at Offsetting Behaviour]

Summarising topics for students using movie trailers

Following on from yesterday's post about communicating research results to the general public, I read this 2018 article I found by Lara Goudsouzian (DeSales University), published in the Journal of Microbiology and Biology Education (open access). Goudsouzian discusses the use of movie trailers as a way to quickly introduce new topics for students (in biology, presumably). The paper doesn't include a lot of detail, but apparently the Apple iMovie app is able to easily create movie trailers based on photographs or video clips using one of several templates (and the process is described here). Goudsouzian created five trailers over a semester long course, and then her students:

...were asked to self-report their attitudes toward the movie trailer introductions using a five-point Likert scale...

The results are summarised in Figure 2 from the article:

It's pretty clear that students found the trailers to be interesting, unique and unexpected, and that they increased students' interest in the topics. However:

While some students did report using the movie trailers to study for exams, most disagreed with the assertion that the movie trailers could be used as a study tool for summative assessments...

That much is obvious from the bottom bar in the figure. Goudsouzian tries to argue that an increase in student interest would lead to higher student engagement, which in turn leads to greater learning, but I'd like to see some actual evidence of that before I believe it. However, Goudsouzian does offer this in the conclusion:

Rather than creating movie trailer introductions for students, instructors might choose to assign this work to students, as a way of familiarizing them with the topic before entering the class or as a means to review the material after it has already been covered.

That's very similar to this example I linked to in yesterday's post. And it makes me wish that I had read this article a couple of months ago, when planning how to approach flexible assessment in my ECONS101 and ECONS102 classes. We have adopted weekly video reflections for the trimester starting shortly (as one of two options for students, the other being in-class tests). However, an alternative might have been getting students to create trailers for each topic (having said that, the brief for a video reflection is open-ended enough that a trailer could satisfy the requirements, and would be pretty novel and interesting for the tutors to grade). We definitely need assessment items that make it more difficult for students to cheat (see here for more on that point). 

Reducing the opportunities for cheating is not the only reason to consider alternative types of assessment. Finding novel ways for teachers to keep students engaged is also important, and one way to keep students engaged is to have engaging assessments. In ECON100, we used to have a group video assignment, until it got to the stage where the novelty had clearly worn off and the students weren't taking it seriously (I linked to the last winning videos here). At that point, we ditched the video assignment and we haven't brought it back since. The takeaway is that every so often we need to refresh our assessment styles (if not our teaching styles). That not only keeps the students engaged, but gives us as teachers a chance to try something a little different. And even if only for that purpose, movie trailers may be a good option.