It's not time to give up artificial sweeteners just yet

Yesterday in my first ECONS101 class of the new semester, we talked about the difference between causation and correlation. So, it was interesting to see a story in the New Zealand Herald illustrating exactly that distinction this morning:

A New Zealand health expert says an American study linking diet drinks with stroke is a warning that artificial sweeteners are not risk-free.

Published in the journal Stroke, the longterm study of post-menopausal women found that those who consumed higher amounts of diet drink had a 23 per cent greater risk of having a stroke than those who drank small amounts or none.

The higher users also had a 29 per cent higher risk of coronary heart disease and a 16 per cent higher risk of death.

Higher use was defined as consuming the equivalent of two or more cans of "diet drinks such as Diet Coke or diet fruit drinks" a week. Low use was less than one a month.

The study involved more than 81,000 women in the Women's Health Initiative Observational Study.

The original study is here (ungated), by Yasmin Mossavar-Rahmani (Albert Einstein College of Medicine) and eight other co-authors. There is also an interesting related editorial by the editors of the journal Stroke here (also ungated). The authors of the study used data from over 81,000 post-menopausal women in the U.S. Three years into the study, the women were asked how often they drank artificially-sweetened beverages (ASBs), like Diet Coke or diet juices. The study then followed those women over time to see what happened, and found that the women who said they drank more ASBs were more likely to suffer coronary heart disease, or to die.

The problem here is that this study doesn't show a causal impact of ASBs on heart disease, stroke or mortality. Although it was carefully designed, and the authors controlled for a bunch of factors also known to affect heart disease or mortality risk, like quality of diet, body mass index, etc., the study still only shows a correlation. As the editorial and the New Zealand Herald article note, this correlation might be causal (and is consistent with a causal story), but it might not. It might be a result of reverse causation. Perhaps instead of ASBs causing heart disease or stroke, being at higher risk of heart disease or stroke causes women to be more likely to drink ASBs; maybe in order to lose weight or to reduce their risk of poor health (maybe under doctor's orders). This is supported by the fact that when they look at sub-groups by BMI, the correlation is only statistically significant for obese women, but not for overweight or normal-weighted women.

Also, the sample was limited to post-menopausal women. I'm unsure how representative of the general population that sample is, even if it is large. Besides which, people's dietary behaviours change over time. This study followed women for an average of 11.9 years, and then based the risk models on one self-reported question about ASB use over the three months at the start of that period. Perhaps women who were more likely to drink ASBs were also more likely to drink sugar-sweetened beverages as they got older? Perhaps they were more likely to be yo-yo dieters, which is bad for you.

So, it is far too soon to conclude that ASBs pose an unnecessary risk to health. At the very least, we would need more careful studies that follow a more widely representative sample of people over time. And even better, if they were randomised so that some drank ASBs and others didn't (although it would be pretty difficult to get ethical clearance for that!).

Could Twitter increase student engagement?

Engaging students is probably the most important aspect of teaching. Engaged students are more likely to learn, and retain their learning for longer. I expend quite a lot of effort trying to engage with students, including through social media. My classes have had their own Facebook groups for the last several years. I know of other lecturers who have used Twitter, Instagram, WhatsApp, or WeChat, but I haven't engaged with those (either because I'm not convinced that enough students are using them to make it worthwhile or, in the case of Instagram which many students use, it isn't clear how it would be an improvement over Facebook). Also, while I know that students find the Facebook groups helpful for their learning, I haven't rigorously evaluated the effect.

So, I was interested to read this 2017 article (which appears to be open access) by Abdullah Al-Bahrani (Northern Kentucky University), Darshak Patel (University of Kentucky), and Brandon Sheridan (Elon University), published in the Journal of Economic Education. In the article, Al-Bahrani et al. evaluate the impact of Twitter on student learning in economics principles courses in three U.S. universities. All students were provided, roughly three times per week, with "short articles relating concepts from class to the real world" (probably not too dissimilar to what I do on this blog). At each institution, one section of the class received this information via the learning management system (like Moodle), while one section received this information via Twitter. Al-Bahrani et al. then looked at whether there was any difference in learning between the two groups, and found that:

In all specifications, the treatment coefficient is statistically insignificant, therefore we find no evidence that communicating through Twitter impacts students’ learning differently than a traditional LMS.

So, Twitter is no better for providing students with links to additional material than a learning management system like Moodle. Admittedly, there was a lack of statistical power in their analysis as they only had a sample of 163 students across the three institutions. So, maybe there was an effect, but their sample size was too small to detect it. However, either way I do agree with their conclusion that:

...the impact of Twitter on the educational experience may not necessarily be in the form of grades or learning, but may rather be in the form of engagement, teacher evaluations, and fostering interest in the topic.

Student engagement is important, and finding ways of engaging students on 'their turf' is even more important. More on that in a future post.

Why summer ice cream prices don't respond to changes in demand

New Zealand has been suffering through a heatwave. One of the effects was a shortage of ice cream, as the New Zealand Herald reported last month:

Hot temperatures have led to such a demand in ice cream and cold drinks that some businesses have had to turn customers away.

Havelock North McDonald's ran out of soft serve ice cream and milkshakes on Wednesday evening, forcing customers to look elsewhere.

A shortage occurs when the quantity of a good demanded exceeds the quantity of the good supplied. Some customers will miss out on the good. We might expect the price to increase to eliminate the shortage. Consider the perfectly competitive market in the diagram below:

Before the heatwave, the demand for ice creams is D0 and the supply is S. The market is in equilibrium with the price P0 and the quantity of ice creams traded is Q0. When demand increases from D0 to D1, the equilibrium should move to the intersection of D1 and S, where the price has increased to P1 and the quantity of ice creams traded has increased to Q1. However, if the price stayed at P0, the quantity supplied remains Q0, but the quantity demanded is QD - there is a shortage (or excess demand).

However, the market diagram above assumes a perfectly competitive market. In a perfectly competitive market, buyers and sellers are price takers - they have no control over the price, which is set by the market (at the intersection of supply and demand). The perfectly competitive market assumes there are many buyers and many sellers, and the sellers are all selling an identical (homogeneous) product. This is not a reasonable assumption for most markets, including the market for ice creams. In most markets, there are a many buyers, but few sellers, or the sellers are selling a differentiated product. That gives the sellers some market power - the power to choose their own price.

The diagram below shows what happens when a firm with market power faces an increase in demand. The firm is profit maximising, so it operates at the profit-maximising price and quantity where marginal revenue intersects with marginal cost - with the original (red) demand curve D0 and (red) marginal revenue curve MR0, this leads to the price P0 and the quantity of ice creams traded is Q0. When demand increases to D1 (and marginal revenue increases to MR1), the profit-maximising price increases to P1, and the quantity increases to Q1. However, if the firm kept the price at the original price P0, then the quantity demanded is QD. There is no excess demand in this case, unless the firm hadn't planned for the possibility of extra sales.

So, regardless of whether we are considering a firm with market power or a firm in a perfectly competitive market, when the demand for ice creams increases, we should expect the price to increase. So, it might be surprising that the price doesn't adjust. Why wouldn't the price adjust?

There are a few reasons that sellers don't automatically adjust prices in response to changes in demand. The first reason is menu costs - it might be costly to change prices (they're called menu costs because if a restaurant wants to change its prices, it needs to print all new menus, and that is costly). The second reason is that changing prices creates uncertainty for consumers, and if they are uncertain what the price will be on a given day, perhaps they choose not to purchase (in other words, the cost of price discovery for consumers makes it not worth their while to find out the price). The third reason is fairness. Research by Nobel Prize winner Daniel Kahneman (and described in his book Thinking, Fast and Slow) shows that consumers are willing to pay higher prices when sellers face higher costs (consumers are willing to share the burden), but consumers are unwilling to pay higher prices when they result from higher demand - they see those price increases as unfair.

Finally, in this particular case, the price of McDonald's ice creams are set at the national level. So, the seller doesn't have control over the price and can't adjust it in response to changes in demand. So, even though the simple economic models might suggest a particular outcome (an increase in price), it is easy to explain why the real world outcome differs from the model.

The economics of clearing landmines

Some years ago, I was involved in a project measuring the value of a statistical life (VSL) in Thailand (ungated earlier version here) and Cambodia (ungated version here). Part of the point of that work was to overcome earlier cost-benefit analyses of landmine clearing activities, which had wildly underestimated the benefits of clearing landmines, and often suggested that the costs outweighed the benefits.

The problem with the earlier studies is that they used the human capital approach to valuing the benefits of lives saved from clearing landmines. The human capital approach estimates VSL based on the total value of output that an average person would produce over their lifetime - essentially, it is estimated based on the total wages they would earn. However, time in work is only part of what we contribute to society, and the human capital approach therefore must underestimate the real VSL. An alternative is to use a non-market valuation approach like contingent valuation. Essentially, this involves asking people what they would be willing to pay for a small reduction in the risk of dying. Then, the average that people are willing to pay can be scaled up to work out what they would be willing to pay (on average) for a 100 percent reduction in the risk of death, which is the estimated VSL.

Our work in Thailand and Cambodia showed that the estimated benefits of landmine clearance were much larger than previously estimated. However, in spite of the higher benefits, the cost-benefit calculus only favoured landmine clearance in some areas. There were many places (typically remote, far from roads, where few people lived) where the costs of clearing landmines still outweighed the benefits.

However, benefits from lives saved (and injuries averted) are not the only benefits from clearing landmines. In a recent NBER working paper, Giorgio Chiovelli (London Business School), Stelios Michalopolous (Brown University), and Elias Papaioannou (London Business School) look at the effects of clearing landmines in Mozambique. Specifically, they estimate the impact on economic activity. Mozambique is interesting to investigate, because it is the first country ever to move from being classified as "heavily contaminated by landmines" (in 1992) to "landmine free" (which it was certified as in September 2015).

However, good data on economic activity are scarce in Mozambique due to the years of conflict. So, Chiovelli et al. make use of night-time lights data from satellite images. This is a fairly new and exciting data source, which relies on the observation that areas that are more illuminated have higher economic activity (this has been shown in many studies, but for a graphic example, look at photos that compare neighbouring North Korea and South Korea, such as this one).

Chiovelli et al. exhaustively compiled data on the landmine clearance activities in Mozambique over the period from 1992 to 2015, so that they could evaluate the impacts of clearances in different parts of the country occurring at different times. They then examined the economy-wide impacts, recognising that the main impact of landmines was on reducing market access through making roads and rail impassable. They found that:

...a one-standard deviation increase in the number of cleared CHA [Confirmed Hazardous Areas] increases log luminosity by 0.072 standard deviations... Clearing a locality from all contaminated hazardous areas increases the likelihood of the locality being lit by roughly 4%; this estimate should be compared with an average value of the locality being lit of 9.7% in 1992.

The effect is reasonably large, as:

...cleared localities (as opposed to not-contaminated ones) enjoy a boost in economic activity comparable to that of being one of the few localities endowed with a colonial railroad...

They then test for heterogeneous impacts of landmine clearance, and find that:

...reducing the number of contaminated areas along roads-railroads and clearing areas around villages and towns, especially the ones with cantinas is associated with significant increases in luminosity. On the other hand, removal of landmines in remote, rural areas (the residual category) does not seem to lead to increases in luminosity.

This is interesting, because it complements the findings from my earlier studies. The areas that are remote, with few people, not only would have lower benefits of landmine clearance due to fewer lives saved (and injuries averted), but also have lower benefits in terms of increases in economic activity.

Chiovelli et al. then move on to look at spill-over effects, and find that there are increases in economic activity even for areas with no landmines. This arises because those areas also benefit from landmine clearance, when roads and rail are cleared and areas become more accessible.

Overall, the takeaway message is that there are significant economic benefits from clearing landmines. However, that still doesn't necessarily mean that the benefits outweigh the costs in all areas.

[HT: Marginal Revolution, last June]