Taking advantage of loss aversion in education

Many years ago (I forget exactly when), I introduced extra credit into my ECON110 class (which is what is now ECONS102). The idea was to provide an incentive for students to attend class, since they could earn extra credit for completing various in-class exercises. A couple of years later, I briefly changed the way that I framed the extra credit, from being "extra marks that would be gained from attending", to "extra marks that would be lost by not attending".

If students were purely rational, the change from 'gain framing' to 'loss framing' the extra credit should have had no impact on student attendance. However, I was looking to exploit the fact that most people are quasi-rational, rather than purely rational. Quasi-rational decision-makers are loss averse, meaning that they value losses more than equivalent gains. For a loss averse person, losing $20 makes them unhappy to a greater extent than winning $20 makes them happy.

Does a change from 'gain framing' to 'loss framing' work? That is the question that this new article by Antal Ertl, Éva Holb (both Eötvös Lóránd Science University), and Barna Bakó (Corvinus University of Budapest), published in the Journal of Economic Behavior and Organization (open access), tries to answer. They use data from a field experiment at Corvinus University of Budapest, where students enrolled in a compulsory macroeconomics course for business students were randomised into one of three conditions: (1) Gain group, which earned points in each of four tests and the final examination as usual; (2) Loss group, which started each test and the final exam with full points, but had points deducted for each incorrect answer; and (3) Hybrid group, which was the same as the Gain group for the tests, but switched to the loss framing for the final examination.

Ertl et al. have a sample of 321 students who consented to be part of the research, completed an initial questionnaire at the start of the term, and earned a non-zero grade. Randomisation was conducted at the level of the tutorial group (so all students in a tutorial were in the same treatment), in such a way that each teacher had groups across more than one treatment. One wrinkle in their analysis is that the best three out of the four tests would count towards a student's grade, meaning that students may end up putting differential effort into each test, depending on how they have performed in the other tests already completed. So, in addition to looking at the effect of treatment on each test mark individually, Ertl et al. look at the effect on the 'best three' tests collectively, as well as the exam mark.

If randomisation were perfect and the treatment groups were balanced, the comparison between the Loss group and the Gain group would demonstrate the overall effect of loss framing on student performance. The comparison between the Loss group and the Hybrid group for the final exam, compared with the same comparison for the best three tests, would demonstrate whether students adjust in such a way that the loss framing has less impact over time (because the Hybrid group would be in their first loss-framed assessment, while the Loss group would be in their fifth such assessment). The treatment groups weren't perfectly balanced, with students sorting into tutorial groups in part based on whether they worked part-time. So, Ertl et al. control for working part-time, the tutorial day and time, and the tutorial group teacher, as well as other demographic and background variables.

In their main analysis, they find support for the positive effects of loss framing:

For the Loss treatment, the effect on the average of the Best 3 Tests is 3.2 percentage points, although the difference is not statistically significant. The treatment effect on the Final Test score, however, shows a large difference of 9.6 percentage points when not controlling for Best 3 Tests’ scores, i.e., how well students did throughout the semester before the Final Test.

After controlling for performance in the best three tests, the effect of the loss framing on performance in the final examination is a statistically significant 7.8 percentage points. Turning to the comparison of the Loss and Hybrid groups, Ertl et al. find that:

...the estimated effect sizes for Loss and Hybrid are essentially the same for the Final Test, once we take into account how well students did perform throughout the semester...

These results are consistent with loss framing leading to better student performance, and there being no novelty effect - the effect of loss framing doesn't appear to decline over time. Ertl et al. go on to show that the effects are similar for both male and female students, but larger for students who did not take advanced mathematics in high school than for those that did. They also show that the treatment did not seem to negatively affect students' perceptions of the course, because the teaching evaluations were similar for the different treatment groups.

Finally, Ertl et al. do provide a note of caution in their conclusion:

previous studies have highlighted possible psychological and motivational costs associated with loss framing... These findings suggest that the mechanism by which loss framing improves performance may, at least in part, operate through heightened tension and concern about avoiding mistakes rather than through enhanced intrinsic motivation. Moreover, in extreme cases, loss-framed grading may even produce adverse effects — for example, low-performing students might become discouraged early in the semester after ‘‘losing’’ too many points. Once it becomes apparent that only a passing grade is attainable at best, the loss-framed structure may make this limitation increasingly salient, potentially exacerbating anxiety and disengagement. Over time, this could have broader implications for students’ well-being and their willingness to enroll in courses or programs that employ such systems.

Ertl et al. don't directly test for these effects, but they should be a concern. We may be able to improve student performance through loss-framing assessments, but that might come at a cost to student mental health and wellbeing.

And that brings me back to the example I started with, from my ECON110 class. When I switched extra credit from gain-framed to loss-framed, student attendance in class did improve slightly. However, the bigger impact seemed to be the number of students who would contact me by email, seeking special consideration for missing the extra credit, offering to provide medical certificates or other evidence to explain their absence, and asking for extra chances to complete the in-class exercises. It turned out to be administratively much more costly for me, and so the change was short-lived (to the extent that I cannot even remember which year I tried this in). Those reactions could suggest a negative psychological effect of the switch from gain framing to loss framing.

So, not all interventions that are effective for promoting student performance should be adopted. We need to carefully consider both the benefits and the costs of the intervention first. Taking advantage of student loss aversion might be worth exploring further, but I would want to see a wider evaluation that included student wellbeing outcomes before adopting it.

This week in research #139

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

• List (open access) comments on how to address the generalisability of research
• Lehner et al. (open access) find that the opening of a Walmart Supercenter is associated with a 2.2 percentage point (18%) increase in poverty, and that the increase is largest for younger and less-educated adults

And the latest paper from my own research (led by my former PhD student Muhammad Irfan, along with Ushan Goonawardane, and Craig Robertson), which was also covered in the New Zealand Herald:

• Our new article (open access if you register for free) in the New Zealand Medical Journal performs a comparison of methamphetamine contamination of 423 properties across New Zealand, before and after the requirement to test for methamphetamine was eased in May 2018, and finds a significant increase in methamphetamine contamination

Customers shouldn't pay less when they use a self-checkout, they should pay more

The New Zealand Herald reported last week:

State representative Nikki Lucas has introduced a bill that would require retail businesses selling food in the state to offer a 10% discount to those who used the self-checkout lane.

“Retail businesses increasingly rely on self-checkout systems to reduce staffing and operational costs by shifting responsibilities traditionally performed by employees onto consumers,” she wrote...

Consumer NZ head of advocacy Gemma Rasmussen said her organisation thought there was validity to the argument in New Zealand, too.

Call me radical, but I think that Lucas and Rasmussen have this backwards. Customers shouldn't pay less when they use a self-checkout, they should pay more. To see why, I'm going to rely on the concept of price discrimination - where the seller sells the same good or service to different groups of consumers for different prices.

Consider two groups of consumers (impatient, and patient), and two options (self-checkout, and regular checkout). The first group of consumers is impatient, and they want to get out of the store as soon as possible, and for that reason they prefer to use self-checkout. This group can be said to have a short time horizon for their purchases. This short time horizon makes their demand for goods less elastic (less sensitive to price). The second group of consumers is more patient, and they are willing to wait. This group can be said to have a longer time horizon for their purchases, which makes their demand for goods more elastic (more sensitive to price).

If supermarkets want to price differently for each group, which group should pay the higher price? The answer to that question is shown in the two diagrams below. Both diagrams show a firm with market power (a supermarket), and each diagram corresponds to one of the sub-markets. The sub-market on the left represents the patient buyers, who have more elastic demand - notice that the demand curve D1 is relatively flat (which means that a change in price will have a big effect on the quantity that these consumers demand). The sub-market on the right represents the impatient buyers, who have less elastic demand - notice that the demand curve D2 is relatively steep (which means that the same change in price would have a smaller effect on the quantity that these consumers demand, than it would for the patient consumers). The marginal cost (MC) is the same in both sub-markets - it doesn't cost the supermarket any more to sell a product to an impatient buyer than what it costs them to sell that same product to a patient buyer. [*]

The supermarket will maximise profits by selling the quantity where marginal revenue (MR) is equal to marginal cost (MC) - this is the standard short-run profit-maximising condition (as I discussed in this post). In the impatient sub-market, the profit-maximising quantity occurs where MR2=MC, which is Q2. In order to sell that quantity in the impatient sub-market, the supermarket should set the price equal to P2. The problem with that high price P2 is that in the patient sub-market, no consumers would be willing to buy the good at all. The supermarket can increase profits if it charges a different price in the patient sub-market from the price it charges in the impatient sub-market. In the patient sub-market, the profit-maximising quantity occurs where MR1=MC, which is Q1. To sell that quantity in the patient sub-market, the supermarket should set the price equal to P1. In other words, the supermarket should charge a higher price to the impatient consumers, and a lower price to the patient consumers.

The problem here is that supermarkets don't know (for sure) which group (impatient or patient) any particular consumer belongs to. But by offering different checkout options, the customers can sort themselves into the impatient (less elastic demand) group and the patient (more elastic demand) group, because the impatient consumers use the self-checkout. In other words, the supermarket should charge a higher price to the users of the self-checkout.

This is an example of menu pricing (or second-degree price discrimination) - where the consumers are presented with a menu of options, and they select the one they prefer.  Crucially, the seller knows that some menu options appeal to consumers with more elastic demand, and other options appeal to consumers with less elastic demand. In this case, there are two menu options - self-checkout, or regular checkout, and the supermarket knows that the self-checkout appeals to the impatient consumers who should be charged a higher price.

So, customers who use a self-checkout right now shouldn't be arguing to lower prices. They should think themselves lucky that supermarkets aren't optimising, because if they were, the prices at self-checkouts would be higher than at regular checkouts.

*****

[*] You could argue that it doesn't cost the same to offer purchase through regular checkouts and self-checkouts. However, how big is the cost difference, really? Let's say that it takes two minutes to scan your items, but would take three minutes through the regular checkout, because the payment process tends to take a bit longer at a regular checkout. With self-checkout, the supermarket would save three minutes of labour. Say that the supermarket pays their checkout staff $30 per hour (somewhat more than the minimum wage). By using the self-checkout, you've saved the supermarket $1.50 of labour in this example (3/60 * $30). Except, that calculation doesn't take into account that the self-checkout is not a zero-labour option. There is usually a checkout person who has to watch over the consumers using the self-checkout. So, the saving is actually a bit less than that. It almost certainly isn't close to the 10 percent discount that Lucas is arguing for. Most of the cost of the items that you buy at the supermarket is the wholesale cost that the supermarkets pay, not the checkout labour cost.

Generative AI, cognitive offloading, and escaping the 'illusion of competence'

Last week, for maybe the first time, I found myself telling a student not to use AI. That might sound extraordinary. After all, generative AI hit the big time with the release of ChatGPT in November 2022, and increasing numbers of students have been using it ever since. Many lecturers immediately freaked out, and many institutions initially reacted by banning or restricting AI use, before realising that they were fighting a losing battle against the incoming tide of generative AI and trying to impose 'guardrails'.

I've never asked my students not to use generative AI. In fact, I've encouraged it. I even have custom AI tutors set up for each of the papers I teach, that use a knowledge base of materials from the paper to give students targeted assistance. I have little to fear from generative AI, because the vast majority of assessment in my papers is in-person and invigilated (that's one of the beautiful things about teaching first-year papers - I can argue that basic concepts and applications can be authentically assessed in an exam environment).

Anyway, back to the story. The student was using our class AI tutor during class, to give them a solution to a problem we were working on during class. I pointed out that it defeated the purpose of doing the problem in class, if they used Jane (our ECONS102 AI tutor is named after Jane Marcet, the author of the 19th-Century popular economics book, Conversations on Political Economy) to solve it for them. The problem wasn't the use of Jane per se (after all, I encourage them to use her). It was that by using Jane to solve the problem for them, they were missing out on a key learning opportunity.

Probably, I was a little hard on the student. After all, they were using the tools available to them, and engaging in cognitive offloading - reducing the demand or mental load that they face by offloading a task onto generative AI. And increasingly, students are engaging in this cognitive offloading, sometimes in helpful ways, but often in ways that are detrimental. That is one of the conclusions from this 2026 report (with non-technical summary on The Conversation) by Jason Lodge and Leslie Loble (both University of Technology Sydney).

The report has a lot of quotable quotes. For instance, they note that:

It is not possible to engage in critical thinking when one has nothing to think critically about. A person does not simply think critically in a vacuum. A scientist thinks critically about a flawed methodology by drawing on a vast store of knowledge about experimental design. A historian thinks critically about a primary source by drawing on their knowledge of the document’s social, political, and historical context...

This, to me, highlights the key challenge that education faces with generative AI. In order for students to be well prepared for engaging with generative AI, they need to be able to evaluate AI output. And without a thorough grounding in disciplinary knowledge, their evaluations would at best be superficial. And that is why I hold the line on having assessment in my papers that explicitly excludes the use of generative AI. My papers build the foundation on which students' later use of generative AI, and their evaluation of AI outputs, can build.

Lodge and Loble note that:

Every task or learning activity is essentially now a group activity. It just so happens that the other member or members of the group are machines that have practically all human knowledge at their fingertips (in their databases/algorithmic weights). Like any other group activity, students can benefit from that collaboration or get the smart kid to do all the work for them.

That is absolutely what is happening. Students' learning activities are now mostly group activities, even when they are the only human in their group. In group work, how the work is shared is important, and that is where cognitive offloading comes in. Lodge and Loble distinguish between two forms of offloading:

• Beneficial offloading occurs when AI is used to manage extraneous cognitive load (e.g., checking grammar), freeing a learner’s limited working memory to focus on essential, intrinsic tasks.

• Detrimental offloading (outsourcing) occurs when a learner uses AI to bypass this intrinsic cognitive effort (the desirable difficulties) required to build long-term knowledge schemas. This offloading also seems to extend to vital metacognitive and self-regulated learning capabilities, compounding the negative impact of outsourcing on learning.

Importantly, Lodge and Loble note that students typically don't understand metacognition. They haven't intentionally engaged in thinking about their own thinking and understanding how they learn or managing that process. In my experience, many students tend to have been passive recipients of learning approaches, without really engaging with the process themselves. And even those that do engage usually haven't thought deeply about how they learn. And so, when they use a tool that gives them ready answers, it may seem to students that they are learning more efficiently. Lodge and Loble label this an 'illusion of competence', noting that:

Research has long shown that fluent learning materials, such as high-quality videos, can lead people to greatly overestimate how much they have learned by mistaking the ease of processing (fluency) for the depth of learning...

Lodge and Loble's report doesn't stop at the point of diagnosing the problem though. They present three main solutions, that involve:

• shifting generative AI use towards beneficial offloading, where students free up cognitive resources to focus on intrinsic learning. Lodge and Loble offer the example that "AI can be used to provide scaffolding, structured practice, and feedback, all aimed at managing the cognitive burden on the learner and enabling progressive independence...";
• deliberately designing AI interactions to include metacognitive responsibilities, so that students must pause, reflect, and assess their own understanding; or
• shifting the fundamental role of AI from being an 'answer oracle' to a tool that provokes intrinsic load. Lodge and Loble offer examples such as asking students to teach the AI (which plays the role of a confused student), setting up AI as a Socratic tutor, or asking students to independently verify AI outputs.

Those solutions have implications for how I design and use AI tutors in my papers. In order to limit students from engaging in detrimental cognitive offloading, the AI tutor shouldn't simply act as an answer machine. Instead, they should encourage students to attempt problems themselves, offer hints or scaffolding when they get stuck, and ask them to explain or justify their reasoning. And, importantly, the AI tutor could also prompt students to reflect on what they understand, what they don't understand, and whether they could solve the problem on their own. The master prompt for my AI tutors does instruct them to take a Socratic approach, but they don't adhere to it strictly. I'll certainly be putting more thought into the master prompt to see if I can dissuade them from being answer machines and to incorporate more of the metacognitive elements in the future.

The solutions provided by Lodge and Loble are useful, and hopefully they prompt other lecturers to think intentionally about students' (and possibly their own) engagement with generative AI. I especially like the second option, because I believe that we all (and not just students) can benefit from better understanding our learning process, and recognising when we are engaged in genuine and effortful learning. This is not the first time I've encountered concerns about cognitive offloading in the context of generative AI and education (see here). And it is interesting that the same, or a similar, set of solutions keep being presented. In particular, integrating technology as a complement, rather than a substitute, for thinking is important. Generative AI has dramatically lowered the cost of getting answers. It hasn't lowered the cost of learning. A better understanding of metacognition is therefore important too.

Perhaps, then, the lesson from my interaction with the student isn't that they shouldn't have been using generative AI in class. It's that they need to understand when using generative AI in class supports their learning, and when it substitutes for the thinking that learning requires.

I think most universities are now considering explicitly including generative AI in the curriculum, in order to better prepare students for future careers that will no doubt involve substantial interactions with generative AI. Perhaps we should also be considering explicitly including metacognition in the curriculum?

Read more:

• Neuroscientific insights into learning and pedagogy, especially in the age of generative AI