Behavioural economics has seen a rise and fall when it comes to presenting absolutely incredible findings when it comes to explaining human behaviour. The rise was coming up with a plethora of amazing and sometimes quite intuitive findings. Phenomena, biases, heuristics and other behavioural quirks that we could all identify with. The fall then started with some of them not replicating. And then continued with a lot of them not replicating. And some thankfully did replicate, but only under certain conditions. These ever so robust findings became ever so context dependent. Welcome to behavioural economics folks. Now let’s look at loss aversion. A key concept within behavioural economics. If there was ever a phenomenon within behavioural economics that can remain somewhat uncontested it’s this one. Right?!
The Original Paper Loss aversion is a concept from the Kahneman and Tversky (1979) paper on Original Prospect Theory. In this paper, they show that expected utility theory, a core theory of behavioural economics, has a clear reference (starting) point, and quite importantly: works different for gains and losses. Yes, the utility curve, or rather, disutility curve, for losses is very different than that for gains. As clearly explained by this picture.
This picture shows that both curves show diminishing marginal returns (going from 1 to 2 is very different (more impressive) than moving from 1001 to 1002), but one curve is not like the other. Even I, with very poor eyesight, can tell that the losses curve is a lot steeper. This is also where the idea of “losses loom larger than gains” , comes from. Within the original K&T paper, lambda, the loss aversion parameter which looks a “Y” doing a backflip, is jammed in front of the function calculating the change in utility in the domain of gains, to make it a function calculating the change in utility in the domains of losses. Very simple. How big is lambda? Well, within the K&T paper it’s estimated to be 2.25. Meaning that losses do indeed loom larger than gains. About double so. This number set rise to a movement. A wave, if you will. “Losses are felt twice as much as gains” is a phrase many a behavioural economist has taken to heart, whether that interpretation is the correct one or not. But despite this paper being written by a Nobel Prize Winner (not winners, as Tversky had already passed), that doesn’t mean it’s the be-all-end-all.
Failure to Replicate and Alternative Explanations It didn’t take too long for people to become hung-up on the exact number that lambda, or loss aversion in general, was supposed to represent. And even worse, some studies didn’t find loss aversion at all. The theory was challenged, however, in 2018, in a paper by David Gal and Derek D. Rucker on the Social Science Research Network. The two assert that “current evidence does not support that losses, on balance, tend to be more impactful than gains”, and cite a lack of context as undermining the original research. They do also cite the research and studies that did not manage to find a loss aversion parameter as large as the original lambda, or no loss aversion at all. Just for the sake of keeping this post more informal and not just a long list of references, I’m not going to copy all the research rejecting loss aversion here. Do check out the reference list of this post to find the original article, which gives a great overview of the work already done in this domain, and how loss aversion might not be as robust as was initially assumed.
What I am going to address here are the several explanations that have been proposed to explain the lack of confirming loss aversion. Mukherjee et al. (2017) have proposed that there is an effect of size: loss aversion doesn’t work for small payoffs, because people just don’t care enough. It doesn’t hurt enough to invoke what is known as loss aversion. This is often referred to as magnitude dependent loss aversion. Other explanations focus specifically on the size of lambda. The original 2.25 was found with a small sample size (many a paper from the 70s and 80s have sample sizes smaller than 50 participants). It can as such be expected that we the sample size increases, the variability, but also the sheer size of lambda, decreases. So we might just be looking at a much smaller parameter, but indicating loss aversion nonetheless. Last but not least, an entirely new concept has been proposed: Loss Attention. Loss attention is an interesting concept to me. It moves away from focusing on the weighting of outcomes and moves to the effect losses can have on how we distribute attention over (a set of) outcomes (Yechiam & Hochman, 2013).. This theory finds support in research that shows that losses lead to more autonomic arousal than gains even in the absence of loss aversion (Hochman & Yechiam, 2011). So it’s not our weighing of outcomes, but how we divide our attention. And it can be easily argued that a loss might capture a lot more of our attention than a gain does. This does seem to be what the physiological evidence alludes to.
Conclusion To be loss averse, or to not be loss averse? That is not really the question, but a nice nod to Shakespeare is never a bad idea. What is a bad idea is to test something on a sample smaller than 50 people, find something wild and then have the behavioural science community run wild with it. Which is to some extent what happened here. It is not surprising that some results disappear or at least decrease when testing larger samples, this is genuinely to be expected. It is also not very surprising that an effect found in context or setting, with a certain method applied, does not replicate with a different method, or in a different context/setting. As often seen in behavioural science, or behavioural economics if you will, there might be a result, a real phenomenon, it’s just heavily context-dependent. To the surprise of no one.