By analyzing a dataset of more than 130 million chess games by 70,000 players, researchers at the University of Toronto and the Wharton School at the University of Pennsylvania have discovered “personal bests” act as reference points in human behaviour.
“If your personal best is 28 push-ups and you get 29, then you feel much, much happier than if you do 27, even though they’re very close,” says Ashton Anderson, an assistant professor in the department of computer and mathematical sciences at U of T Scarborough and the tri-campus graduate department of computer science.
“So things are really torqued around at that point – called a reference point.”
Anderson and Etan Green, an assistant professor in operations, information and decisions at Wharton, have published their findings in the Proceedings of the National Academy of Sciences (PNAS).
Reference points such as goals, expectations or the status quo are among those studied by behavioural economists, says Anderson, whose research looks at designing algorithms to support online systems and analyzing datasets of human behaviour.
The Nobel Prize-winning prospect theory revealed humans evaluate potential outcomes as gains or losses.
“People dislike losing five dollars more than they like winning five dollars. People are just loss averse,” says Anderson. “It's very powerful theory, but in order for it to be predictive and useful in the real world, we need to know where these reference points are in people's minds.”
The study began when Anderson and Green were postdoctoral fellows at Microsoft Research in New York City. They looked at chess games played on a server where each player’s best rating was prominent, including notifications that were issued when someone had beat their personal best.
“Everyone was very aware of [their personal best],” he says of their dataset.
Interested in publicly funded research in Canada? Learn more at U of T’s #supportthereport advocacy campaign
So if personal bests are reference points, how do we expect people to behave?
Anderson says there were very clear predictions from that, including that individuals exert extra effort the closer they get to their personal best.
The other clear prediction is that individuals stop playing as soon as they surpass their personal best. They won’t risk playing another game and sinking below where their personal best used to be. But eventually they’ll come back and try again.
“We defined leaving as not playing another game for an hour. We can define it in other ways, like not playing for a day. The results are the same. They won’t drop off completely. They’ll just leave it for a while.”
To bridge models in social sciences and computer science, Anderson believes finding more reference points will make social science theories more predictive.
Although the study is constrained to chess player data, Anderson expects personal bests can be applied to other areas.
“You can imagine making one’s best more salient in any kind of online system. People are genuinely motivated by their personal bests. Ten years ago I had no idea how many steps I was walking every day, and now everyone does. If you had some notion of your maximum or [what] your best [is] ... that might motivate you.”
Computation is having a significant impact on the social sciences, where historically lab studies were limited to the number of participants.
“There's a lot of life being lived in these platforms. Now we have datasets of billions of people, searching for what they want to find, talking to their friends, finding romantic partners and so on.”
“[Big] datasets and the computational resources that we have to process them is akin to the invention of the telescope. Before, we were just looking at the stars with our bare eyes.”