 |
| Neil Garrett |
This post is by Neil Garrett who recently wrote a paper with Tali Sharot, entitled Optimistic Update Bias Holds Firm: Three Tests of Robustness Following Shah et al.. The paper is to appear in a special issue of Consciousness and Cognition on unrealistic optimism, guest edited by Anneli Jefferson, Lisa Bortolotti, and Bojana Kuzmanovic.
Much of the recent research on optimism has centred around the phenomenon of optimistic belief updating. A well established finding is that healthy individuals are reluctant to revise beliefs when in receipt of “bad news” compared to “good news”. One of the tasks that has been devised to show this in the lab, examines how beliefs about life events (such as being burgled or involved in a car accident) change when individuals find out the events are more or less likely to occur than they initially thought. For example, how much does someone alter their beliefs when provided with evidence that they are more likely to have a car accident than they thought (an instance of bad news) versus when provided with evidence that they are less likely to have a car accident than they thought (an example of good news)? This is referred to as the belief update task. In a recent paper, we set out to examine whether optimistic updating continued to be observed when variations to the belief update task were made.
One aspect we examined asked whether the type of life events individuals are asked to consider makes a difference. A lot of past research on optimistic belief updating has focused on beliefs about unpleasant life events. Individuals can also receive good and bad news about pleasant life events however. For example, finding out that one is more (good) or less (bad) likely to earn a high salary in the future than thought. An obvious tweak to the belief update task therefore was to examine whether an optimistic update bias also existed for pleasant life events. Alongside this, another alteration we made was to test whether updating remained optimistically biased when the events examined were everyday events; events that could plausibly happen in the next few weeks (e.g., being invited to a party). Previously, the events under consideration tended to be grave, significant life events that could occur once or twice in someone’s lifetime, but were unlikely to occur over shorter timescales such as the forthcoming month.
We administered a large online version of the belief update task which applied these changes to the task design. Specifically, participants were asked to report how likely they thought different life events - a mix of everyday pleasant and unpleasant life events - were to occur to them in the next month. They were then shown information about the actual likelihood of these events occurring and asked to re-estimate how likely they thought the events were to occur to them. The results revealed that belief updating - the extent to which participants changed their beliefs - was optimistically biased once again; individuals adjusted beliefs more after receiving good news than they did after getting bad news. Note that in this design, good news could either constitute learning that a pleasant event (e.g., being invited to a party) was more likely to occur, or that an unpleasant event (e.g., catching a cold) was less likely to occur. The opposite cases - learning that a pleasant event was less or that an unpleasant event was more likely - were designated as instances of bad news. This optimistic update bias observed was just as strong for pleasant life events as for unpleasant life events.
Another aspect we examined was whether evidence for biased updating was limited to a specific method of analysing the data collected using the belief update task. Hence we employed alternative approaches to those used previously including models which allowed us to control for possible confounds in a more sensitive way and an analysis comparing updating to a rational “Bayesian” agent. These alternative approaches all revealed an optimistic update bias.
Finally, we reexamined whether altering how “good” and “bad” news were defined, played a role in determining the bias. Usually in the belief update task, news is divided into good and bad types based on whether participants’ beliefs about themselves are at odds with information provided to them in the task. But another valid approach would be to separate news into good and bad according to whether their beliefs about someone like them is at odds with the information provided. The difference between these two might not initially seem obvious but imagine you hold the belief that you have a 40% chance of having a car accident and you also hold the belief that someone like you has a 60% chance. Perhaps you think you take additional car accident preventative measures compared to the average you, making you less accident prone. If I then inform you that the average for someone like you is actually 50%, have I presented you with good or with bad news?
Cases like these are ambiguous because they depend on which beliefs you compare (beliefs about you or beliefs about someone like you) to the information provided. By running a version of the belief update task where we elicited both sets of beliefs, we were able to define news as good and bad using each of these approaches. An optimistic update bias was observed under both of them.
There are a number of other ways that the belief update task and other tasks used to study biases in beliefs can and ought to be varied. Doing so is important for fully understanding the conditions under which optimistic belief updating is observed, the factors that motivate it and characterising it’s boundaries. But the fact that the bias found using the belief update task is robust to variations in task design, to alternate analysis methods and is found by different research groups, supports the view that it is a robust and pervasive phenomenon.
Much of the recent research on optimism has centred around the phenomenon of optimistic belief updating. A well established finding is that healthy individuals are reluctant to revise beliefs when in receipt of “bad news” compared to “good news”. One of the tasks that has been devised to show this in the lab, examines how beliefs about life events (such as being burgled or involved in a car accident) change when individuals find out the events are more or less likely to occur than they initially thought. For example, how much does someone alter their beliefs when provided with evidence that they are more likely to have a car accident than they thought (an instance of bad news) versus when provided with evidence that they are less likely to have a car accident than they thought (an example of good news)? This is referred to as the belief update task. In a recent paper, we set out to examine whether optimistic updating continued to be observed when variations to the belief update task were made.
One aspect we examined asked whether the type of life events individuals are asked to consider makes a difference. A lot of past research on optimistic belief updating has focused on beliefs about unpleasant life events. Individuals can also receive good and bad news about pleasant life events however. For example, finding out that one is more (good) or less (bad) likely to earn a high salary in the future than thought. An obvious tweak to the belief update task therefore was to examine whether an optimistic update bias also existed for pleasant life events. Alongside this, another alteration we made was to test whether updating remained optimistically biased when the events examined were everyday events; events that could plausibly happen in the next few weeks (e.g., being invited to a party). Previously, the events under consideration tended to be grave, significant life events that could occur once or twice in someone’s lifetime, but were unlikely to occur over shorter timescales such as the forthcoming month.
We administered a large online version of the belief update task which applied these changes to the task design. Specifically, participants were asked to report how likely they thought different life events - a mix of everyday pleasant and unpleasant life events - were to occur to them in the next month. They were then shown information about the actual likelihood of these events occurring and asked to re-estimate how likely they thought the events were to occur to them. The results revealed that belief updating - the extent to which participants changed their beliefs - was optimistically biased once again; individuals adjusted beliefs more after receiving good news than they did after getting bad news. Note that in this design, good news could either constitute learning that a pleasant event (e.g., being invited to a party) was more likely to occur, or that an unpleasant event (e.g., catching a cold) was less likely to occur. The opposite cases - learning that a pleasant event was less or that an unpleasant event was more likely - were designated as instances of bad news. This optimistic update bias observed was just as strong for pleasant life events as for unpleasant life events.
 |
| Tali Sharot |
Another aspect we examined was whether evidence for biased updating was limited to a specific method of analysing the data collected using the belief update task. Hence we employed alternative approaches to those used previously including models which allowed us to control for possible confounds in a more sensitive way and an analysis comparing updating to a rational “Bayesian” agent. These alternative approaches all revealed an optimistic update bias.
Finally, we reexamined whether altering how “good” and “bad” news were defined, played a role in determining the bias. Usually in the belief update task, news is divided into good and bad types based on whether participants’ beliefs about themselves are at odds with information provided to them in the task. But another valid approach would be to separate news into good and bad according to whether their beliefs about someone like them is at odds with the information provided. The difference between these two might not initially seem obvious but imagine you hold the belief that you have a 40% chance of having a car accident and you also hold the belief that someone like you has a 60% chance. Perhaps you think you take additional car accident preventative measures compared to the average you, making you less accident prone. If I then inform you that the average for someone like you is actually 50%, have I presented you with good or with bad news?
Cases like these are ambiguous because they depend on which beliefs you compare (beliefs about you or beliefs about someone like you) to the information provided. By running a version of the belief update task where we elicited both sets of beliefs, we were able to define news as good and bad using each of these approaches. An optimistic update bias was observed under both of them.
There are a number of other ways that the belief update task and other tasks used to study biases in beliefs can and ought to be varied. Doing so is important for fully understanding the conditions under which optimistic belief updating is observed, the factors that motivate it and characterising it’s boundaries. But the fact that the bias found using the belief update task is robust to variations in task design, to alternate analysis methods and is found by different research groups, supports the view that it is a robust and pervasive phenomenon.