Confidence Snowballing and Relative Performance Feedback

Feedback is usually considered to be one of the most powerful tools we can use to affect others’ beliefs, performance, decisions or mood. But can feedback actually make things worse by creating and increasing biases? In my previous blog, I discussed that feedback may sometimes hurt if feedback in one aspect of our lives spills over to affect our decisions and performance in other unrelated aspects of our lives. In this blog, I want to talk about another situation where feedback may be hurtful by creating confidence biases.

In a lab experiment, I and my co-author Chris Starmer tested whether feedback that is not informative of participants’ future successes or failures affects their confidence of succeeding or failing. While designing this experiment, it was tricky to come up with a type of feedback that is meaningful enough yet not informative.

To achieve this we asked participants to complete a visual perception task which was either very difficult or very easy. They saw a set of 20 pairs of black circles, each for one second on their screens and then had to guess whether the left or right circle had more white dots in it.

After completing the task, we collected data on their confidence levels by asking them to make series of choices that were relevant for their pay. We then gave participants feedback about their performance ranking relative to their three other group members. So they know that they either scored in the bottom or top half of their groups.

We regrouped them with others in the session who were also either in the top or bottom halves of their groups: so in effect similarly scoring people were grouped together in the new stage. They then had to complete a set of the visual perception task again and we collected their confidence levels again.

We did this three times in total, so there were three stages where participants had to do the task, provide us with their confidence levels, receive feedback and get regrouped for the new stage. How they were regrouped was known to them in our main treatment condition. In additional two conditions we manipulated what they know about how they are regrouped and what their feedback is.

How would a rational participant change their confidence from one stage to the next? Or should they change their confidence at all? Indeed our design ensured that a rational participant should not change their confidence at all since in every stage they are regrouped with people similar to them: In the first stage, participants do not know the quality of the participants they are matched with in a group so their confidence on average should be 50%-50% to be in the top or bottom of their group. Later on, they learn that they were in top/bottom but they also know that now their new group members were also in the top/bottom just like them.

So the situation becomes such that they “fail” and their competition becomes weaker or they “succeed” and their competition becomes stronger. In that situation, on average confidence should not change from one stage to the next… why would it, right?

Yet what we observe is that after learning they were in the top half, participants became overconfident in the next stage. And after learning that they were in the bottom half, participants became underconfident. This seems to be because of reference group neglect, where participants focus too much on the feedback about themselves and disregard the quality of their new group members - something that was speculated and partly shown in a previous experiment by Camerer and Lovallo (1999).

In a very simple lab experiment, we managed to show that on average unbiased participants become biased as they receive feedback from one stage to another. So the feedback snowballs their confidence from being unbiased to being systematically biased either in the overconfidence or underconfidence direction.

Think of recent high school graduates who completed their studies with top grades. They get accepted to the best colleges where students are graded on a curve… We show that disregarding the tougher competition will result in overconfident students who may be little motivated to study hard. The same can happen to employees in organizations whose bonus payments depend on relative rankings. Thinking carefully how to frame and present performance feedback is this crucial to achieve desirable outcomes for students, employees or for a society at large.

Please see our paper for further reference “Confidence Snowballing and Relative Performance Feedback” here.

First published on 22/06/2020