Imagine you are a game show contestant and the host asks a simple question: “The island of Corsica belongs to which country?”
If you’re like most people, you’re probably thinking “Italy.” If asked if that were your final answer, you’d say yes — you’re probably confident “Italy” is the answer.
Unfortunately, though, your confident memory is wrong. Corsica belongs to France.
Over the last five years, I’ve collaborated with Dr. Henry L. Roediger, III to investigate the degree to which our confidence predicts the accuracy of our memories (or our knowledge about the world, as in the example above). The main message of this research — recently published in the journal Psychological Science— is that a subjective sense of high confidence is only sometimes predictive of accuracy. In some cases, high confidence in an answer can mean we’re very likely to be correct. On the other hand, high confidence can sometimes be predictive of an incorrect answer.
In our study, we had undergraduate students memorize words belonging to different categories. For example, students may have studied the words “oriole,” “eagle,” and “cardinal” from a “birds” category. Later, we gave them a recognition memory test, meaning we presented individual words one at a time and asked them whether they had been studied earlier. For items that had been studied, confidence ratings were strongly associated with accuracy — the more confident someone was, the more likely they were to be right.
On this test, however, we also tested students on new (i.e., unstudied) words from studied categories (e.g., “robin” from the “birds” category). These words were tricky: When students saw these words on a test, they were more confident when accepting them as studied (i.e., answering incorrectly; committing a false positive error) than they were when saying they had not been studied earlier. This means that confident answers were in many cases more likely to be wrong than right.
Here’s a figure from the article. What it shows is that for straightforward items that were studied (called targets in the paper), confidence is predictive of accuracy. For tricky items like “robin” that weren’t studied, though (called related lures), the more confident someone was, the more likely they were to be wrong.
In summary, in our research, a high confidence response could mean one of two different things. When a high confidence response was provided for a straightforward, simple test question, the answer was likely to be correct. On the other hand, when students responded with high confidence to a tricky test item, they were likely to get it wrong.
What’s the implication? Consider an analog in an eyewitness testimony situation. If you see someone commit a crime and are asked to choose that person from a lineup, if the perpetrator is actually in the lineup, you should have reasonably high confidence and accuracy in choosing him (or her). If the police were to replace that person with a very similar lookalike, however, you would be very likely to identify that person — an innocent individual— with high confidence.
The bottom line is that we can’t always trust our confidence. Sometimes it’s right, but many times it is likely to be wrong. My Ph.D. dissertation, currently in progress, is tackling some of these issues, so stay tuned for more.
If you’d like to read the paper, take a look here. You can also read some other research we’ve published on this topic, too. If you like, follow me on Twitter for more like this.