Cognitive Daily

Some people—even people who really know their stuff—just don’t “test well.” You can talk to them face to face, and they seem perfectly well informed and intelligent, but when the money’s on the line, when they’ve sharpened their number 2 pencils and it’s time to sit down for the big exam, they just crumble.

Of course, others simply use “not testing well” as an excuse for true slackerdom, but mere laziness can’t explain the fact that many otherwise well-prepared students don’t perform well under pressure. They score well on the practice SAT, but not on the official test that determines their collegiate future in the United States. Even after taking the test several times, they still can’t match their practice scores.

What’s wrong with these people? Or is there anything wrong?

Certainly anxiety and self-doubt play important roles in the phenomenon, but Sian Beilock of Miami University and Thomas Carr of Michigan State suspected that there was something more fundamental going on. They devised a test to see if working memory might be at the root of the problem (“When High-Powered People Fail: Working Memory and ‘Choking Under Pressure’ in Math,” Psychological Science, 2005).

Working memory is the short-term memory we use to solve problems in our heads. For example, in the subtraction problem 51 – 38, we must “borrow” from the 5, remembering that it’s been reduced to 4, then remembering the result of 3 in the ones place, before finally putting together the final answer of 13. That’s three separate items to keep in memory. A longer problem, like 451 – 278, requires even more working memory. Turn it into a multiplication problem, 451 278, and you’ll exceed most people’s working memory capacity.

It’s known that anxiety or other unpleasant emotions can reduce the available working memory, so this knowledge can lead to a prediction about the impact of working memory during a high-pressure testing situation. Perhaps people who have larger working memory capacity perform better in stressful testing situations. After all, if everyone’s working memory is reduced by anxiety, then the people with larger working memories will still be better off.

Beilock and Carr’s experiment was deceptively simple: test people in high- and low-pressure situations, then test their working memory. Their test involved a simple arithmetic function called Gauss’s modular arithmetic task. In reality the test only required division and subtraction, but since it was expressed in an unfamiliar way, it ruled out the effects of participants who were simply better at arithmetic. During the low-pressure phase, test-takers were simply told they were “practicing.”

Then the experimenters turned the screws. They told participants that they would get $5 if they improved their practice scores by 20 percent. To increase the pressure, test-takers were “paired” with a second volunteer. If both participants increased their scores, then both would be rewarded, but if either participant failed to improved, then no one would get the reward. Next they were told that their partner had already improved her score by 20 percent, so their own score would determine if both partners would receive a reward. Finally, the experimenter set up a video camera and indicated that the test would be taped for examination by local math teachers and professors. (They did stop short of threatening to do bodily harm to the test-takers who failed!)

Finally, test-takers were given a chance to cool down, and subsequently given low-pressure tests of working memory. Beilock and Carr divided the results between the participants who scored high on working memory and low on working memory, and came up with the following:


Instead of performing worse under pressure, test-takers with low working memory did about the same. However, the people with high working memory performed much worse. In low-pressure situations, they did significantly better than low-working-memory participants, but under pressure, their scores declined to the same level as everyone else.

Beilock and Carr speculate that people with larger working memories grow to rely on working memory more for problem-solving. When the availability of working memory is decreased by an anxiety-producing situation, then their problem-solving ability also declines. So working memory is indeed tied to performance under stress, but in exactly the opposite way you might expect.

Oh, and by the way, all test-takers were informed of the purpose of the experiment afterwards—and everyone got $5.


  1. #1 Kevin McGrew
    May 23, 2005

    I just discovered your great blog. A great resource.

    You might want to check out my blog on intelligence testing and theories. I’m going to add your URL to my favorite links section of my blog.

  2. #2 Don
    October 31, 2005

    I think we crack under pressure because we know of the consequence of not succeeding in our goals. Its that fear that cause panic, and during panic our brain and nervous system doesnt function nearly as well as normal situations. Like in a fire, many death are caused by panic when it could be easily avoided.

  3. #3 Bill Tozier
    April 27, 2006

    I just had a related experience yesterday, and I wonder if the dynamics might be more nuanced than simple attrition of working memory that can be applied to the task being tested.

    What I’ve realized, one year into a new career track (engineering) after 20+ years’ success in another couple (theoretical biology and computer science) is that all my planning estimates for the time my work—homework, writing assignments, and other communication tasks—are off by a factor of 4x or 5x.

    It strikes me that in many stressful testing situations, what’s being pushed is the student’s ability to communicate understanding. When being tested, one needs to (1) recognize the explicit and tacit expectations of the questioner, (2) obtain (recall) the domain knowledge and raw facts necessary to answer the explicit question, (3) construct a plan for a cogent response, and then (4) execute and convey that response with the appropriate formalism and idioms.

    In my new domain with its weird idioms and cultural assumptions, my first response to hearing many pedagogic utterances is typically some mix of, “Oh, cool, I’ve never thought of that,” and “Why would anybody in their right minds consider that reasonable?” Talking after the fact with my professors, they have the same response to my work product, at about the same rate. We all are brimming over with domain knowledge: what differs between us is disciplinary framework, and culture.

    Not surprisingly, this leads to increased pressure in testing situations.

    The point being: there are very important issues of pragmatics—and I’m thinking especially here of relevance theory—in the pedagogic cycle. Even in simple tasks like the one examined in this experiment. Instructors aren’t lacking the answer after all; they’re looking for cues reported back from the student that somewhere in that head is a cultural framework and body of knowledge that together might suffice for useful future applications. Two objectives: knowledge and communication.

    I’m thinking back to years spent grading intro biology stuff—and for that matter some of the manuscripts I’m sent for peer review—and especially the “memory dump” pattern that we all know and love. The canonical “memory dump” appears on the face of it as a blind listing of facts, and we’re tempted to see it as a demonstration of failure to synthesize knowledge. I wonder, though, if the author of a “memory dump” might be feeling challenged in the communicative aspects of the process. In conversations, where they can read cues and adjust the flow of talk on the fly, they never get a chance to spew random facts. [Well, not most of them.] They come over as “smarter”.

    So, turning to the results here, I wonder if people use their high working memory more in difficult communication tasks. Beyond just doing the math, they also need to read expectations, frame responses, and dynamically edit and adapt the flow of conversation in social situations.

    What is this “pressure” after all but a separate layer of concurrent social challenge, beyond the simple mathematical task? Being asked not just to succeed on an atomic task, but to learn and improve it? Undertaking this with a partner? Being recorded and shown to a large audience of absent strangers?

  4. #4 Rodney
    December 19, 2009

    One alternative way to interpret these findings is to say that having more working memory could lead participants to have thoughts that are not task specific while doing the task. Meaning they are able to worry about their performance while people with low working memory have no room to worry about their performance. It would be good to do another study in which they measure participants instnaces of worriesome thoughts during the “pressure task” to see if my hypothesis has any merit.

New comments have been disabled.