High-Speed Scanning in Human Memory
Sternberg, S. (1966). High-speed scanning in human memory. Science, 153(3736), 652–654. https://doi.org/10.1126/science.153.3736.652
You hold a few digits in memory, a test digit appears, and you decide as quickly as you can whether it is one of the digits you are holding. Sternberg measured how the time to answer changes with the number of digits in memory, s.
In both experiments, reaction time rose in a straight line by about 38 ms for each added digit. The slopes for “yes” and “no” answers were about the same. Sternberg concluded that people compare the test digit with each item in memory one at a time, at 25–30 items per second, and finish checking the whole set even after finding a match (an exhaustive serial scan). A search that stopped at the match would give “yes” answers about half the slope of “no” answers.
Run trials
Press F for yes and J for no, or tap the buttons. Answer as fast as you can without making errors. Press Esc to stop.
Your data from this session
Your trials stay in this page only. Nothing is saved; reloading or closing the page erases them. Means use correct trials only (and, in Experiment 1, only trials where the list was recalled correctly), as in the article.
All trials
How this demo differs from Sternberg’s experiments
Both experiments
- Sternberg’s participants pulled one of two levers. Here you press a key or tap a button, and the browser times your response to within roughly ±10 ms.
- His participants had practice trials (24 in Experiment 1, 60 per set in Experiment 2) and many test trials. Feedback and payoffs rewarded speed with few errors. This demo has no practice, no payoffs, and as many trials as you choose, so your means will be much noisier than his.
- Each run here begins with a countdown so you can place your fingers on the keys.
- His lines were fitted to means over 6–8 participants with about 95–500 trials per point. Your line comes from one person and a few trials.
Experiment 1
- Sternberg showed each digit for 1.2 s. Here each digit shows for 1.1 s followed by a 0.1-s blank, so you can see where one digit ends and the next begins when the same position is reused.
- He chose the list length at random on each trial. Here list lengths and yes/no answers cycle through a shuffled, balanced order, so a dozen trials cover every length from 1 to 6.
Experiment 2
- Each of Sternberg’s participants worked with set sizes 1, 2, and 4 in one session, with nonoverlapping positive sets. Here you run one set size per block, and each block draws a new set.
- The test digits follow his fixed population of 15, so 4 of every 15 tests are “yes” at every set size, and the time from a response to the next test digit is 3.7 s, as in his procedure.