The Sternberg Short-Term Memory Experiment

Saul Sternberg published the original experiment of this kind in the early 1960's and since then it has become one of the 'classic' examples of the information processing paradigm in cognitive psychology. It involves a simple technique to study mental processes which has been used in countless studies ranging far beyond the original purpose that Sternberg designed it for.

One reason for the significance of this experiment is that it reveals the complex nature of conscious thought using objective measurements. Short-term memory is often equated with conscious thinking - it is also referred to as 'working memory', the part of the mind where thinking takes place. Although we have direct access to our own thoughts, it is difficult for cognitive psychologists to study thought processes because (a) we cannot directly observe other peoples' thoughts, and (b) we cannot be sure that our own introspections are valid.

You are given a very simple task to perform: A small group of items called the positive set is presented for you to memorize. You are then shown a single test item that may, or may not, have been shown before. You have to respond 'yes' or 'no' accordingly. This procedure is repeated over several trials in which the number of items in the positive set is varied. You are cautioned to respond as fast as you can without making errors. An example is given below, in which the items are digits. Only three trials are shown, but generally more would be given.

                    Trial 1       Trial 2       Trial 3
Positive set         2,5,3       8,4,9,2,7        4,2
(hold in memory)
Test item             5             6              4
(did this occur
in positive set?)
Correct response     YES            NO            YES
Your response, and the time to respond 'YES' or 'NO' are recorded. Subjectively, when doing this, you may feel your responses are almost instantaneous. However, accurate recording of the time, called reaction time (RT), shows that approximately half a second may elapse between presentation of the test item and your response. This illustrates point (b) above, namely that introspection may be misleading.

Sternberg's discovery was that RT varies with the number of items in the positive set. He measured average RT over large numbers of trials using positive sets containing from one to six items. With increasing set size, reactions took longer: From less than half a second for a single item, to almost one second for six items. Accurate measurements in milliseconds (1 ms = 1/1000 second) showed that for each item in the positive set there was an increase of about 40ms in RT. The implication was that a response to the test item involved some kind of search through working memory: To answer 'YES' or 'NO' requires that you first compare the test item with each item in the positive set before deciding on the response. If there is only a single item in the positive set the decision to respond can be almost instantaneous, but if there are six items to be compared the decision should take longer.

A further discovery was that mean RT for 'YES' and 'NO' responses was approximately the same. This seemed to contradict the logic of searching through a list: If the test item is in the positive set you can stop searching and respond 'YES' as soon as it is found (as in Trial 1 above). But suppose the test item is not in the positive set. You must always search the entire positive set before you can decide if this is the case and respond 'NO' (such as in Trial 2 above). By this logic 'YES' responses would on average take less time than 'NO' responses. However, Sternberg could explain this result terms of an exhaustive search process. He argued that short-term memory search always goes completely to the end without stopping when an item is found. Although it is counter-intuitive, the results support such a theory. It also makes sense in terms of a very rapid automatic search that cannot be stopped if the item is found. It goes to the end anyhow, and then the response is made. So 'YES' and 'NO' responses can plausibly take equally long.

This discovery illustrates point (a) above: Sternberg was able to make objective inferences about mental process that can not be observed directly. Since Sternberg formulated his theory about mental processing a great deal of research has been done to investigate these claims. Some futher details are presented below, and there is disagreement about his theory regarding the search process. However, we are not concerned with the exact details here, only the broad picture.

One investigation used the varied-set procedure, in which the subject had to memorize a different set of items on each trial. The items were randomly selected and presented to the subject sequentially for about one second each; there was a delay of two seconds during which rehearsal could occur, and then the test item was shown. Reaction time was measured from this point until when the subject pulled one of two levers to indicate a response. Different items in different set sizes were presented in random order. We shall use this procedure for our own experiment.

Sternberg reported results averaged over eight subjects (where POSITIVE and NEGATIVE responses are 'YES' and 'NO' respectively). His original plot and figure caption are shown below:

Note that these results indicate a linear relationship between set size and mean RT. Sternberg gives the regression equation for this relationship as mean RT = 397.2 + 37.9(set size). You can see the linear relationship between the plotted points. They are all close to the regression line.

Note further that Sternberg used mean RT as the dependent variable. Each subject did about a hundred trials for each set size in random order: The circle plotted on the graph above represent the mean of all those trials over eight persons. Subjects had to practice so they made only about 1 or 2 errors in a hundred trials.

Memory set sizes up to six items were used by Sternberg because short-term memory capacity is limited. He did not want to test capacity but retrieval speed, so he made the task quite easy, with emphasis on speed. Eight or nine items would cause great difficulty remembering all of them. Errors would increase, and the test of retrieval speed would become invalid. One can, however, extrapolate the straight line shown above to beyond set size six, and see what the theory would predict.

Use of the task in assessing effects of neurotoxins

In a more recent study by Smith & Langolf (1981), the Sternberg short-term memory experiment was used to measure the effect of exposure to toxic chemicals. Twenty-six factory workers who had traces of elemental mercury in their urine were tested with the task. The authors conclude that memory scanning time was affected by exposure to mercury. In other research, this task has been used to detect harmful effects resulting from exposure to solvents in the workplace. This experimental task has provided valuable evidence of subtle subclinical cognitive harm when no directly observable neurological disturbances are apparent in the people exposed to toxic substances.

14 September 1999