PS 350 Unit 1 cognitive methods paper example

Thirty-Eight Milliseconds per Item: A Cognitive Methods Paper on Sternberg's Memory-Scanning Task, From What Participants Did to the Inference About Serial Search, and Why That Inference Was Later Questioned

Student Name

Psychology, Herzing University

PS 350: Cognitive Psychology

Instructor Name

September 8, 2026

Thirty-Eight Milliseconds per Item: A Cognitive Methods Paper on Sternberg's Memory-Scanning Task, From What Participants Did to the Inference About Serial Search, and Why That Inference Was Later Questioned

The Problem Cognitive Methods Solve

Mental processes cannot be observed directly. A researcher cannot watch a person search memory; she can only see what the person does on a task and how long it takes. Cognitive psychology therefore relies on paradigms, carefully designed tasks whose results can be read as evidence about processing. This paper takes one paradigm, the memory-scanning task introduced by Sternberg (1966), describes what participants actually did, names the measure the task generated, and follows the inferential step from that measure to a claim about how memory is searched, including the reasons that step was later questioned.

What Participants Did

In each trial of the task Sternberg (1966) devised, a participant saw a short list of digits, the memory set, presented one at a time. The set contained between one and six digits. After a brief pause, a single test digit appeared, and the participant pressed one lever if the test digit had been in the set and another if it had not. Participants were instructed to respond as quickly as possible while remaining accurate. The researcher varied the size of the memory set across trials and recorded how long each response took, measured from the appearance of the test digit to the lever press.

The design has three features that matter for the inference. The set was small enough to be held in memory easily, so errors were rare and speed was the informative measure. Set size was varied systematically. And responses were recorded separately for trials where the test digit was in the set and trials where it was not.

Why Reaction Time Is the Measure

Reaction time has a long history as evidence about mental processing. Stroop (1935) used the time participants took to name the ink colors of printed color words, and found that naming was much slower when the word named a different color, as when the word red was printed in green ink. The slowing showed that reading the word interfered with naming the color even though participants were told to ignore the word, an inference about processes no one could see, drawn entirely from timing. Sternberg's paradigm belongs to the same tradition but asks a different question: not whether two processes interfere, but how one process, searching a list in memory, unfolds over time. In both cases the logic is the same. If a manipulation changes how long a response takes, and the change follows a predictable pattern, the pattern constrains what the underlying process can be. Reaction time is useful precisely because it is sensitive to differences of a few hundredths of a second, far below what participants could report about their own thinking.

The Measure

The measure was reaction time, the delay between the test digit and the response, averaged across many trials at each set size. Sternberg (1966) found that reaction time increased in a straight line as set size increased: each additional digit in the memory set added about 38 milliseconds to the response. Just as striking, the slope was almost the same whether the answer was yes or no. Errors were few, so reaction time carried nearly all the information.

The Inferential Step

The step from a straight line to a claim about memory is the heart of the paradigm. Sternberg reasoned as follows. If each additional item adds a constant amount of time, the simplest explanation is that the items are compared with the test digit one at a time, each comparison taking about 38 milliseconds. That is serial search. If the search stopped as soon as a match was found, yes responses should be faster than no responses on average, and their slope should be about half as steep, because on a yes trial the match is found halfway through the set on average. The slopes were the same. Sternberg therefore inferred that the search was exhaustive: participants compared the test digit with every item in the set before responding, even when a match had already been found.

The inference has three parts, and each depends on an assumption. The linear increase supports serial comparison only if the time per comparison is constant. The equal slopes support exhaustive search only if a self-terminating search would produce different slopes. And the conclusion about memory assumes that reaction time reflects the search itself rather than other stages such as encoding the test digit or preparing the response, which Sternberg addressed by noting that those stages would add a constant, changing the intercept but not the slope.

Why the Inference Was Questioned

A straight line can be produced by more than one process. Townsend (1990) showed that parallel models, in which all items are compared at the same time, can produce the same linear pattern if the capacity available for comparisons is shared among items, so that each comparison slows as more items are added. On such a model, the slope reflects limited capacity rather than one-at-a-time search. Townsend argued that serial and parallel processing can look alike in this kind of data but can be distinguished with more elaborate designs. The memory-scanning data alone, in other words, do not uniquely identify serial search.

What the Paradigm Still Establishes

The challenge does not undo the paradigm's main contribution. The finding that recognition time rises linearly with set size, at a nearly constant rate for yes and no responses, has been reproduced many times and remains a benchmark that any model of short-term memory search must explain. What changed is the interpretation: the data constrain possible models without selecting one. That is a general lesson about cognitive methods. A measure can be precise and reliable while the inference from it remains open, and the strength of a paradigm lies partly in making clear what additional evidence would distinguish the remaining explanations. Reading a cognitive result well means holding the measure and the inference apart, crediting the first and testing the second.

References

Sternberg, S. (1966). High-speed scanning in human memory. Science, 153(3736), 652-654. https://doi.org/10.1126/science.153.3736.652

Stroop, J. R. (1935). Studies of interference in serial verbal reactions. Journal of Experimental Psychology, 18(6), 643-662. https://doi.org/10.1037/h0054651

Townsend, J. T. (1990). Serial vs. parallel processing: Sometimes they look like Tweedledum and Tweedledee but they can (and should) be distinguished. Psychological Science, 1(1), 46-54. https://doi.org/10.1111/j.1467-9280.1990.tb00067.x