Gf task analysis via Sternberg's unified reasoning theory

During my vacation I read a relatively old article by Robert Sternberg on human reasoning. I read his work via my CHC lens and have summarized the key information below. I believe this information can be translated by good practitioners into useful insights regarding the performance of individuals on various Gf tests. I have taken the liberty to insert CHC abbreviations (Gf, I, RG, etc.) in my summary of his writing.

Sternberg, R. (1986). Toward a Unified Theory of Human Reasoning, Intelligence, 10, 281-314 (1986)

• “Human reasoning has been a topic of serious study at least since Aristotle and continues to be an important topic of psychological theory and research today. The omnipresence of reasoning in our lives encourages us both to understand the processes by which we reason and to identify the sources of error that sometime lead us to mistaken conclusions” (1986, p.281)

In 1986, Sternberg specified a unified theory of human reasoning (Gf). His unified theory of Gf conceptualized reasoning as “the controlled and mediated application of three processes-- selective encoding, selective comparison, and selective combination--to inferential rules.” According to Sternberg, the presence of any of these processes defines a task/problem as Gf whereas any task where the solution is not dependent on any of these processes is not to be considered Gf.

As per the CHC taxonomic definition of Gf, selective encoding and comparison primarily underlie the narrow Gf ability of inductive reasoning (I) while selection combination is characteristic primarily of general sequential (deductive) reasoning (RG).

Sternberg’s Gf processes defined

• Selective Encoding. In many everyday problems and tasks (as well as psychometric tasks) individuals are bombarded with a large array of information, only a portion of which is relevant to understanding and solving the task at hand. Selective encoding is the process employed to distinguish relevant from irrelevant information. In many such tasks individuals must decide which bits of information/stimuli are relevant to solving the problem. This relevance decision-making process is believed to occur within working memory (Gsm-MW).

• Selective Comparison. The solution to most reasoning problems requires individuals to retrieve declarative and/or procedural knowledge from the vast stores of acquired information. Given the breadth and depth of an individual’s domains of acquired knowledge, a mechanism is needed to decide which stored information is potentially problem-relevant. Selective-comparison is the process by which individuals retrieve only those bits of information that are potentially relevant to problem solution. This process involves accessing (Glr) the stores of acquired knowledge (e.g., Gq, Gc, Grw).

• Selective Combination. Once information has been selectively encoded and compared, the two components are combined in working memory (Gsm-MW).

Consistent with the CHC-based definition of Gf (novel problem solving), Sternberg provides one caveat. Namely, selective encoding, comparison and combination only define a reasoning situation/task to the “extent that they are executed in controlled, rather than automatized, fashion” (p. 286). According to Sternberg’s “graduated” view, automatization lies along a continuum with problems that rely less on automatization requiring greater degrees of Gf.

Sternberg I vs. RQ distinction

According to Sternberg (1986), the essence of inductive reasoning (I) derives primary from “selective encoding and selective comparison processes, both of which involve sorting of relevant from irrelevant information. The only constraint is that there be no logically determinate solution to the problem; in other words, it should not be the case that the use of certain information is logically correct and the use of other information logically incorrect” (p. 293).

In contrast, the defining feature of deductive reasoning (RQ) derives primarily from “selective combination processes, with the constraint that there be one or more logically determinate solutions to the problem. In other words, certain combinations of information must be logically correct, and others logically incorrect (p. 294).

Mediators of Gf performance

According to Sternberg (1986), a Gf problem “may be made easier or harder simply by varying the availability or accessibility of such rules through the use of mediators” (p. 292). A mediator is defined as any intervening variable that will increase or decrease the availability or accessibility of the inferential rules to be used in a particular Gf problem. Below is a non-exhaustive list of potential Gf mediators provided by Sternberg.

• Prior Probability. The subjective likelihood estimate that an individual brings to a reasoning task for the use of a given inferential rule (or set of inferential rules).

• Posterior (Contextual) Probability. The probability that an individual will apply a certain kind of inferential rule to a problem can be influenced by “internal context clues” in the problem task. An example provided by Sternberg, relevant to verbal Gf verbal test items (e.g., verbal analogies), is that an individual can often figure out the meanings of unknown words vis-à-vis knowledge provided by prefixes, stems, and suffixes.

• Entrenchment. Some inferential information (rules) are more familiar (entrenched) in one’s experience than other rules, with entrenched rules being more readily applied to problem solutions. For example, individuals are more familiar with using positive (versus) negative information in Gf tasks (Sternberg, 1986).

• Prior Knowledge. An inferential rule relevant to Gf problem solution cannot be employed if an individual does not know the rule. The availability (in contrast to the accessibility – characteristic of the three mediators described above) of a rule is considered a prior knowledge mediator.

• Working-memory capacity (Gsm-MW). The resource-limited constraints of the proverbial information processing “bottleneck” (working memory) hinders the amount of space required for encoding and combining problem-relevant information. Large complex Gf problems that require the mental juggling of increasing amounts of information and rules are rendered more difficult by the limited resources of the working memory system.

• Representational capacity. People vary in their ability to represent information in different (e.g., linguistic; spatial) formats. According to Sternberg (1986), “two people may be equally adept at applying a given procedural rule such as inferring the relation, but have differential difficulty on a given problem because of their differential ability to apply that procedural rule to a given mental representation. One individual might find it easier to apply the rule in a spatial domain, and the other to apply the rule in a linguistic domain” (p. 292). The fluency or efficiency in applying various kinds of processes to Gf tasks may depend upon the quality of the mental representation used during the problem-solving process.

• Content-induced biases. The type of content in a Gf problem might influence task performance. For example, research has demonstrated that categorical syllogisms with emotionally charged content are more difficult than those without emotional content.

• Form-induced biases. The form or structure of a Gf task may also introduce bias. Form-induced biases are biases that are introduced by the form rather than the content of a given reasoning problem.