The psychometric view of social intelligence has its origins E.L. Thorndike's (1920) division of intelligence into three facets, pertaining to the ability to understand and manage ideas (abstract intelligence), concrete objects (mechanical intelligence), and people (social intelligence). In his classic formulation: "By social intelligence is meant the ability to understand and manage men and women, boys and girls -- to act wisely in human relations" (p. 228). Similarly, Moss and Hunt (1927) defined social intelligence as the "ability to get along with others" (p. 108). Vernon (1933), provided the most wide-ranging definition of social intelligence as the person's "ability to get along with people in general, social technique or ease in society, knowledge of social matters, susceptibility to stimuli from other members of a group, as well as insight into the temporary moods or underlying personality traits of strangers" (p. 44).
By contrast, Wechsler (1939, 1958) gave scant attention to the concept. Wechsler did acknowledge that the Picture Arrangement subtest of the WAIS might serve as a measure of social intelligence, because it assesses the individual's ability to comprehend social situations (see also Rapaport, Gill, & Shafer, 1968; Campbell & McCord, 1996). In his view, however, "social intelligence is just general intelligence applied to social situations" (1958, p. 75). This dismissal is repeated in Matarazzo's (1972, p. 209) fifth edition of Wechsler's monograph, in which "social intelligence" dropped out as an index term.
Defining social intelligence seems easy enough, especially by analogy to abstract intelligence. When it came to measuring social intelligence, however, E.L. Thorndike (1920) noted somewhat ruefully that "convenient tests of social intelligence are hard to devise.... Social intelligence shows itself abundantly in the nursery, on the playground, in barracks and factories and salesroom (sic), but it eludes the formal standardized conditions of the testing laboratory. It requires human beings to respond to, time to adapt its responses, and face, voice, gesture, and mien as tools" (p. 231). Nevertheless, true to the goals of the psychometric tradition, the abstract definitions of social intelligence were quickly translated into standardized laboratory instruments for measuring individual differences in social intelligence (for additional reviews, see Taylor, 1990; Taylor & Cadet, 1989; Walker & Foley, 1973).
The George Washington Social Intelligence Test
The first of these was the George Washington Social Intelligence Test, (GWSIT; Hunt, 1928; Moss, 1931; Moss, Hunt, Omwake, & Ronning, 1927; for later editions, see Moss, Hunt, & Omwake, 1949; Moss, Hunt, Omwake, & Woodward, 1955). Like the Stanford-Binet Intelligence Test or Wechsler Adult Intelligence Scale, the GWSIT was composed of a number of subtests, which can be combined to yield an aggregate score. The subtests are:
Judgment in Social Situations;Memory for Names and Faces;Observation of Human Behavior;Recognition of the Mental States Behind Words;Recognition of Mental States from Facial Expression;Social Information; andSense of Humor:
The first four subtests were employed in all editions of the GWSIT. The Facial Expression and Social Information subtests were dropped, and the Humor subtest added, in later editions.
Hunt (1928) originally validated the GWSIT through its correlations with adult occupational status, the number of extracurricular activities pursued by college students, and supervisor ratings of employees' ability to get along with people. However, some controversy ensued about whether social intelligence should be correlated with personality measures of sociability or extraversion (e.g., Strang, 1930; Thorndike & Stein, 1937). Most important, however, the GWSIT came under immediate criticism for its relatively high correlation with abstract intelligence. Thus, Hunt (1928) found that aggregate GWSIT score correlated r = .54 with aggregate score on the George Washington University Mental Alertness Test (GWMAT), an early IQ scale (see also Broom, 1928). A factor analysis by R.L. Thorndike (1936) indicated that the subtests of the GWSIT loaded highly on the same general factor as the subtests of the GWMAT. Woodrow (1939), analyzing the GWSIT with a much larger battery of cognitive tests, found no evidence for a unique factor of social intelligence. R.L. Thorndike and Stein (1937) concluded that the GWSIT "is so heavily loaded with ability to work with words and ideas, that differences in social intelligence tend to be swamped by differences in abstract intelligence" (p. 282).
The inability to discriminate between the social intelligence and IQ, coupled with difficulties in selecting external criteria against which the scale could be validated, led to declining interest in the GWSIT, and indeed in the whole concept of social intelligence as a distinct intellectual entity. Spearman's (1927) model of g afforded no special place for social intelligence, of course. Nor is social intelligence included, or even implied, in Thurstone's (1938) list of primary mental abilities.
Social Intelligence in the Structure of Intellect
After an initial burst of interest in the GWSIT, work on the assessment and correlates of social intelligence fell off sharply until the 1960s (Walker & Foley, 1973), when this line of research was revived within the context of Guilford's (1967) Structure of Intellect model. Guilford postulated a system of at least 120 separate intellectual abilities, based on all possible combinations of five categories of operations (cognition, memory, divergent production, convergent production, and evaluation), with four categories of content (figural, symbolic, semantic, and behavioral) and six categories of products (units, classes, relations, systems, transformations, and implications). Interestingly, Guilford considers his system to be an expansion of the tripartite classification of intelligence originally proposed by E.L. Thorndike. Thus, the symbolic and semantic content domains correspond to abstract intelligence, the figural domain to practical intelligence, and the behavioral domain to social intelligence.
Within Guilford's (1967) more differentiated system, social intelligence is represented as the 30 (5 operations x 6 products) abilities lying in the domain of behavioral operations. In contrast to its extensive work on semantic and figural content, Guilford's group addressed issues of behavioral content only very late in their program of research. Nevertheless, of the 30 facets of social intelligence predicted by the structure-of-intellect model, actual tests were devised for six cognitive abilities (O'Sullivan et al., 1965; Hoepfner & O'Sullivan, 1969) and six divergent production abilities (Hendricks, Guilford, & Hoepfner, 1969).
O'Sulivan et al. (1965) defined the category of behavioral cognition as representing the "ability to judge people" (p. 5) with respect to "feelings, motives, thoughts, intentions, attitudes, or other psychological dispositions which might affect an individual's social behavior" (O'Sullivan et al., p. 4). They made it clear that one's ability to judge individual people was not the same as his or her comprehension of people in general, or "stereotypic understanding" (p. 5), and bore no a priori relation to one's ability to understand oneself. Apparently, these two aspects of social cognition lie outside the standard structure-of-intellect model.
In constructing their tests of behavioral cognition, O'Sullivan et al. (1965) assumed that "expressive behavior, more particularly facial expressions, vocal inflections, postures, and gestures, are the cues from which intentional states are inferred" (p. 6). While recognizing the value of assessing the ability to decode these cues in real-life contexts with real people serving as targets, economic constraints forced the investigators to rely on photographs, cartoons, drawings, and tape recordings (the cost of film was prohibitive); verbal materials were avoided wherever possible, presumably in order to avoid contamination of social intelligence by verbal abilities. In the final analysis, O'Sullivan et. al developed at least three different tests within each product domain, each test consisting of 30 or more separate items -- by any standard, a monumental effort at theory-guided test construction. The six cognitive abilities defined by O'Sullivan et al. were:
Cognition of behavioral units: the ability to identify the internal mental states of individuals;Cognition of behavioral classes: the ability to group together other people's mental states on the basis of similarity;Cognition of behavioral relations: the ability to interpret meaningful connections among behavioral acts;Cognition of behavioral systems: the ability to interpret sequences of social behavior;Cognition of behavioral transformations: the ability to respond flexibly in interpreting changes in social behavior; andCognition of behavioral implications: the ability to predict what will happen in an interpersonal situation.
After devising these tests, O'Sullivan et al. (1965) conducted a normative study in which 306 high-school students received 23 different social intelligence tests representing the six hypothesized factors, along with 24 measures of 12 non-social ability factors. A principal factor analysis with orthogonal rotation yielded 22 factors, including the 12 non-social reference factors and 6 factors clearly interpretable as cognition of behavior. In general, the six behavioral factors were not contaminated by non-social semantic and spatial abilities. Thus, O'Sullivan et al. apparently succeeded in measuring expressly social abilities which were essentially independent of abstract cognitive ability. However, echoing earlier findings with the GWSIT, later studies found substantial correlations between IQ and scores on the individual Guilford subtests, as well as various composite social intelligence scores (Riggio, Messamer, & Throckmorton, 1991; Shanley, Walker, & Foley, 1971). Still Shanley et al. conceded that the correlations obtained were not strong enough to warrant the conclusion (e.g., Wechsler, 1958) that social intelligence is nothing more than general intelligence applied in the social domain.
In one of the last test-construction efforts by Guilford's group, Hendricks et. al (1969) attempted to develop tests for coping with other people, not just understanding them through their behavior -- what they referred to as "basic solution-finding skills in interpersonal relations" (p. 3). Because successful coping involves the creative generation of many and diverse behavioral ideas, these investigators labelled these divergent-thinking abilities creative social intelligence. The six divergent production abilities defined by Hendricks et al. were:
Divergent production of behavioral units: the ability to engage in behavioral acts which communicate internal mental states;Divergent production of behavioral classes: the ability to create recognizable categories of behavioral acts;Divergent production of behavioral relations: the ability to perform an act which has a bearing on what another person is doing;Divergent production of behavioral systems: the ability to maintain a sequence of interactions with another person;Divergent production of behavioral transformations: the ability to alter an expression or a sequence of expressions; andDivergent production of behavioral implications: the ability to predict many possible outcomes of a setting.
As with the behavioral cognition abilities studied by O'Sullivan et al. (1965), the very nature of the behavioral domain raised serious technical problems for test development in the behavioral domain, especially with respect to contamination by verbal (semantic) abilities. Ideally, of course, divergent production would be measured in real-world settings, in terms of actual behavioral responses to real people. Failing that, testing could rely on nonverbal behaviors such as drawings, gestures, and vocalizations, but such tests could well be contaminated by individual differences in drawing, acting, or public-speaking ability that have nothing to do with social intelligence per se.
Still, Following the pattern of O'Sullivan et al., (1965), a battery of creative social intelligence tests, 22 for divergent production of behavioral products and another 16 representing 8 categories of cognition of behavior and divergent production in the semantic domain, was administered to 252 high-school students. As might be expected, scoring divergent productions proved considerably harder than scoring cognitions, as in the former case there is no one best answer, and the subject's responses must be evaluated by independent judges for quality as well as quantity. Principal-components analysis yielded 15 factors, with six factors clearly interpretable as divergent production in the behavioral domain. Again, the divergent-production abilities in the behavioral domain were essentially independent of both divergent semantic production and (converent) cognition in the behavioral domain.
A later study by Chen and Michael (1993), employing more modern factor-analytic techniques, essentially confirmed these findings. In addition, Chen and Michael extracted a set of higher-order factors which largely conformed to the theoretical predictions of Guilford's (1981) revised structure-of-intellect model. A similar re-analysis of the O'Sullivan et al. (1965) has yet to be reported.
In summary, Guilford and his colleagues were successful in devising measures for two rather different domains of social intelligence: understanding the behavior of other people (cognition of behavioral content), and coping with the behavior of other people (divergent production of behavioral content). These component abilities were relatively independent of each other within the behavioral domain, and each was also relatively independent of the non-behavioral abilities, as predicted (and required) by the structure-of-intellect model.
Despite the huge amount of effort that the Guilford group invested in the measurement of social intelligence, it should be understood that the studies of O'Sullivan et al. (1965) and Hendricks et al. (1969) went only part of the way toward establishing the construct validity of social intelligence. Their studies described essentially established convergent and discriminant validity, by showing that ostensible tests of the various behavioral abilities hung together as predicted by the theory, and were not contaminated by other abilities outside the behavioral domain. As yet, there is little evidence for the ability of any of these tests to predict external criteria of social intelligence.
Tests of the remaining three structure-of-intellect domains (memory, convergent production, and evaluation) had not developed by the time the Guilford program came to a close. Hendricks et al. (1969) noted that "these constitute by far the greatest number of unknowns in the [Structure of Intellect] model" (p. 6). However, O'Sullivan et al. (1965) did sketch out how these abilities were defined. Convergent production in the behavioral domain was defined as "doing the right thing at the right time" (p. 5), and presumably might be tested by a knowledge of etiquette. Behavioral memory was defined as the ability to remember the social characteristics of people (e.g., names, faces, and personality traits), while behavioral evaluation was defined as the ability to judge the appropriateness of behavior.