Table of contents
1. Intelligence - Introduction
1.2. Sternberg’s Theory Of Successful Intelligence
1.3. Social Intelligence
1.4. Multiple Intelligences
2. Intelligence and Human Abilities
2.1. Intelligence As A Unitary Concept
2.2. Carroll’s Three Stratum Theory of Intelligence
2.3. The Berlin Structural Model Of Intelligence
2.4. Problem Solving Ability And Learning Test Concept
3. Intelligence and Speed of Mind
3.1. Studies On The Connection Between RT And Intelligence
3.1.1. The Hick-Paradigm
3.1.2. The Sternberg-Paradigm
3.1.3. The Posner-Paradigm
3.1.4. Odd Man Out-Paradigm
3.2. Starting Point
3.3. A Developmental Perspective
3.4. Cultural Invariance
3.5. Uniformity Of RT Measures
3.6. Do Different RT Paradigms Yield Different Correlational Patterns With Intelligence?
3.7. Information Psychology And The MS-IQ Correlation
3.8. Predicting Intelligence From RT Measures
3.9. The Issue Of Movement Time
3.10. Specificity Or Singularity Of Mind
3.11. RT-IQ Correlation And Its Dependence On Complexity
3.12. Inspection Time And Intelligence
3.13. Psychophysiological Measures Of Speed And Their Correlation With Intelligence
3.14. The Issue Of Causation
4. CriticismOf Mental Speed Approach
5.2. Emotions In General
5.3. The Prototypical Model Of Emotions
5.4. The Concept Of Basic Emotions
5.4.1. Criticism of the Basic Emotion Concept
5.5. Vascular Theory of Emotion
5.6. Neural Structures And Emotional Processes – Evidence From Mental Deficit Studies
5.6.1. The Issue of Schizophrenia
5.6.2. The Issue of Autism
5.7. Brain Structures And Emotions
5.7.2. Facial Agnosia
5.8. Withdrawal/ Approach System
5.8.1. The Prefrontal Cortex
5.8.2. The Amygdala
5.8.3. Withdrawal/Approach System in the Context of the Dimensional and Categorical View on Emotions
5.8.4. Handedness and recognition of emotions
5.9. Regularity In Emotional Experience
6. ConceptsOf Emotional Intelligence
6.1 EI Conceptualized According To Salovey And Mayer
6.1.1 Is EI an Intelligence?
6.1.2 Further Criticism on the Ability Concept of EI
6.1.3. The Schutte EI Scale
6.1.4. The Arrival of Objective Measures
6.2. The Bar-On EQ-I
6.3. EI – Conclusions
7. RecognitionOf EmotionsIn Faces
7.1 Early Investigations
7.2. Beyond The Use Of Only Positive Vs. Negative Emotions
7.3. The Role Of Context And Other Sources Of Information
8.INTEGRATION OF THE RESEARCH REPORTED
9.2.1. Construction of Tests
22.214.171.124. Reaction Time Tests
126.96.36.199.1. Emotional reaction time task
188.8.131.52.2. Non-emotional reaction time task
184.108.40.206. The Posner Paradigm Task
220.127.116.11.1. Emotional Posner paradigm test
18.104.22.168.2. Non-emotional Posner paradigm test
9.2.2. Designation, Derivation and Meaning of the Developed Testscores
9.2.3. Intelligence Test
10.1. TEST CHARACTERISTICS
10.1.1. IQ Testscore
10.1.2. Mental Speed Testscores
10.1.2.1. Data Cleaning
10.1.2.1.1. Ambiguous items
10.1.2.2. Distribution of MS testscores
10.1.3. Dependence of Testscores on Situational and Personal Variables
10.1.3.1. Age of subjects
10.1.3.1.1. Age and IQ
10.1.3.1.2. Age and MS
10.1.3.2. Gender of subjects
10.1.3.2.1. Gender and IQ
10.1.3.2.2. Gender and MS
10.1.3.3.1. Testorder and IQ
10.1.3.3.2. Testorder and MS
10.1.3.4. Button control
10.2.RELATIONSHIP AMONG MS TESTS
10.2.2. Relationship Between the Emotional and Non- Emotional MS Tests
10.2.2.1. Emotional and non-emotional RT tests
10.2.2.2. Emotional and non-emotional PP tests
10.2.2.3. Comparing RT and PP tests
10.3.RELATIONSHIP BETWEEN IQ AND MS TESTSCORES
10.3.1. IQ and RT Testscores
10.3.2. IQ and PP Testscores
10.3.3. Factor Analysis Including IQ, RT, and PP Testscores
10.3.4. Regression Analysis
10.3.5. AMOS Model Building
10.4.RESULTS SEPARATED FOR CULTURAL GROUPS
10.4.1. Relationship between emotional and non-emotional MS testscores
10.4.1.1. Emotional and Non-emotional RT Tests
10.4.1.2. Emotional and Non-emotional PP Tests
10.4.1.3. Comparing RT Tests with PP Tests
10.4.2. Relationship between IQ and MS Testscores
10.4.2.1. IQ and MS Tests
10.4.2.2. Factor Analysis Including IQ, RT, and PP Testscores
10.4.3. The AMOS Model – One Model for All?
11.1.The Use Of Mental Speed Measures
11.2.Relationship Among Emotional And Non-emotional MS Tests
11.3.Relationship Between IQ And MS Testscores
12. FOLLOW-UP STUDY
12.2.1. Sample Characteristics
22.214.171.124. Data Cleaning
126.96.36.199. Distribution of Testscores
188.8.131.52. Reliability – Internal Consistency
184.108.40.206. Correlation to Personal Variables
220.127.116.11.1. Age of subjects
18.104.22.168.2. Gender of subjects
12.3.2. Relationship among MS Tests
22.214.171.124. Relationship in the Follow-up Study
126.96.36.199.1. Emotional and non-emotional RT tests
188.8.131.52.2. Emotional and non-emotional PP tests
184.108.40.206.3. Comparing all MS scores
220.127.116.11. Relationship between Main and Follow-up Study
18.104.22.168.1. Stability in MS scores
22.214.171.124.2. Stability in correlation to IQ
126.96.36.199.3. Stability in the AMOS model
12.4.1. The Level Of Testscores
12.4.2. The Level Of Relationship Pattern
12.4.3. Stability In The AMOS Model
13. GENERAL DISCUSSION
(A.) List of Subjects’ Countries of Origin
(B.) Instruction for Emotional RT
(C.) Instruction for Non-emotional RT
(D.) Instruction for Emotional Posner Paradigm Test
(E.) Instruction for Non-emotional Posner Paradigm Test
(F.) FA with all MS testscores and the APM score
(G.) FA for RT and LTM scores separated for cultural groups
(H.) FA for RT, LTM and APM scores separated for cultural groups
(F.) FA for the Japanese Sample Employing all Testscores
(G.) FA for the International Student Sample Employing All Testscores
(K.) Intelligence Differences according to Cultural Background
(L.) Absolute Values for the MS Testscores in Milliseconds
Table 1:Dependence of correlations between RT measures and intelligence on the complexity of RT tasks
Table 2:The pattern of possible conditions in the Posner paradigm
Table 3:List of all MS testscores
Table 4:MS testscores and normal distribution
Table 5:MS tests and their reliability
Table 6:MS tests and their correlation with age
Table 7:Correlations among emotional and non-emotional RT tests
Table 8:Correlation among LTM scores
Table 9:Correlations among RT and LTM scores
Table 10:Factor analysis of RT and LTM scores – rotated matrix
Table 11:Relationship between IQ and RT testscores
Table 12:Relationship between IQ and LTM scores
Table 13:Eigenvalues and explained variance for the three extracted factors
Table 14:Rotated factor matrix for RT, LTM and the IQ Score
Table 15:Correlations among emotional and non-emotional RT tests for the Japanese and the international sample
Table 16:Correlation among LTM scores for the Japanese and international sample
Table 17:Correlations among RT and LTM scores for the Japanese and the international sample
Table 18:Relationship between IQ and RT testscores for the Japanese and the international sample
Table 19:Relationship between IQ and LTM scores for the Japanese and the international sample
Table 20:Factor composition separated for the Japanese group, the international student group and the total sample
Table 21:Comparing the fit indices for the Japanese and the international student sample
Table 22:Internal consistency for the MS testscores in the follow-up study
Table 23:Correlations among emotional and non-emotional RT tests in the follow-up data
Table 24:Correlations among LTM scores in the follow-up data
Table 25:Correlations among RT and LTM scores in the follow-up data
Table 26:Factor analysis of RT and LTM scores for the follow-up study
Table 27:Test-retest reliability for the MS testsscores
Table 28:Correlation between the MS tests from the follow-up study and Raven’s APM scores
Table 29:Factor analysis of MS testscores and intelligence for the follow-up study
Figure 1:From Eysenck; Intelligence: the new look (p. 334)
Figure 2:Model of psychostructure, from Lehrl and Fischer (p. 885)
Figure 3:Theoretical Circular Model, from Russel, 1980 (p.1164)
Figure 4:AA stimulus (shape same/ meaning same) for the emotional and non-emotional PP test
Figure 5:Ab stimulus (shape different/ meaning different) for the emotional PP test
Figure 6:Aa stimulus (shape different/ meaning same) for the emotional PP test
Figure 7:Ab stimulus (shape different/ meaning different) for the non-emotional PP test
Figure 8:Aa stimulus (shape different/ meaning same) for the non-emotional PP test
Figure 9:Correlations among LTM scores for emotional and non-emotional scores
Figure 10:Correlations among RT and LTM scores
Figure 11:AMOS Model (N=99, cases with missing values were deleted)
Figure 12:Modified AMOS model; changes base on the follow-up data
The EI construct, which developed mainly during the last decade, has been split into an ability-EI and a trait-EI. Originally EI was proposed as an intelligence, needed because emotional information is processed differently from non-emotional information. However, due to the lack of adequate objective measures for ability-EI, trait-EI became popular. In this study, an innovative approach building on the use of mental speed (MS) measures was theoretically justified and applied to emotional information processing in order to tackle the criterion problem for emotional test items and to investigate the relationship between mental speed and intelligence. Emotional and non-emotional tests following a simple reaction time (RT) and the Posner paradigm were constructed and administered along with Raven’s APM to a sample of 121 college and university students at Hiroshima University, Japan. Approximately half of these students were Japanese nationals, while the other half were international students, coming from countries all over the world.
The APM score correlated low but significantly with both RT tests, but neither with the emotional nor the non-emotional long term memory (LTM) access times. In varimax factor analyses (FA) the emotional MS tests grouped together to one factor, while IQ and non-emotional MS tests formed two more factors. Additionally, an AMOS (Analysis of Moment Structures; an SPSS compatible program for structural equation modelling distributed by Smallwaters Company) model confirmed the existence of two factors, one for the emotional and one for the non-emotional testscores.
This general picture did also not change substantially, when the data were analysed separately for the two groups (Japanese, international students). Although there were some changes in the size of single correlations among MS testscores, the results of the FA as well as the AMOS model was supported in the separated group data.
In order to test temporal stability, the MS tests were re-administered to 48 students (20 Japanese and 28 international students) from the original sample. The test-retest reliability for most of the MS scores was low (between .35 and .80) compared to the reliability of psychometric tests. Additionally, the APM score from the main study was correlated significantly with all MS testscores in the follow-up data. Nonetheless after the speed and the emotional factor of the AMOS model were allowed to correlate to each other also on the level of single testscores, the corrected model showed very favourable fit indices for the follow-up data as well as for the data from the main study. Thus, the EI core assumption was supported in this investigation.
Human intelligence is one of the best researched subjects in scientific psychology. Yet in the recent past the construct of intelligence itself became a matter of discussion and different researchers introduced new aspects of intelligence or proposed alternative constructs to substitute for so-called ‘academic’ intelligence.
Such new constructs were proposed e.g. from Gardner (1998; multiple intelligences), Sternberg (e.g. Sternberg & Kaufman, 1998; practical intelligence), Salovey & Mayer (e.g. 1989-90; emotional intelligence) and others. The starting point of their criticism was similar – intelligence as measured with traditional tests seemed insufficient to describe human abilities. Soon the term “academic” intelligence was created to emphasize the context of academic settings and achievement. Academic intelligence tests are acknowledged as the best single predictor of success in a variety of school subjects. However, the same tests do not predict the performance in a variety of life settings, including such important issues as work performance or the quality of interpersonal relationships, very successfully.
1.2. Sternberg’s Theory of Successful Intelligence
Hence, new concepts were supposed to fill this gap. One citation from Sternberg (1979) reflects well the different approach he was taking toward human intelligence: “…factors (such as inductive reasoning or deductive reasoning), are constructs to be accounted for rather than the constructs that do the accounting.” (p. 228). This approach also was an answer to Cronbach’s (1957, 1975) demand to merge two disciplines of science – experimental psychology and the study of individual differences. For Sternberg’s theory of mental abilities, the psychometric approach needed to be re-evaluated along the lines Cronbach had laid out. To develop a concept on intelligence is not facilitated by factor analyzing the results of different people, but by analyzing the difference among the mental tasks available. A deeper understanding of human problem solving abilities needs the understanding of the elements that constitute a certain task.
Sternberg later introduced the idea of practical intelligence, which is related to superior work performance of people, who are regarded less intelligent by traditional testing procedures, in comparison with worse work performance of supposedly more intelligent people. Practical intelligence, hence, should describe the ability “to implement options and to make them work […] is involved when intelligence is applied to real-world contexts”(Sternberg & Kaufman, 1998, p. 494). The key aspect of practical intelligence is the acquisition and use of “tacit” knowledge.
Sternberg (1985) divides tacit knowledge in three categories – tacit knowledge about managing self (how to manage one’s own daily condition, productivity and so on), managing others (how to use social relationships) and managing career (how to establish a reputation). In a series of experiments, Sternberg tried to evaluate the contribution of tacit knowledge to successfully handling real-world situations. The tacit knowledge measure was significantly correlated with important parameters of success in the field of business (i.e. with level of company reputation, years of schooling beyond high school, and salary, status as expert vs. novice).
Practical intelligence is but one of two concepts vital for interpreting individual differences in intelligence. The other construct is creative intelligence, which is also not well represented in traditional psychometric testing procedures (Neisser et al., 1996). Creative intelligence is mainly used to generate problem-solving options (Sternberg & Kaufman, 1998) and forms together with analytical (academic) and practical intelligence “successful” intelligence. Successful intelligence is the ability “…to adapt to, shape, and select environments to accomplish one’s goals and those of one’s society and culture.” (Sternberg & Kaufman, 1998, p. 494).
Sternberg also criticizes the different kinds of problems used in academic intelligence tests compared to real world problems. Items in intelligence tests are formulated by people different from oneself, are clearly defined, come with all information needed, have only one single correct answer and often have little or no intrinsic value. Real world problems, however, tend to be poorly defined, require a lot more information seeking and have different, albeit adequate, solutions.
1.3. Social Intelligence
Social intelligence was already proposed by Thorndike but due to difficulties in measurement and ambiguity in the meaning of the term social intelligence, this construct never attracted as much research as the academic intelligence concept did.
With the academic intelligence construct under pressure, however, interest in social intelligence revived again. Hence, also measurement was improved and 1989 Costanzo and Archer published a measure for the ability “to observe and interpret the expressive behavior of others” (p. 226). This new measure – the Interpersonal Perception Task (IPT) – used video-taped scenes from real-life interactions between different people. The IPT’s construction also drew on the experiences gathered with a predecessor – the Social Interpretation Task (SIT, Archer & Akert, 1977). The task of the subject was to answer a certain question from a given social interaction. For instance, there was a dialogue between two women, while one child was present. The task was to infer from the film clip, who the mother of the child was. One advantage of the IPT certainly is the availability of a ‘correct’ answer.
The newly available measurement further stimulated interest and research in the area of social intelligence. Brown and Anthony (1990), for instance, assessed its usefulness in relation to academic intelligence. The correlation between social intelligence and traditional intelligence was quite low (.15), and a regression analysis used both measures to predict scores on dependent variables.
1.4. Multiple Intelligences
While Sternberg just added another ability, Gardner (1998) reinvented the whole construct of intelligence. Gardner insisted that different aspects of thinking were represented by different – so-called multiple – intelligences. Gardner drew his inferences mainly from the study of highly gifted people and brain damages. As for the gifted people, he found that the vast majority was gifted only in one area. This area could be quite narrow, such as handling mathematical operations with an incredible speed, or less narrow, such as the ability to handle interpersonal situations with success. However, the giftedness usually does not generalize to other areas. Gardner proposed the following eight forms of intelligence: (1) linguistic (e.g. in writing a poem or reading comprehension), (2) logical-mathematical (e.g. in deriving mathematical proofs), (3) spatial (e.g. using street maps), (4) musical (e.g. singing songs or composing music), (5) bodily-kinesthetic (e.g. in dancing or playing sports), (6) interpersonal (e.g. understanding and interacting with others), (7) intrapersonal (e.g. understanding oneself), and (8) naturalist intelligence (e.g. discern patterns in nature, like Charles Darwin did).
Besides the study of giftedness, Gardner also used examples from selective brain damages to illustrate and support his theory. Like giftedness, brain damage can result in the complete loss or strong handicap in one special area of thinking. For instance, people do not understand language anymore, while they are still able to express themselves adequately.
Additional criteria for intelligences are: the existence of certain core operations, a distinctive developmental history, a distinctive evolutionary history, evidence from cognitive-experimental research as well as from psychometric tests and a symbol system for any of the eight intelligences.
In this study, however, the third of the mentioned constructs will be investigated – the concept of emotional intelligence (EI) by Salovey and Mayer (1989-90).
Salovey and Mayer (1989-90) did not reinvent the idea of intelligence as Gardner did. Their research line resembles the idea of Sternberg, namely to add a new construct to the original academic intelligence construct, which should in combination raise the predictional power for life settings. EI is supposed to play a critical role for success in interpersonal situations.
This study investigates the relationship between one aspect of emotional intelligence and academic intelligence. The theoretical part of this paper will outline the theoretical basis of intelligence constructs in the first chapter. The second chapter is dedicated to show that mental speed (MS) measures are useful parameters of academic intelligence. Chapter three contains a description of EI theories and empirical results on emotion recognition.
2. Intelligence and Human Abilities
Unfortunately it is not possible to open this chapter with a simple definition of intelligence, since it is defined very differently by different researchers.
It might be interesting, though, to investigate the term intelligence itself. Spiker and McCandless (1954) characterized different definitions for scientific terms. On the one end, they positioned terms that “are reducible to very simple terms” (p. 255). These terms cannot be linguistically explained; only direct experience with the referents of such terms like “blue” or “above” can lead to an understanding of their meaning. Such terms are called “primitive predicates” and usually in scientific practice, scientists do not bother to define constructs in such primitive predicates. While primitive predicates derive their meaning from experience, intelligence belongs to a second form of concepts, which are described in words and builds on the relations among different aspects, i.e. item characteristics, predictive power, testing procedures and so on.
Consequently, it could be argued that there is not one construct of intelligence but as many as there are measuring procedures. According to this view, each intelligence test is a definition of intelligence, with its own particular meaning. This operational definition, however, clarifies our understanding of intelligence only within rigid restrictions (Spiker & McCandless, 1954). Hofstätter (1966), a philosopher grounded in Leibniz’s work, tries to tackle the core of the construct of intelligence. He supposes that intelligence is the difference between animal and human beings. However, even animals are able to learn. When punished for a behavior an animal usually avoids that behavior in future situations. This consequence resembles inferential thinking. It is built, however, solely on the power of experience. Animals, Hofstätter therefore argues, are guided fully by experiences, not by inferences drawn from knowledge. Human beings also use experiences as guidance to a large part. Leibniz supposed that three parts of human behavior is guided by simple experience. Only one fourth of human behavior is building on inferential processes. The basis for such inferential processes, which he calls intelligent behavior, is an implicit statistic in human beings (Hofstätter, p. 232). Although experience, in form of memories, is essential for this statistic to work meaningfully, it is not sufficient. The step from experience guided behavior to behavior guided by one’s own inferences needs the prediction of future order. Such predictions are easier, when memories are ordered using a language. Mathematics is such a language and therefore Hofstätter argues that a person using e.g. probability calculations to predict future events is acting intelligently, while a person using only his or her memory is not acting intelligently.
It is not necessary to use mathematics as the language to structure experiences; rather it is the possibility to find redundancy, or order, in the world that is the core of intelligence.
Approaching the term “intelligence” from the non-scientific origin reveals additional aspects. The term “intelligence” existed in common language long before it was used in psychological science. Spiker and McCandless (1954) point out that there are at least two interesting aspects to the common sense meaning of intelligence. First individuals are characterized as intelligent, when they are generally bright or bright in special fields. The second aspect contains the meaning of a trait versus a state. Intelligence is supposed to be a characteristic that does not change easily or rapidly.
Hence, a great deal of work has been put into showing, that intelligence scores derived from intelligence tests do not differ between times of testing (retest reliability). Schaie, Willis, Jay and Chipuer (1989), however, expanded this idea of stability beyond a simple comparison of test scores. In their investigation, they wanted to show the structural invariance among different ages of subjects. They argue that a stable test should also show the same factorial structure, independent of the age of the tested subjects. In a longitudinal study running from 29 to 90 years, Schaie et al. found that it is valid “to track the same basic construct across age and cohorts in adulthood” (p. 660). However, marker variables for intelligent behavior change over a life-span.
2.1. Intelligence as a unitary concept?
Aunitary construct of intelligence assumes that there is some kind of an underlying ability that does not dependent on the different content of items or on different cognitive processes performed on these items. The first to argue for such an underlying ability was Spearman, who analyzed a correlation matrix derived from relationships among a battery of measures of intelligence (Brody, 1999). The positive correlation among all measures was called “positive manifold”, a term introduced by Spearman in 1904. From the positive manifold, Spearman concluded that there, indeed, was an underlying element that he labeled g. Gottfredson (1996) defined g as “the ability to deal with complexity, regardless of the specific content to be mentally manipulated” (p. 20).
While Spearman’s view prevails today, the history of psychology also experienced different approaches. Thurstone for instance tried to substitute the general factor of intelligence by introducing several independent abilities, which should be unrelated to each other.
While some papers (e.g. Neisser et al., 1996) depict these opposing views on g, they do not mention the fact, that Thurstone himself abandoned the idea of independent abilities. Brody (1999) shows the rapprochement between Thurstone’s and Spearman’s view, when Thurstone investigated the correlational structure among the seven abilities.
After the controversy between Spearman and Thurstone was settled, other scientists such as Guilford still did not acknowledge the existence of g. While Humphreys (1962) only vaguely criticized Guilford, Brody (1999) clearly stated that empirical data found this far does not support Guilford at all. Since Thurstone had abandoned his theory of independent primary factors and Guilford’s model fell because of lacking empirical support, Spearman’s g -factor model attracted attention again. However, scientists took a modified look on Spearman’s idea. Humphreys already points out that a hierarchical model, proposed for instance by Vernon (1950), would be able to solve the puzzle. Such a hierarchical model places Spearman’s g -factor on the top of the model. Hence, g would be the one element influencing all other elements. The place of g, however, seems to be the only certainty scientists can offer so far. The general factor of intelligence is found either when correlating many different measures of cognitive ability or when primary factors, such as those proposed by Thurstone, are analyzed again (Eysenck, 1986).
At the level below g, Horn (1976) recognizes three broad factors that are found consistently and are independent enough not to collapse into one factor. These factors are verbal productive thinking, crystallized intelligence and fluid intelligence. The latter two factors Gc and Gf were introduced by Cattell (Brody, 1999) to split up g into two components. While crystallized intelligence is conceptualized as cultural component, particularly learning experiences, fluid intelligence represents a biological component of intelligence.
In accordance with Horn (1976), Carroll (1993) places Gc and Gf at the level below g along with other factors. Neisser et al. (1996) mention Carroll’s hierarchical model favorable and Brody (1999) writes:
The vexed history of attempts to derive a taxonomic model of intelligence based on an analysis of performance on diverse tests of intellect is now close to definitive resolution in Carroll’s three stratum theory. (p. 21)
Daniel (1997) also notes in his survey on contemporary measures of intelligence that g is accepted conceptually, while the assessment procedures clearly emphasize the level below g.
On the psychometric model established by Carroll, Daniel writes:
Never before has a psychometric-ability model been so firmly grounded in data. This [Carroll’s] multifactor model provides a common frame of reference for test analysis and interpretation and for many avenues of research. It is an invaluable tool for establishing the construct validity of scores on psychometric-ability tests and for shedding light on the meaning of scores derived from other types of tests. (p. 1043)
2.2. Carroll’s Three Stratum Theory of Intelligence
Carroll (1993) derived a model of intelligence based on the analysis of approximately 460 data sets, including a total of 130,000 subjects. This meta-analysis, unique in its sampling, included factor analytic studies from 1925 until 1987.
While Carroll (1993) investigated separately the structure of all the very different domains of intelligence, the interest for this study lies on the development of his three stratum theory of intelligence and the place of cognitive speed in this theory. The question how EI, which was not then introduced in psychological science would fit into this theory, cannot be answered on an empirical basis since the factor analytic studies used by Carroll did not operationalize emotional aspects. Consequently, EI factors could not be found. The main ability areas for the structure of intelligence were the domains of language, reasoning, memory and learning, visual perception, auditory reception, idea production, cognitive speed, knowledge and achievement, psychomotor abilities and two more categories called miscellaneous domains of ability and personal characteristics and higher-order factors of cognitive ability.
For the present study the domain of cognitive speed is most interesting and elaborated in the following paragraphs. The main ability areas above refer to the broad abilities included in Carroll’s theory just below the general factor of intelligence. The interested reader, hence, should refer to the corresponding chapters in Carroll’s book.
As for the cognitive speed factor it is important to understand that cognitive speed, according to Carroll, does not seem to be a single unitary concept. The existence of a domain of cognitive speed already suggests, that the factorial concept of speed is complex. Indeed Carroll at one point discusses the possibility that the general factor of intelligence, a second-order factor in many factorial analyses of psychometric intelligence test batteries, might derive from the speed element in all of the first order factors. Since many tests employ some form of speed limits, the correlation among these factors might result in a second-order factors that reflects exactly this speed component.
Then, however, the general factor would not reflect a general thinking ability, at least not solely. He concluded, however, that the speed factor itself could not be the general factor of intelligence itself, rather cognitive speed seemed to be the dominant factor for a variety of tasks in each domain, while the level of ability was the more dominant factor in a variety of other tasks. Hence, it can be stated that tasks differ in their degree to reflect speed or level of performance measures. Accordingly, a general cognitive speed factor is better positioned on the level below g.
From the area of elementary cognitive processes Carroll (1993) tentatively derived a multitude of first order speed factors such as simple reaction time (RT) tasks, choice RT tasks, movement time, semantic processing speed, visual and/or memory search (slope parameters), visual and/or memory search (RT and/or intercept parameters), and speed of mental comparison. Additionally he found a second order cognitive speed factor.
Carroll’s three stratum theory should not be depicted in form of a tree structure, in which certain factors dominate lower-order abilities. Nonetheless, the three stratum theory is a hierarchical model rather than a taxonomic model. Particularly it is important to keep in mind that specialized factors have loadings on more than just one higher-order factor and that the three strata are not perceived as rigid levels of abilities but rather as probable positions, between which other strata might exist. Hence, it is sometimes difficult and inappropriate to assign a certain factor to one stratum.
2.3. The Berlin Structural Model of Intelligence
While Carroll’s three stratum theory reflects an international converging picture of intelligence, there is yet to show one piece of work done by a group of scientists around Adolf Jäger. Building on a structural model of intelligence (called: “Berliner Intelligenzstrukturmodell”, BIS), they developed a highly complex measure of intelligence capable of distinguishing very different aspects of intelligence, including productive thinking.
Three basic assumptions underlie the BIS (Jäger, Süß, & Beauducel, 1997). First, any performance that uses intelligence draws from every cognitive ability, though with very different emphasis for different tasks. The different aspects can be traced back by the variance in any performance. Second, performance and the constructs established to reflect performances can be classified. In the case of the BIS a bimodal classification (content and operations) was used. Third, ability constructs are structured hierarchically, and hence are differently general or broad.
The classification into content and operations allows to distinguish the following four operations (1) processing speed, (2) memory, (3) creativity, and (4) processing capacity and the three content based components (1) verbal, (2) numerical, and (3) figural (Jäger, 1984).
The BISM was derived from factor analysis of about 2,000 different tasks collected by Jäger and others since 1967. FA of the final 191 items distinguished readily four operational factors (those enlisted above). In each of these factors content was mixed. These content based factors were found, however, when each operational factor was analyzed separately. Finally, g was established as integral over all factors.
Hence, the BISM essentially is a descriptive model. Jäger et al. (1997) do not argue for a certain theory the model evolved from, but rather used empirical data and identified seven factors. These factors themselves are, however, theoretically well established.
Although the BISM is a thoroughly researched model, its corresponding testing instruments are available only in German so far.
2.4. Problem Solving Ability and Learning Test Concept
Finally two more views on intelligent behavior include the issue of problem solving ability and learning test concepts. Intelligent people are supposed to solve problems more easily than less intelligent people. This holds true, as long as the measure for problem solving capacity is something like school grades (Dörner & Kreuzig, 1983), while there generally exist only low correlations with problem solving in real world situations (Funke, 1983). Like Sternberg, Dörner and Kreuzig argue that real world problems need divergent thinking (creativity) rather than convergent cognitive processes. Intelligence should describe the main ingredient for cognitive performance (Dörner, 1984) and should be more meaningful the more complex a situation is. The measurement of complex cognitive performance (e.g. the “tailor shop” problems), however, disappointed (Funke) and the conclusions derived from this research line seems open to critics (Tent, 1984).
Guthke (1982) summarizes the learning test concept (introduced by Vygotski (1934) and Kern (1930)). The main issue is the difference to the state-orientation of the traditional testing procedures. Intelligence should not be conceptualized in terms of a one-time score but in terms of learning from experience. Hence, the identification of an individual’s ability to learn is the target of the learning test concept.
Vygotski’s model was often cited as an opposition to psychometric testing procedures (Guthke, 1982), though Vygkotski did not reject the psychometric approach. Rather they used traditional test items or tests and added guidance, e.g. feedback about the correctness of solutions, about the effectiveness of strategies used and others. After such a training session the test items were usually presented again and the scores of the individuals recorded. The idea, of course, was that the training session between the two times of testing produced an effect that would change the rank order of the individuals from the first time because of different learning abilities in subjects. Difficulties in measurement, however, impeded also the success of this concept (Guthke).
More recently, Daniel (1997) describes two different groups of dynamic assessment techniques. The one group is characterized by the use of nonstandardized clinical intervention by the examiner. Usually, the interest of this kind of examination is the individual ability to learn, his or her ability to be more or less successful with different kind of material or with different instructions. Due to the lack of standardization, traditional reliability scores and other comparisons basing on the idea of stable assessments are not meaningful for this group of clinicians. The second group is concentrating on exactly these problems. The first standardized and normed instrument available in this line of research is the New Swanson Cognitive Processing Test, which yields scores for working-memory task performance before and at various other points during the procedures.
3. Intelligence and Mental speed
When Eysenck (1986) wrote his survey on intelligence and the future directions of research in intelligence, he operated with the model depicted in figure 1.
illustration not visible in this excerpt
Figure 1. From Eysenck; Intelligence: the new look (p. 334)
Intelligence is divided in three kinds of intelligence: biological intelligence, psychometric intelligence and social intelligence. He did not, however, try to separate these different aspects in the form of three different intelligence concepts, rather he perceived biological intelligence as basis for psychometric intelligence, which itself is a basis for social intelligence.
Measures of RT lie at the very core of biological intelligence. Eysenck also argues that power and speed tests indeed measure much the same kind of intelligence. Essentially, these developments bear the meaning that, no matter what kind of task is used, the important variable is the time to answer the task. Intelligence is, thus, comparable to a biological notion of the speed of information processing and the speed of neuronal conduction (Sternberg, 1998).
Baumeister (1998) describes a scientific beginning of the notion that individuals differ in their RTs towards observational stimuli. The so-called personal equation states the problem in scientific observation that different individuals have different time lags between observation and recording of a stimulus.
While RT measures were already used at the turn of the last century, the arrival of psychometric tests and broad changes in psychological science led to the neglect of RT measures. Particularly a study conducted by Wissler showing no substantial correlations between mental tests (RT measures) and students’ grades stopped interest in the RT approach (Neubauer, 1993). It was only in the 40s, when interest in RT measures was evoked again by demonstrating that retarded individuals show remarkable slow RTs compared with a control group (Baumeister). Such findings led to a rediscovery of the connection between RT measures and intelligence resulting in numerous studies, which allow the present statement that indeed “one of the most firmly established correlates of intelligence is RT” (Baumeister, p. 258).
3.1. Studies on the Connection between RT and Intelligence
Just recently Schweizer and Moosbrugger (1999) evaluated speed of information processing in comparison with attention and intelligence as a basic parameter of mental capacity. Using two different intelligence tests (LPS and HAWIE) they found substantial correlations between intelligence and RT (between -.41 and -.51). Deary, Der, and Ford (2001) found a correlation between intelligence and simple choice RT of -.31 and between intelligence and four-choice RT of -.49 in a big community sample (900 individuals aged 56 years). Such studies, however, shed only a crude light on the complex association between speed of information processing and intelligence.
While there are numerous studies investigating this association between speed of mental information processing (e.g. via RT) and intelligence, only a small overview on such studies is given in this chapter. First, however, the basic research paradigms used in most of these studies are described.
3.1.1. The Hick-Paradigm
The apparatus used in this paradigm consists of basically three elements, the stimulus presentation device, target keys and a home key. The stimulus presentation device originally consisted of several lights that were turned on and off during the trials. A different number of lights could be used to manipulate complexity. Two kinds of times are measured with this instrument, the time from onset of stimulus presentation until the individual releases the home key, and the time from the moment the home key is released until the target key is touched. The first measure is a measure of the RT, the second one a measure of movement time. Usually the RT is the measure of choice in this paradigm and is used to calculate the average RT and the standard deviation of the RT.
A different parameter described just recently is the shape of the RT distributions. Juhel’s (1993) notion to use the RT distribution of one person as parameter builds on the fact that RT distributions are manifestly positive skewed and, hence, the use of traditional statistics is not always suitable (i.e. when the distribution is highly skewed and only a small number of items is used). Juhel suggests to use either higher order parameters of RT distributions (e.g. skewness or kurtosis) to characterize an individual’s distribution or the use of explicit models to characterize individual distributions. While the overall correlations (over all intelligence tests used) between mean RT, standard deviation of RTs and intelligence dropped to -.201 for mean RTs and -.332 for standard deviation of RTs, the skewness was associated -.396 with intelligence.
3.1.2. Sternberg – Short Term Memory Scan Paradigm.
This paradigm was invented by Saul Sternberg to test the short-term memory. The basic assumption is that the time needed to determine, whether an element is a member of the short term memory or not, is a function of the number of elements in the short term memory. The usual way to test this assumption is to offer a different number of elements (e.g. numbers or letters), which the individual has to keep in memory. After a short break a target element is offered and the individual is asked, whether this target element was part of the set of elements. In the speed of information processing paradigm, however, the interest is the individual difference in time needed to answer the question for any given number of elements.
3.1.3. The Posner-Paradigm.
The Posner Paradigm is used to assess the speed of long-term memory retrieval. In each trial two letters are presented to the individuals tested. These trials differ in the way that letters (1) either are physically and semantically identical (‘AA’), (2) are physically different but semantically identical (‘Aa’) or (3) are either physically and semantically different (‘AB’). Additionally, the response requirements are manipulated. Either the individual has to answer whether the two letters have the same shape or the same meaning. To answer the latter question only visual discrimination is needed, while the answer for the first question needs long-term memory retrieval. Subtracting the response times, thus, indicates the time needed for long-term memory retrieval only.
3.1.4. Odd Man Out Paradigm.
In this paradigm the apparatus is somewhat similar to the Hick paradigm, but the sequence of lights is different. Usually, there is a number of lights going on in any trial with all but one of these lights being close together and one light going on fairly distant from the others. This distant light is the odd man. The individual then has to release the home key and touch the target key below the odd man. This paradigm offers somewhat more complex tasks, without including higher cognitive processes.
3.2. Starting Point
Although Roth (1964) must be credited for finding that measures of slope DT are related to intelligence, Nettelbeck (1998) suggests Jensen’s chronometric research as a starting point. Indeed Jensen’s interest in RT measures and their association with intelligence is a still ongoing project starting in 1979, when Jensen published his first RT study (Vernon, 1998). Within the next years Jensen collected data on the suggested relation between general intelligence and speed of processing that amounted to some 2,000 subjects. In these studies Jensen built on the Galtonian notion that intelligence could be properly conceptualized using mental speed tests.
Jensen also interpreted the causal direction in favor of RT. In other words, he argued that the cause for higher intelligence was a quicker way to process information. Nettelbeck phrased Jensen’s viewpoint as follows:
If all intelligent activities are dependent on the resources of an information processor with limited capacity to cope with input at any point in time before reliance on long-term retention, then more rapid encoding and storage must bestow advantage; a faster processor should be capable of more accurate mental activity and therefore greater knowledge acquisition per unit time. (p. 235)
One of the outstanding points in Jensen’s research was the commitment to a single research paradigm resembling the Hick paradigm. Jensen used other tasks (e.g. the Sternberg paradigm) in parts of his work, but referred back to his research paradigm in almost every study. Eight lights, ordered in a semicircle with the same distance to the home key, formed Jensen’s research apparatus. The task of the individual was to press the lit signal, choices of reaction were introduced by either using just one light, or using two, four or eight lights. RT (between stimulus onset and release of the home key) as well as movement time (between release of home key and press of target key) were recorded. With the collected data set Jensen could finally reach a somewhat definite estimation of the relation between RT, as measured in his paradigm, and intelligence, most often conceptualized by Raven’s Matrices. This estimation is -.201, rising to -.309 (Vernon, 1998) when corrected for attenuation due to unreliability.
Beauducel and Brocke (1993) also used the Hick paradigm to assess the correlation between RT parameters and psychometric intelligence. The psychometric intelligence tests used were Raven’s APM and the BIS, the RT measure consisted of two sequences of 15 trials each per bit level (0,1,2, and 3).
Beauducel and Brocke could confirm only some of the results predicted from earlier work in this area. The intercept (of the line of regression over the RT means per bit level) correlated with general intelligence (of both intelligence measures) significantly. The standard deviation, which was often used in Studies by Jensen and others, correlated significantly only at bit-level 0 with general intelligence from Raven’s APM but at bit-levels 0,1, and 2 with general intelligence from the BIS.
Beauducel and Brocke also correlated RT measures with subcomponents of the BIS. They found significant correlations particularly for the standard deviation and the mean RTs. Processing speed of the BIS was the component most significantly and most often correlating with RT measures, while creativity did have almost no significant correlation. Other components showed a substantial number of significant correlations, particularly processing capacity and numbers, or at least some significant correlations, such as verbal and figural content components or memory.
The study of Beauducel and Brocke shows clearly that RT measures substantially correlate with measures of intelligence. They do so, however, differently for different measures of RT and different measures of intelligence. Broadly speaking, general intelligence seems to be a measure quite well correlating with RT measures (particularly the mean RT and also the standard deviation to a smaller degree). Higher correlations, however, can be found with certain subcomponents of intelligence. Moreover, correlations do not appear at all complexity levels of RT paradigms in a uniform way. Rather there seems to be a certain complexity level up to which correlations are raised. Beyond this level, however, the correlations between intelligence and RT measures tend to suffer a decrease in value.
3.3. A Developmental Perspective
Anew issue in the study of RT measures is a developmental perspective. Hence, like interindividual differences in intelligence scores are only somewhat stable during the aging process, also RT parameters are not stable over life-time. Nettelbeck (1998) concludes from the literature that RT improves until about the age of 11 to 14 years, after which RTs remain mainly stable.
Agrawal and Kumar (1993) correlated age with RT and intelligence. The subjects’ age ranged from 20 years to 80 years, and analyses were divided according to sexes. In men and women they found correlations between age and intelligence and age and RT measures with intelligence scores decreasing with increasing age and RTs increasing with age. The authors do not report on a differential effect of age on the correlation between intelligence and RT.
Lavergne and Vigneau (1997) investigated the issue of changing RTs and differences in the correlation between RT and intelligence measures in a cross-sectional study conducted in Canada on 111 individuals from various primary schools. They grouped children in three age groups (9,10, or 11 years) and assessed verbal, nonverbal and global IQ scores from the Wechsler Intelligence Scale for Children (WISC-III), scores from a French-Canadian test battery of mental abilities and the response times for the easiest items (solved by more than 90%) on this test battery.
Regression analyses revealed that response times predicted intellectual performance of the 11 year olds quite well, while they were only a mediocre predictor for performance at age 10 and an even worse predictor at age 9.
Lavergne and Vigneau explain these results with the utilization of strategies used at a lower age level compared to higher age. Essentially, they argue that the items, though easy for all pupils, impose a much heavier load on the younger children (although they solve them correctly). Hence, for variance within this age cognitive strategies are responsible. Children vary according to their use of more or less adaptive strategies on the items. At the older age, all children face items so simple that these items do not impose a substantial load on their cognitive abilities. Hence, the use of strategies is less pronounced and the time scores are more pure measures of mental speed. In other words, at age nine more noise (in form of cognitive strategies) reduces the power of the time score measures.
3.4. Cultural Invariance
The relation between intelligence and RT also holds true, when regarding individuals from very different backgrounds. As Ja-Song and Lynn (1992) hypothesized, correlations between decision time in RT experiments and intelligence were significant in a sample of 299 Korean children aged between 9 and 10 years. General intelligence, once again, was measured with Raven’s matrices (Standard Progressive Matrices). Generally, standard deviation measures of choice RT yielded higher correlations with intelligence than means of RTs. The odd-man-out paradigm was most highly associated with intelligence (-.22) compared with simple RTs (-.10) and choice RT (-.09). All of these correlations, except the one between choice RT and intelligence, were statistically significant.
3.5. Uniformity of RT Measures
Neubauer and Knorr (1997) tried to separate empirically four different aspects of RT measures: stimulus perception, stimulus discrimination, response choice and motoric response. For this purpose they constructed two different tasks, one of which used stimulus perception, stimulus discrimination and motoric response only. No response choice was included in this task. The second task additionally required a choice reaction. For the first task, correlations with intelligence remained low, while the correlations for the second task were significant on all levels of complexity and increased with level of complexity. The authors conclude that choice reaction produces the correlation with intelligence.
An issue of high importance to the development of any theory about the relation between RT measures and intelligence is the problem of stability of results. Several studies (e.g. Beauducel & Brocke, 1993; Neubauer & Knorr, 1997) showed significant correlations of comparable magnitude between reation time measures and intelligence.
Schweitzer (1991), however, focused on the issue of stability and tried to replicate a theoretical model for the association of RT and intelligence in two different samples. Comparable to the work of Neubauer and Knorr (1997) he separated RT tasks in two different elementary cognitive processes – a composition of stimulus perception and comparison with a stimulus in memory (p-component) and the decision about reaction and the reaction itself (r-component). In two different samples from 1988 and 1990 four different models for the relation between intelligence and RTs were evaluated: (1) an additive model containing no interaction between the components of the RT task (r+p+p+...+p), (2) a multiplicative relation among p-components, while the r-component was added (r+p*p*…*p), (3) a model for statistical reasons (r*(p+p+…+p)), and (4) all components multiplicatively (r*p*p*…*p).
While the additive statistical model (1) indicates serial information processing, the multiplicative models assume parallel processing, making it virtually impossible to clearly separate the impact of different task components.
A first difference was apparent from the different values of the correlations between RT and intelligence scores. While the highest correlation in the sample from 1988 was -.372, the highest correlation in the sample from 1990 was -.592. As for the models compared, the additive model (1) fit the data of the 1988 sample well. For the 1990 sample, model (1), however, did not fit. All other models showed a better fit for the 1990 data. Particularly model (4), the one with multiplicative relations among components, fit this sample’s data most adequately.
Schweitzer (1991), thus, concluded that the stability of the results obtained in his study is rather disappointing, although a model proposing multiplicative relations among components of RT measures showed greater stability compared to using the correlations between RT and intelligence itself. From the stability of model 4, however, parallel processing of RT components must be taken as a serious option in formulating theories on the relation between intelligence and RT.
3.6. Do Different MS Paradigms yield Different Correlational Patterns with Intelligence?
Instead of comparing the contribution of different elements of RT measures Neubauer, Riemann, Mayer and Angleitner (1997) assessed the correlation between different elementary cognitive tasks such as the Hick, Sternberg and Posner paradigm and general intelligence (Raven’s APM).
Comparing the three paradigms, the authors found that the correlation between general intelligence and the Posner paradigm (r’s from -.16 to -.41) was the highest, followed by the Hick paradigm (r’s from -.14 to -.32). The Sternberg paradigm showed the lowest correlations with general intelligence (r’s from -.15 to -.25). The second inference concerns the parameter most acceptable in any paradigm. Besides mean RTs and standard deviation, the slope and the intercept were calculated. However, the highest correlations were obtained with the mean RTs and the standard deviations.
Luo and Petrill (1999) focused on the issue of g and the possibility to substitute psychometric measures for g with elementary cognitive tests. A wide variety of ECTs (learning, probe memory, simple RT, stimulus discrimination, self-paced learning, and inspection time (IT); all tests were take from the Cognitive Assessment Tasks battery by Detterman) were used to predict school performance in 568 pupils attending grades one to six. In factor analysis elementary cognitive tasks formed a set of factors that were best described with one general factor of ECT’s, a factor of learning and memory ECTs, a factor of response time ECTs and a factor of IT tasks. The path analyses with two different estimations of g (one from the WISC-R, the other from the Specific Cognitive Abilities battery, SCA) showed strong connections between ECT factors and general intelligence. The general ECT factor and the learning and memory ECT factor were strongly related with the Wechsler and the SCA general intelligence scores, while the other ECT factor scores showed no substantial correlation with psychometric general intelligence.
When predicting scholastic achievement from different models mixing ECTs, SCA and WISC-R scores, models containing ECTs showed a somewhat stronger predictive power. The increase in variance explained, however, was by no means substantial. Unfortunately, Luo and Petrill did not compare the predictive power of ECTs alone with the predictive power of psychometric g measures.
3.7. Information Psychology and the MS-IQ Correlation
Lehrl and Fischer (1988) from the Erlangen school of information psychology, tried to explain associations between intelligence and RT measures on the basis of information psychology. Information psychology is the search for parameters to describe, explain and predict human information processing. The basic framework, in which information psychology operates, is the model of psychostructure depicted in figure 2.
 Implicitly one has to acknowledge that intelligent behavior, as characterized by the individual possibility to find order and redundancy, is only possible in a world that indeed shows some redundancy.
 R.D. Roberts (2003); http://www.psych.usyd.edu.au/difference5/papers/ects/introduction.html
 The Raven Matrices are particularly well researched in intercultural contexts. Rushton, Skuy, and Fridjhon (2003) for instance found that the Raven’s APM scores had the same construct validity for African, Indian and White students.