Is speed of information processing related to fluid or to crystallized intelligence?

Is speed of information processing related to fluid or to crystallized intelligence?

INTELLIGENCE7, 91-106 (1983) Is Speed of Information Processing Related to Fluid or to Crystallized Intelligence?* JOSEPHINE C . JENKINSON Institute...

839KB Sizes 0 Downloads 32 Views

INTELLIGENCE7, 91-106 (1983)

Is Speed of Information Processing Related to Fluid or to Crystallized Intelligence?* JOSEPHINE C . JENKINSON

Institute of Special Education Victoria College

Speed of information processing was examined in relation to the theory of fluid and crystallized intelligence. Sixty sixth grade children completed the Standard Progressive Matrices test and the Mill Hill Vocabulary scale, both untimed, as measures of fluid and crystallized intelligence, and three RT tasks: Memory Scanning, Picture Identification, and Sentence-Picture Comparison. As predicted, RT parameters, with the exception of slope on two tasks, were negatively correlated with scores on both tests. There were no consistent differences between the two tests in the size of correlations with RT, and partial correlations, eliminating the effects of fluid and crystallized intelligence respectively, failed to support any causal relationship between fluid and crystallized intelligence in either direction. Some limitations of the research, specifically in the possible effect of error rates on the speed-intelligence relationship, and in the application to individuals of task parameters derived from group performance, are noted.

INTRODUCTION Since its initial proposal in published form by Cattell (1963), the theory of fluid and crystallized intelligence has been widely accepted for its usefulness in acc o m m o d a t i n g some apparently opposing views about the nature of intelligence, particularly those c o n c e r n i n g intelligence as a single, general ability, as opposed to several separate abilities, and those concerning the relative roles of heredity and e n v i r o n m e n t in the development of intelligence. The theory is presented in its most c o m p r e h e n s i v e form by Cattell (1971), but is also outlined in several other sources (Cattell, 1963; Horn, 1966, 1968, 1972; Horn & Cattell, 1966). The theory originates from second-order factor-analytic studies of primary mental ability factors which indicated the existence of more than one general fac-

*This research was carried out in the Faculty of Education, Monash University, in fulfillment of the requirement for the degree of Master of Education. Correspondence and requests for reprints should be addressed to the author at Institute of Special Education, Victoria College, Burwood Campus, 221 Burwood Highway, Burwood, 3125 AUSTRALIA 91

92

JENKINSON

tor (Cattell, 1963). Cattell argues that abilities which involve intelligence to any degree are organized at a more general level into two principal classes or dimensions. The first, fluid intelligence, is said to be the major measurable outcome of the influence of biological factors such as heredity, basic sensory structures, and the intactness or lack thereof in the nervous system on intellectual development, and is defined most clearly by tests of reasoning which are relatively "culturefair" in content. The second, crystallized intelligence, is said to be the "principal measurable manifestation of a unitary influence of experiential, educative, and acculturational factors" (Horn & Cattell, 1966), and is defined most clearly by tests in which knowledge of the content is most likely to have been acquired through specific educational or cultural experiences, such as measures of verbal comprehension. The two dimensions are claimed to be operationally independent, although there is considerable overlap in that some primary abilities show loadings on both factors, and correlations of approximately .4 or .5 are usually found between measures of the two factors. The overlap is explained by Cattell (I 971) as resulting from the "investment" of fluid abilities in the acquisition of crystallized skills and knowledge. This investment aspect assumes that in the development of the individual there is initially a single, general, relation-perceiving ability which has a biological basis. This ability is not specific to any one sensory or motor area, or to any single memory process; hence it is termed fluid ability. As development proceeds, fluid ability is invested in all kinds of complex learning situations which are provided by cultural and educational experiences. The efficiency with which skills and knowledge are acquired will thus depend on the amount of fluid ability which is available for investment in learning, as well as on cultural opportunities. Some of the common variance in measures of crystallized intelligence will be the result of fluid intelligence having been invested in a variety of learning situations. The mechanical process by which investment takes place is through the application of basic elementary capacities to learning which Horn (1968) terms "anlage" functions. An "anlage" function: represents very elementary capacities in perception, retention, and expression, as these govern intellectual performance. For example, span of apprehension--the number of distinct elements which a person can maintain in immediate awareness--is an elementary capacity and yet one which determines, in part, the complexity with which one can successfully cope in an intellectual task. (Horn, 1968, pp. 243--4) Thus, the theory predicts that fluid intelligence is, at least in part, a causal factor in determining the level of crystallized intelligence. The reverse, however, does not hold: the level of fluid intelligence is not claimed to be affected by the level of acquisition of crystallized abilities. But the validity of "investment" as the integrating concept between fluid and crystallized intelligence is difficult to

SPEED OF INFORMATIONPROCESSING

93

establish empirically. One attempt to establish fluid intelligence as an antecedent of subsequent crystallized intelligence through cross-lagged panel correlations met with conflicting results, the relationship varying at different levels of socioeconomic status (Schmidt & Crano, 1974).

Intelligence as Efficiency of Information Processing The application of information processing concepts to the study of intelligence suggests one direction in which the nature of the relationship between fluid and crystallized intelligence might be investigated. Hunt (Hunt, 1978; Hunt, Frost, & Lunneborg, 1973; Hunt, Lunneborg & Lewis, 1975) has established a relationship between the efficiency of basic information processing mechanisms and measures of verbal ability. Efficiency, in several of Hunt's experiments, is measured by speed of response to tasks requiring the encoding and manipulation of stimulus information which is generally very familiar to all subjects, for example, alphabetic and numerical symbols. These "mechanistic" processes are thus independent of individual differences in knowledge, but are the means by which words and concepts or other linguistic symbols are retrieved and manipulated. Their efficiency determines, in part, a person's level of verbal ability. These processes correspond to the "anlage" functions which Horn (1970) argues make up fluid intelligence. Jensen (1978) proposes that the speed of basic cognitive processes is fundamental to a general factor of intelligence which corresponds to Cattell's factor of fluid intelligence, and he presents evidence that speed of information processing is related to nonverbal matrices tests which are among those that operationally define fluid intelligence. In Hunt's experiments, however, speed of information processing is related to verbal ability as measured by tests of knowledge of word meanings, syntactic rules, and semantic relations denoted by words (Hunt et al., 1975)---tests which also operationally define crystallized intelligence. Hunt's interpretation of the so-called verbal intelligence test is that it measures not just the extent of acquired knowledge, but the ability to retrieve and use meaningful information in the test situation (Hunt, 1978). If this interpretation is correct, then speed of information processing may be part of a more general underlying ability which is reflected in the immediate operation of crystallized, as well as fluid, intelligence.

The Study In previous research, the relationship between speed and intelligence has usually been investigated using a single measure of intelligence. This study set out to investigate the relationship further by administering tests of both fluid and crystallized intelligence to a single sample, and examining their relationship to speed on a variety of simple information processing tasks using partial correlations. If speed is related to crystallized intelligence only by virtue of its initial investment in the acquisition of knowledge through the operation of fluid intelligence, then

94

JENKINSON

partialling out fluid intelligence could be expected to significantly reduce the correlation between speed and crystallized intelligence. Conversely, partialling out crystallized intelligence should not reduce the correlation between speed and fluid intelligence, since crystallized intelligence has no causal role in the operation of fluid intelligence. If, on the other hand, speed is part of a more general organizational factor which is present in both fluid and crystallized intelligence, then partialling out either fluid or crystallized intelligence respectively should reduce the correlations between speed and both measures of intelligence. PROCEDURE

Subjects Group tests were initially administered to 191 sixth grade children in six schools, drawn at random from a list of primary schools in the Melbourne metropolitan area of Moorabbin. From this pool of subjects, 60 children were selected at random for administration of three information processing tasks. Children who indicated that a language other than English was spoken in the home were excluded because of the difficulty of providing a fair measure of crystallized intelligence for this group. The final sample consisted of 28 males and 32 females with a mean age of 135 months, sd = 4 months.

Intelligence Tests The Standard Progressive Matrices Test (English Version) and the Mill Hill Vocabulary Scale, Junior Form 2 (1975 item order revision) were administered in group form. These tests were selected on the basis of their development and standardization without set time limits, so that any correlation obtained between speed of information processing and test performance could not be attributed simply to speed of completion of test items. The Standard Progressive Matrices Test was originally designed as a measure of Spearman's general ability factor. To the extent that it depends on logical reasoning with nonverbal material rather than on acquired information, it can be regarded as a relatively pure measure of fluid intelligence. The Mill Hill Vocabulary Scale measures recall of acquired verbal information, with a minimum demand on intellectual reasoning ability (Raven, Court, & Raven, 1977). It is thus a relatively pure measure of crystallized intelligence. Both Synonyms and Definitions sections were given. The Definitions section was scored with the aid of the key to the Synonyms section in the parallel form, and according to the consistency of the response with the meaning given in the Concise Oxford Dictionary (Sixth Edition, 1976). Because there are no satisfactory Australian norms for either test, and the age range of the sample was relatively narrow, raw scores were used in the analysis of results.

SPEED OF INFORMATIONPROCESSING

95

Information Processing Tasks Three reaction time tasks were selected to allow for a variety of types of information processing, while maintaining the requirement that the information content should either be equally familiar to all subjects, such as simple words or sentences or highly familiar codes such as letters or digits, or it should be equally novel to all subjects (Smith, 1968). This ensures that individual differences in task performance can be attributed to differences in processing efficiency and not to differences in knowledge of the content. Each of the tasks assumes an additive model of information processing. The essential principle underlying this model is that response latency, or processing time, increases as a linear function of task complexity, where complexity is defined as the amount of stimulus information which needs to be considered in reaching a correct decision. The regression of response time on complexity yields a number of individual processing parameters. The slope parameter represents the time taken for processing each additional amount of stimulus information, independently of time taken to form a representation of the test stimulus, selection of the correct response, motor activity, and other processes, all of which are represented by the zero intercept value. The mean or median response time at each level of complexity can be assumed to represent all of the components of processing which take place between presentation of the stimulus and activation of the response, with additional time accumulating as complexity increases.

Memory Scanning. This task was devised by Sternberg (1966) to investigate speed in scanning short-term memory. Subjects memorize a short list of digits presented visually. The list, termed the memory set, is then followed by a probe item, and the subject is required to indicate whether or not the probe item was in the memory set. The list of digits is varied in content and size, but is kept within the normal capacity of short-term memory, usually not exceeding six items. According to Sternberg, the time taken to respond is a linear function of the number of digits in the memory set. Sixty trials were presented, divided evenly between two, four, and six digits, and between positive and negative sets. Digits in the memory set were presented simultaneously for 4 seconds, followed by a 4-second interval before the probe digit was presented. Picture Identification. This was devised for the present research to provide a task relying purely on the processing of visual information. Subjects were presented with a set of pictures or diagrams, ranging in size from one to eight, and a target picture presented simultaneously. The task was to indicate whether or not the target picture was in the stimulus set. The pictures were small black and white line drawings of the type used to measure Thurston's perceptual speed fac-

96

JENKINSON

tor (French, Ekstrom, & Price, 1963). Picture combinations requiring very fine discriminations or conceptual judgments were avoided. Sixty trials divided evenly between set size and between positive and negative sets were presented.

Sentence-Picture Comparison. This task was developed by Clark and Chase (1972) to measure speed in the comprehension and verification of simple sentences. The task consists of a sentence such as "Star is above plus", "Plus is not below star", followed by one of two diagrams showing a star above a plus or a plus above a star. This allows for 16 sentence-picture combinations. The subject is required to indicate whether the sentence is true or false with respect to the picture. Complexity is introduced by adding a negative and/or falsifying the sentence. A linguistic model has been developed by Carpenter and Just (1975) to explain latencies on this task according to the number of constituents in the sentence which have to be compared with the picture. Data presented by Tversky (1975) suggests that a linguistic strategy is most likely to be employed when the sentence and picture are presented simultaneously, which was the procedure employed in this study. Taking the true affirmative sentence as the base requiring K comparisons, a false affirmative sentence requires K + 1, false negative K + 4, and true negative K + 5 comparisons. Sixty-four trials divided evenly between sentence type and complexity were presented. Apparatus The three information processing tasks were presented by means of a portable two-field tachistoscope manufactured by Electronic Developments. Stimulus material was reproduced on a white card according to the manufacturer's instructions. A reaction timer scaled in milliseconds was connected to the tachistoscope. Positive responses were made by pressing a button held in the dominant hand, and negative responses by pressing a button held in the non-dominant hand. The tasks were presented individually in two sessions on successive days. Session 1 consisted of Picture Identification, followed after a 5-minute interval by Sentence-Picture Comparison. Memory Scanning was presented in Session 2. Each task was preceded by a practice session. RESULTS

Intelligence Test Results Means and standard deviations for the two intelligence tests were, respectively, 41.4 and 7.1 for Standard Progressive Matrices, and 28.8 and 5.8 for the Mill Hill Vocabulary Scale. The correlation between the two tests was .54.

SPEED OF INFORMATION PROCESSING

97

Reaction Time Tasks Reliability. Reliability coefficients were derived from analysis of variance for repeated measures, using the formula given by Winer (1962) in which reliability is estimated as: MS within people rk ~ 1

--

MS between people

Since each analysis assumed an equal number of repeated measures, errors could not be discarded and reliability was based on reaction times for both correct and incorrect responses, although in subsequent analyses errors were treated separately from correct responses. The resulting reliability coefficients are shown in Table 1. All tasks yielded reliable individual differences at each level of complexity.

Group Results. In analyzing individual results, responses in which errors occurred were discarded on the assumption that these represented processes other than those of immediate interest. An overall error rate of 10% is usually regarded as acceptable on reaction time tasks. However errors may have some significance, and although discarded from the main analysis, are considered in a separate analysis. Slope and intercept values were calculated from individual equations describing the regression of reaction times for correct responses on task complexity. Group means and standard deviations at each level of task complexity were calculated from individuals' median reaction times for correct responses. The group results shown in Table 2 generally conform to expectations, with mean reaction times for each task showing an increase with task complexity. The overall error rate for Memory Scanning was 4.25%, for Picture Identification (almost negligible) 2.7%, and for Sentence-Picture Comparison 11.25%. Clearly

TABLE 1 Reliability of Reaction Time Tasks at Each Level of Complexity (All Responses) Complexity Level

Memory Scanning

Picture Identification

Sentence-Picture Comparison

1

*

.90

2 4 5

.94 .95 * .94 *

.92 .92 * .92 .92

.89 .85 * .83 .84 *

6

8

*No trials were presented at this level.

JENKINSON

98

TABLE 2 Reaction Times (Correct Responses) and Error Rates Errors as

Errors as

Y(

SD

% of all

% of total

(msec)

(msec)

responses

errors

Memory

Slope Intercept Memory Set Size

2 4 6

53.86 912.60 908.65 1055.38 1110.08

Scanning

66.16 581.03 424.40 423.15 422.98

1.75 2.83 8.17

13.72 22.22 64.05

1.84 1.25 2.22 3.19 4.58

14.73 9.47 16.84 24.21 34.74

Picture Identification

Slope Intercept Size of Stimulus Set

1 2 4 6 8

200.12 1315.46 1369.52 1681.22 1924.92 2521.22 2669.43

94.13 365.46 381.81 451.97 556.55 764.22 842.85

Sentence-Picture Comparison Slope Intercept Sentence Complexity"

1

308.44 2762.70 2842.70

162.01 860.66 765.04

2.64

5.80

2 5 6

3190.90 4109.70 4136.38

745.42 930.97 936.32

3.47 17.50 21.39

7.70 38.89 47.53

al = true a_ffwmative 2 = false affirmative 5 = false negative 6 = true negative

for this sample, the presence of a negative in the sentence was a problem. In fact on all tasks, but particularly on Memory Scanning and Sentence-Picture Comparison, the proportion of error responses increased with task complexity, suggesting that errors do not simply occur at random, despite the apparent simplicity o f task content, and that their relationship to both speed and intelligence should therefore be investigated. The implications of variations in both error rates and the fit o f individual data to predicted task models are discussed further below.

Correlations Between Reaction Time and Intelligence Correlations between reaction times for the three information processing tasks and scores on the Standard Progressive Matrices (SPM) and Mill Hill Vocabulary Scales (MHV) are shown in Table 3. Correlations at each level of task complexity are based on individuals' median reaction times for correct responses. The negative correlations indicate shorter reaction times associated with higher intelligence.

SPEED OF INFORMATION PROCESSING

99

TABLE 3 Correlations and Partial Correlations between Reaction Times (Correct Responses) and Intelligence Test Scores Partial Correlations

Reaction Time

SPM

MHV

RT/SPM partialling out MHV

RT/MHV partialling out SPM

Memory Scanning Slope Intercept Memory Set S~e

2 4 6

.06 -.33** -.30** -.43*** -.34**

.15 -.37** -.36** -.32** -.30**

-.03 -.17 -.13 -.32** -.21

.14 -.25* -.27* -.12 -.16

-.24* -.18 -.18 -.16 -.23* -.22* -.25*

-.15 -.17 -.17 -.17 -.20 -.17 -.22*

Picture Identification Slope Intercept Size of Stimulus Set

1 2 4 6 8

-.37** -.31"* -.31"* -.28* -.38*** -.35** -.40***

-.32** -.30** -.30** -.29** -.36** -.32** -.39***

Sentence-PictureComparison Slope Intercept Sentence Complexity

-.23* -.23*

-.17 -.33**

-.17" -.13'

-.05 -.22*

I

-.34**

-.34**

-.20

-.20

2 5

-.29** -.34"*

-.31"* -.28"

-.15 -.24"

-.19 -. 12

6

-.33**

-.31"*

-.20

-.17

*p < .05 **p < .01 ***p < .001

Correlations are comparable to those obtained in previous r e s e a r c h - - a r o u n d the .3 and .4 r a n g e - - a n d are relatively consistent across task complexity. However, slope for Memory Scanning was not related to either measure of intelligence. In previous studies, a relationship between slope on this task and measures of intelligence has been found only with extreme groups (Hunt, 1978), or has been confounded by sex differences in the direction of the correlation coefficient (Chiang and Atkinson 1976). Sex differences were not investigated in this study. However, correlations could have been affected by the validity of slope and intercept parameters as measures of processing speed, which assumed that individual as well as group data fitted a linear model. In Sentence-Picture Comparison the correlation with slope only just reaches statistical significance for

100

JENKINSON

fluid intelligence, and is not significant for crystallized intelligence . In Picture Identification the relationship appears to be present in all stages. The major point of interest, however, is any difference which might exist in the relationship between reaction time and fluid and crystallized intelligence. None of the three tasks show any consistent tendency for the correlations between reaction time and fluid intelligence to be greater than those between reaction time and crystallized intelligence. Any discrepancies which do occur are small and are not consistent across task complexity. Partial correlations are also shown in Table 3. The partial correlations are to some extent limited by the relatively small correlations obtained initially between reaction time and intelligence measures, but in most cases they drop to a level that is statistically nonsignificant when either fluid or crystallized intelligence is partialled out, and, where they remain significant, no consistent trend is indicated. Considered in terms of common variance, speed of information processing accounts for only a negligible part of the performance on either fluid or crystallized intelligence when the overlap between the two abilities is eliminated.

Errors

Correlations between number of errors and reaction times for correct responses at each level of task complexity are shown in Table 4. The same table presents correlations between errors and intelligence test scores. Error rates on the Picture Identification task were too low to produce any relationships of significance either with reaction time for correct responses or with intelligence. On the other two tasks, however, errors are positively correlated with median reaction times for correct responses; that is, subjects who were faster at producing correct responses also tended to make fewer errors. Thus slower, rather than faster, subjects showed a greater tendency to sacrifice accuracy and presumably resorted to guessing on the basis of inadequate processing of information, possibly in an attempt to keep response times within what they regarded as acceptable limits. In relation to intelligence, there is also an interesting pattern: errors are related to intelligence only at more complex levels of processing. The correlations will be affected by the relatively low error rates at less complex levels, but this also suggests that errors result from attempts by subjects of lower intelligence to cope with the greater demands on processing time associated with more complex stimulus information. On the Memory Scanning task, errors are related more consistently to crystallized intelligence than to fluid intelligence as the size of the memory set is increased. Linear Fit

An important assumption underlying the analysis of results, in particular the use of slope and intercept parameters, was that models of information processing, based on group performance, provided a valid description of the perform-

SPEED OF INFORMATION PROCESSING

l 01

TABLE 4 Correlations Between Errors and Reaction Times (Correct Responses) and Between Errors and Intelligence Number of Errors

Median RT

SPM

MHV

-.08 -.24* -. 16

-.11 -.32** -.38***

Memory Scanning Memory Set Size

2 4 6

.32** .49*** .51"** Picture Identification

Size of Stimulus Set

1

-.01

2 4 6 8

.06 .01 .01 .10

.04 .14 .09 .20 .11

.06 .06 .13 .05 .12

Sentence-Picture Comparison

Sentence Complexity

1

.30**

2 5 6

.19 .27* .34**

-.09 .08 -.28* -.37**

-.05 -.01 -.32** -.34*

*p < .05 **p < .01 ***p < .001

ance o f individuals, and that for each subject, as well as for the group as a whole, reaction time increased as a direct linear function o f stimulus complexity. The problem of non-linearity applied particularly to Memory Scanning, in which individual scatter-grams showed widely varying results. Individual estimates of goodness of fit to the linear equation were obtained by squaring the correlation coefficient associated with each subject's regression equation to yield a coefficient of determination (Lewis, 1960). This estimate does not indicate whether a linear equation necessarily provides the best fit to the data, or whether the data would be better fitted by an alternative function which takes higher order components into account. However, since this investigation was limited to a test o f the assumption that slope and intercept values derived from a linear equation were valid as individual parameters, a test of alternative functions was not considered necessary. For Picture Identification, individual coefficients for determination were on the whole satisfactory--the proportion of variance in reaction time which could be attributed to task complexity fell below 30% for only four subjects. There was more variation on the Sentence-Picture Comparison task, although all subjects were affected by the presence of a negative in the sentence. The major problem was in M e m o r y Scanning. Individual coefficients of determination ranged from .0 to .4, amounting to .3 or higher for only 3 of the 60 subjects. Several subjects

102

JENKINSON

TABLE 5 Intercorrelations Betwee n Slope and Intercept Parameters for Reaction Time Tasks Memory Scanning Int.

Picture Identification Slope

Sentence-Picture Comparison

Int.

Slope

Int.

.07 .55***

-.51"** .52***

.25* .33**

.27 .24*

.37"* .54***

M e m o r y Scanning Slope Intercept

-. 13

-.03 .58***

Picture Identification Slope Intercept

.45"** Sentence-Picture Comparison

Slope

-. 15

*p < .05 **p < .01 ***p < .001

failed to show any increase in median reaction times as complexity increased; a few even had a slight negative slope. Thus for the majority of subjects, a linear model did not provide a satisfactory description of their Memory Scanning data. Difficulties raised by the use of slope and intercept parameters for individuals are further illustrated in Table 5, which gives between-and within-task correlations for these values. For Picture Identification, slope and intercept are positively correlated and can be assumed to be measuring processes which are not entirely independent. Both parameters are also positively correlated with Memory Scanning intercept and Sentence-Picture Comparison intercept and slope. For both Memory Scanning and Sentence-Picture Comparison, however, slope and intercept are not significantly correlated. This could be interpreted as support for the two parameters as measures of separate processing stages, but it also has to be considered in light of the poor fit of several subjects to the linear model discussed above. An unexpected result is the substantial negative correlation between slopes for Memory Scanning and Sentence-Picture Comparison. DISCUSSION The relationship between speed of information processing and intelligence was investigated in the context of one particular theory of intelligence--that of fluid and crystallized intelligence as two correlated, but operationally independent, abilities. Significant correlations were obtained for all tasks with both measures of intelligence, but did not clearly favor either fluid or crystallized intelligence.

SPEED OF INFORMATIONPROCESSING

103

Where differences did occur, they were too small to be significant and were not consistent between different levels of complexity on a task. The partial correlations cannot be interpreted as supporting a causal relationship between fluid and crystallized intelligence in either direction. Thus, the relationship between speed and crystallized intelligence cannot be explained solely in terms of the investment of fluid intelligence in the acquisition of verbal or other skills. Although eliminating the effect of fluid intelligence resulted in some reduction in the correlation between speed and crystallized intelligence, the converse also occurred: eliminating the effect of crystallized intelligence also resulted in some reduction in the correlations between speed and fluid intelligence. This suggests that the efficiency of basic functions, such as speed of information processing, is part of a more general factor which is common to both fluid and crystallized dimensions of intelligence. Within the theoretical framework of the theory of fluid and crystallized intelligence, processing efficiency might be regarded as important in the general organization factor which has been incorporated by Horn (1980) into his schematic representation of intellectual functions. The fact that error rates were related to both intelligence and reaction time suggests a possible limitation on the observed relationship between speed and intelligence. In simple information processing tasks, errors are usually attributed to a speed-accuracy tradeoff. However, within individuals, accurate estimation of such a tradeoff is confounded by the fact that some experimental conditions produce increases in both response time and errors (Pachella, 1974). This occurred in the present study, but in addition, errors were related to intelligence only at more complex levels of processing, This suggests that errors may have a differential effect on the relationship between reaction time and intelligence, depending on the complexity of the task. Subjects with less intelligence tend to produce a larger error rate at higher levels of complexity, thereby avoiding the much larger increases in reaction time that are predicted by their lower level of intelligence. A second limitation on the observed speed-intelligence relationship arises from the use of information processing parameters derived from models based on group performance to yield measures of individual performance. This is particularly relevant to Memory Scanning, in which several subjects failed to show any consistent increase in reaction time as task complexity increased, and thus produced results which were poorly fitted by a linear equation, and slope values which were small or even, in some cases, showed a slight negative trend. Individual slopes which are low or zero suggest the use of a strategy of parallel processing by which subjects scan items simultaneously, rather than serially, for a given target, and thus could reflect differences in the amount of practice individuals require before a task becomes automatic (Schneider & Shiffrin, 1977). Furthermore, the negative correlation between slopes for Memory Scanning and Sentence-Picture Comparison suggests a characteristic style or strategy which facilitates speed of scanning information on one task, but hinders it on the other.

104

JENKINSON

Subjects who use a parallel strategy for Memory Scanning, resulting in a low or zero slope, would appear to take longer to scan and compare the constituent elements of more complex sentences, a linguistic operation in which serial processing is presumably more efficient. Since the present study was not designed to identify strategy differences, these interpretations must be regarded as speculative. There is, however, evidence that strategies are important in information processing, and may vary according to both task and subject variables (Sternberg & Ketron, 1982). Hunt and MacLeod (1978) have identified differences in strategies in the Sentence-Picture Comparison task which might affect the relationship between speed and intelligence, and Hunt (1980) has argued for the investigation of strategies as mediators of structure in studying individual differences in cognitive processes. A major criticism of individual difference research in cognitive processes has been the lack of concern for the strategies which individuals might employ (Carroll, 1978; Snow, 1980; Sternberg & Ketron, 1982). Snow, Marshalek and Lohman (1976) note that one of the problems in attempting to relate basic cognitive processes to intelligence is that experimental parameters are derived from simple, relatively automatic tasks that rely mainly on lower-order processes, whereas traditional ability tests probably tap higher-order cognitive processes which are more closely related to "executive" functions than to simple processing functions. Choice of an appropriate processing strategy will also depend on executive functions, and thus should be taken into account in attempting to understand individual differences in intelligence, both in addition to, and in interaction with, processing speed. Unless strategy differences are taken into account, the correlations between simple information processing tasks and intelligence test scores may never reach much more than .3 or .4, a level which Sternberg (1981) notes is commonly found among a wide range of psychological variables. CONCLUSION The results of this study lend support to Humphreys' (1979) criticism that the theory of fluid and crystallized intelligence is hierarchically incomplete in not allowing for a more general, higher-order factor of ability. The theory itself, particularly the aspect of "investment" of basic cognitive functions, may need modification to allow a more prominent role for the immediate operation of these functions in the measurement of acquired verbal knowledge. Guilford's (1980) reinterpretation of Cattell's fluid and crystallized intelligence factors suggests that, at least in tasks requiring access to codes stored in long-term memory, a semantic processing factor, rather than simply semantic knowledge, may be involved. In this context, it must be acknowledged that the two tests of intelligence used in this study, while typical of those which have been distinguished in factoranalytic studies by their respective loadings on fluid and crystallized intelligence, may not necessarily be measuring these factors adequately in the present sample.

SPEED OF INFORMATION PROCESSING

105

Further research may be needed in which additional tests representing the fluid and crystallized intelligence dimensions are included so that these factors may be more clearly differentiated. A further, broader, implication is a possible incompatibility between factor-analytic theories of intelligence based on the content of test items, and theories of cognition which specify the processes required in the manipulation of the item content per se. Some limitations of research into the cognitive correlates of intelligence were noted. In particular, it cannot be assumed that measures derived from information processing models based on group data are valid for individual subjects. These measures may have a different psychological meaning for individuals because of differences in strategies employed to cope with the demands of a task. Such strategies, inferred from differences in error rates and from deviations from predicted task models, may have a moderating effect on correlations obtained between speed, or other indicators of processing efficiency, and intelligence. Future research into the cognitive correlates of intelligence will demand more complex multivariate designs, taking into account individual differences both in processing efficiency and in use of strategies.

REFERENCES Carpenter, P. A., & Just, M. A. Sentence comprehension: A psycholinguistic processing model of verification. Psychological Review, 1975, 82, 45-73. Carroll, J. B. How shall we study individual differences in cognitive abilities? Methodological and theoretical perspectives. Intelligence, 1978, 2, 87-115. Cattell, R. B. Theory of fluid and crystallized intelligence: A critical experiment. Journal of Educational Psychology, 1963, 54, 1-22. Cattell, R. B. Abilities: Their structure, growth and action. Boston, MA: Houghton Mifflin, 1971. Chiang, A., & Atkinson, R. C. Individual differences and inter-relationships among a select set of cognitive skills. Memory and Cognition, 1976, 4, 661-672. Clark, H. H., & Chase, W. G. On the process of comparing sentences against pictures. Cognitive Psychology, 1972, 3, 472-517. French, J., Ekstrom, R., & Price, L. Kit of reference tests for cognitive factors. Princeton, NJ: Educational Testing Service, 1963. Guilford, J. P. Fluid and crystallized intelligence: Two fanciful concepts. Psychological Bulletin, 1980, 88, 406-412. Horn, J. L. Integration of structural and developmental concepts in the theory of fluid and crystallized intelligence. In R. B. Cattell (Ed.), Handbook of multivariate experimentalpsychology. Chicago, IL: Rand McNally, 1966. Horn, J. L. Organization of abilities and the development of intelligence. Psychological Review, 1968, 75, 242-259. Horn, J. L. Organization of data on life-span development. In L. R. Goulet & P. B. Baltes (Eds.L Life span developmental psychology. New York: Academic Press, 1970. Horn, J. L. The structure of intellect: Primary abilities. In R. M. Dreger (Ed.), Multivariate personalitv research. Baton Rouge, LA: Claitor's Publishing Division, 1972. Horn, J. L. Concepts of intelligence in relation to learning and adult development. Intelligence, 1980, 4, 285-317.

106

JENKINSON

Hum, J. L., & Cattell, R. B. Refinement and test of the theory of fluid and crystallized intelligence. Journal of Educational Psychology, 1966, 57, 253-270. Humphreys, L. G. The construct of general intelligence. Intelligence, 1979, 3, 105-120. Hunt, E. Mechanics of verbal ability. Psychological Review, 1978, 85. 109-130. Hunt, E. Intelligence as an information processing concept. British Journal of Psychology, 1980, 71. 449-474. Hunt, E., Frost, N., & Lunneborg, C. Individual differences in cognition: A new approach to intelligence. In G. H. Bower (Ed.), The psychology of learning and motivation (Vol. 7). New York: Academic Press, 1973. Hunt, E., Lunneborg, C., & Lewis, J. What does it mean to be high verbal? Cognitive Psychology, 1975, 7, 194-227. Hunt, E., & Macleod, C. The sentence-verification paradigm: A case study of two conflicting approacbes to individual differences. Intelligence. 1978, 2, 129-144. Jensen, A. R. g: Outmoded theory, or unconquered frontier? Invited Address, Annual Convention of American Psychological Association, Toronto, 1978. Lewis, D. Quantitative methods in psychology. New York: McGraw-Hill, 1960. Pachella, R. G. The interpretation of reaction time in information processing research. In B. H. Kantowitz (Ed.), Human information processing: Tutorials in performance and cognition. Hillsdale, NJ: Lawrence Erlbanm Associates, 1974. Raven, J. C., Court, J. H., & Raven, J. Manual for Raven's progressive matrices and vocabulary

scales: Part one, Section 3: The standard progressive matrices; Part two. Section 5: The Mill Hill vocabulary scales. London: H. K. Lewis, 1977. Schmidt, F. L., & Crano, W. D. A test of the theory of fluid and crystallized intelligence in middle and low socio-economic status children: A cross-lagged panel analysis. Journal of Educational Psychology, 1974, 66, 255-261. Schneider, W., & Shiffrin, R. M. Controlled and automatic human information processing: I. Detection, search, and attention. Psychological Review, 1977, 84, I--66. Smith, E. E. Choice reaction time: An analysis of the major theoretical positions. Psychological Bulletin, 1968, 69, 77-110. Snow, R. E. Intelligence for the year 2001. Intelligence, 1980, 4, 185-199. Snow, R. E., Marshalek, B., & Lohman, D. F. Correlation of selected cognitive abilities and cognitive processing parameters: An exploratory study. (Technical Report), Stanford, CA: Stanford University School of Education, 1976. Sternberg, R. J. Nothing fails like success: The search for an intelligent paradigm for studying intelligence. Journal of Educational Psychology, 1981, 73. 142- ! 55. Steinberg, R. J., & Ketron, J. L. Selection and implementation of strategies in reasoning by analogy. Journal of Educational Psychology, 1982, 74, 399-413. Sternberg, S. High-speed scanning in human memory. Science, 1966, 153, 652-654. Tversky, B. Pictorial encoding of sentences in sentence-picture comparison. Quarterly Journal of Experimental Psychology, 1975, 27, 405-410. Winer, B. J. Statistical principles in experimental design. New York: McGraw-Hill, 1962.