Elsevier

Intelligence

Volume 40, Issue 2, March–April 2012, Pages 163-171
Intelligence

Processing speed and intelligence as predictors of school achievement: Mediation or unique contribution?

https://doi.org/10.1016/j.intell.2012.01.003Get rights and content

Abstract

The relationships between processing speed, intelligence, and school achievement were analyzed on a sample of 184 Russian 16-year-old students. Two speeded tasks required the discrimination of simple geometrical shapes and the recognition of the presented meaningless figures. Raven's Advanced Progressive Matrices and the verbal subtests of Amthauer's Intelligence Structure Test were used as intelligence scales. The teacher-assigned grades in six school subjects that were aggregated into two scales represented real-life school achievement. Latent processing speed and intelligence as individual predictors each accounted for about 18% of the variability in scholastic performance. Taken together, they explained about 28% of the variance of school achievement. Although significantly correlated, each had a unique impact on school achievement; zero-constraining each of the two paths to school achievement resulted in a significantly worsened fit of a model. A mediation effect processing speed  intelligence  school achievement was bootstrapped to obtain an estimate of its statistical significance and was found to be non-distinguishable from zero. The results are inconsistent with the causal hypothesis that states that processing speed is a predictor of real-life scholastic performance because of the impact of processing speed on higher-order cognitive ability, which in turn underlies school achievement.

Highlights

► Relationships between processing speed, intelligence and school grades were analyzed. ► Participants were 184 Russian 16-year-old students. ► Speed and intelligence were equal in strength in prediction of school achievement. ► Although significantly correlated, each had a unique impact on school achievement. ► A mediation effect speed  intelligence  school achievement was non-significant.

Introduction

Since intelligence tests were originally meant to determine children with potential difficulties in school education (Binet, 1905) and the first measurement of general intelligence included analysis of school examination scores (Spearman, 1904), the association between intelligence and scholastic performance is one of the best-established associations and is often referred to in the literature on cognitive ability. The relationship between intelligence scores and school performance that are commonly found in studies are moderate to strong (e.g., Bartels et al., 2002, Brody, 1992, Jencks, 1979, Jensen, 1998, Neisser et al., 1996). These results largely depend on the kind of indexes of school achievement that were examined and whether intelligence was analyzed at a manifest or a latent level. For instance, the observed magnitude of correlation with intelligence varies for different subjects and measures of performance. Achievement in mathematics and sciences tends to be better predicted by cognitive ability than achievement in the languages (e.g., Deary et al., 2007, Krumm et al., 2008, Lu et al., 2011), with a portion of predicted variability in such subjects as arts being the lowest (e.g., Deary et al., 2007). Another issue is the measure of school achievement used in the analysis: achievement test scores are more highly correlated with intelligence than are teacher-assigned grades, probably because the latter tend to reflect, to some extent, not only real performance, but also some of the other characteristics of the child like effort or personal traits (e.g., see Jensen, 1998). On the side of cognitive ability, intelligence modeled at a latent level generally serves as a better predictor of scholastic performance than single test scores; test-specific variance adds much less to the explanation of the variability in school performance. In other words, the variance of school achievement that is predictable by intelligence scores is mostly accounted for by g, and not by the other factors that determine the scores on the different tests (Jensen, 1998).

At the same time, a large number of studies that were published in the last decades demonstrated that g could, in turn, be predicted by a number of basic cognitive processes. Processing speed and working memory are probably the best-established candidates to explain higher-order individual differences in cognitive ability (e.g., Colom et al., 2008, Conway et al., 2002, Fry and Hale, 1996, Jensen, 1998, Kail and Salthouse, 1994). Correlations between single measures of processing speed and intelligence that are commonly reported in the literature are low to moderate; when measures of processing speed are based on response times from different speeded tasks, their correlations with intelligence approach those typically observed between psychometric tests (Grudnik and Kranzler, 2001, Jensen, 2006, Kranzler and Jensen, 1989, Sheppard and Vernon, 2008, Vernon, 1988).

Thus, the next logical step would be to relate these basic processes directly to scholastic achievement. However, studies addressing this problem are still relatively rare. Discussing this issue, Luo, Thompson, and Detterman (2003a) mentioned that the failure of early studies (e.g., Cattell & Farrand, 1890) to observe significant relationships between elementary cognitive processes and scholastic performance has influenced the field. Recent studies seem to come back to this problem; however, research interest has more often focused on the working memory construct as the explanatory factor for school achievement (e.g., Alloway, 2009, Krumm et al., 2008, Lu et al., 2011). The relationship between processing speed and scholastic performance remains much less explored, although processing speed was shown to be almost as a powerful predictor of school achievement as working memory is, in at least one study that analyzed two large datasets (Luo, Thompson, & Detterman, 2006).

A study on the relationship between processing speed and scholastic achievement in fact can address different questions. First, processing speed can be examined as a single predictor of school achievement. For example, Carlson and Jensen (1982) found that reaction time in a task designed in the Hick paradigm (Hick, 1952) and reading comprehension share about 30% of common variance. Very similar results were reported by Luo, Thompson, and Detterman (2003b). In their study, about 30% of the variance of scholastic performance was accounted for by the mental speed factor; the relationship between mental speed and school achievement was found to be invariant across different knowledge domains. Moreover, the latter study addressed another question, namely the etiology of these relationships. The covariance between mental speed and scholastic achievement was found to be mostly genetically mediated (similarly, other studies report that mental speed has a substantial genetic covariation with psychometric g (Baker et al., 1991, Rijsdijk et al., 1998) and intelligence has a mostly genetic covariation with school achievement (Kovas et al., 2007, Petrill and Thompson, 1993, Thompson et al., 1991, Wadsworth et al., 1995, Wainwright et al., 2005)).

The next question that can be addressed is the question on the comparative strength of processing speed and intelligence as possible predictors of school achievement. Luo et al. (2006) formulated a very similar problem in terms of the criterion validity of tasks of basic cognitive processes. In the analyses of two datasets, which are the Woodcock-Johnson III Cognitive Abilities and Achievement Tests normative data and the Western Reserve Twin Project data (with a total of more than 5500 participants), the authors observed zero-order correlations between latent processing speed and achievement factors, which are similar or even higher than the correlations between conventional cognitive ability and achievement factors. In their earlier study, the same authors reported very similar results of almost equal zero-order shared variance between processing speed and scholastic performance, on the one hand, and intelligence and scholastic performance, on the other hand (Luo et al., 2003a). Similarly, Rindermann and Neubauer (2000) observed a correlation between processing speed and school performance (r = .37) that was only slightly lower than a correlation between intelligence and school performance (r = .43). Results reported by Luo and Petrill (1999) also suggest that “the predictive power of g will not be compromised when g is defined using experimentally more tractable [elementary cognitive tasks] ECTs” (p. 157). However, the relative strength of processing speed as a single predictor of school achievement (as compared to intelligence) still remains doubtful, as some studies report significantly lower association between processing speed and school achievement than between intelligence and school achievement. For example, Rindermann and Neubauer (2004) reported the associations with school achievement of β = .09 and β = .53 for processing speed and intelligence, respectively. Colom, Escorial, Shih, and Privado (2007) observed only low zero-order correlations between school grades and processing speed as measured by simple short-term recognition tasks. Of nine school subjects, grades in mathematics showed highest correlations with the measures of processing speed, although even these correlations were quite low (r = −.12 and r = −.17). In the latter study, latent processing speed was not a significant predictor of academic performance, while a combined latent variable for fluid intelligence and memory span capacity accounted for about 29% of variance of school achievement.

Finally, the most intriguing issue on the relationships between processing speed, intelligence, and school achievement is their consistency with the causal mental speed hypothesis. From this point onward, certain theoretical assumptions start playing a major role, as any kind of testing of mediation effects is completely senseless in the absence of strong theoretical and methodological backgrounds. The mental speed theory provides a strong background for this kind of study (Brand, 1981, Deary, 1995, Jensen, 1982, Jensen, 2006, Jensen, 2011); it suggests that processing speed is a basic factor that underlies higher-order cognitive ability, which in turn influences one's success or failure in school. This theoretical model results in another set of questions that can be addressed through empirical studies.

The first question concerns the relationship between intelligence and school achievement, with processing speed as another explanatory variable. Indeed, as soon as processing speed not only underlies intelligence but also serves as a predictor of school achievement, could intelligence add anything else to the explanation of scholastic performance beyond processing speed? In other words, after controlling for processing speed, does the observed relationship between intelligence and school achievement still hold? The recent studies that addressed this issue reported inconsistent results. Luo et al. (2003a), who were the first to use structural equation modeling for the analysis of the unique impact of intelligence on school achievement beyond processing speed, concluded that “the observed correlation between seemingly complex g and scholastic performance is indeed mostly mediated by a set of elemental cognitive functions” (p. 81). The authors compared the proportion of variability in school achievements explained by a single intelligence factor (about 30%) to the corresponding proportion of variability after controlling for processing speed (about 6%). Based on the observed drop of the correlation, Luo, Thompson, and Detterman claimed that processing speed explained a great deal of the common variability between cognitive ability and school achievement. At the same time, it is noteworthy that the higher-order shared variability between intelligence and school achievement was still important, as reducing the corresponding correlation between the residuals in the model resulted in significant chi-square changes. In the later study by Luo et al. (2006), two other methods were used to address the same issue. First, R2 changes in a series of mathematically equivalent structural equation models with different number of predictors of school achievement were analyzed. Low R2 increments in models where cognitive ability factors were predictors of school achievement, in addition to the factors of processing speed and working memory, provided the evidence that the unique contribution of intelligence to school achievement is relatively low. Second, χ2 change, which was caused by constraining the path between cognitive ability and achievement to zero, was evaluated. In both datasets analyzed in that study, such constraints did not result in significant χ2 change, thus demonstrating the non-significant unique impact of intelligence on school achievement beyond processing speed and working memory (the significance of the unique contribution of intelligence beyond processing speed as a single predictor was not analyzed so far in that study). In contrast, Rindermann and Neubauer (2004) reported high direct effect of intelligence on school achievement (βdirect = .54 from the total effect of βtotal = .63, with additional specific indirect effects mediated by processing speed and creativity). Moreover, in the study reported by Rohde and Thompson (2007), the unique contribution of intelligence to scholastic performance differed depending on the measure of achievement. After controlling for working memory, processing speed, and spatial ability in the hierarchical multiple regression models, intelligence scores still accounted for additional 20% and 39% of variance in Wide Range Achievement Test III scores and SAT scores respectively, while its additional contribution to the explanation of the GPA was non-significant.

Yet another question that is much less frequently addressed is whether the entire effect of processing speed on school achievement is mediated by intelligence. In fact, the additional unique contribution of intelligence to scholastic performance discussed above is not in contradiction to the theoretically assumed sequence processing speed  intelligence  school achievement; however, the unique impact of processing speed beyond intelligence would be more problematic for this theoretical model. In those rare studies that examined the direct effect of processing speed, methods very similar to those described above were used. Rindermann and Neubauer (2004) reported the low direct effect of processing speed on school achievement (their model assumed the specific indirect effects of processing speed that was mediated by intelligence and creativity). The authors constrained this direct path to zero and did not observe significant changes in model fit statistics, and hence concluding that “a direct path between processing speed and school performance is statistically not necessary” (p. 582). Vock, Preckel, and Holling (2011) recently reported very similar results. In their study, mental speed was a significant single predictor of scholastic performance (β = .43), but its direct effect on school achievement (β = −.07) was not significant when reasoning and divergent thinking were entered into the model as mediators. These results were regarded as an evidence of “a full mediation for the effect of mental speed on academic achievement” (p. 366).

At the same time, in the two large datasets analyzed by Luo et al. (2006), zero constraints on the path between processing speed and school achievement resulted in significantly higher χ2 values. In other words, the unique impact of processing speed on school performance, not mediated by its effect on intelligence, was important. In the study by Rohde and Thompson (2007), working memory, processing speed, and spatial ability additionally explained 1% to 8% of variance in school achievement, after controlling for intelligence scores. When verbal and math achievement scores were analyzed separately, the additional contribution of basic processes was low for verbal scores and relatively high for math scores, reaching up to 13% of the additionally explained variance (with a good portion of additionally explained variability in math scores due to processing speed).

Finally, the fact that intelligence mediates (at least partly) the relationship between processing speed and school achievement was either explicitly pronounced or implicitly suggested in the previous studies; however, this mediating role of intelligence is itself worthy of particular examination. Indeed, the presence of this mediation effect was previously demonstrated based on the comparison of the proportion of variability in school achievement that is accounted for by processing speed or intelligence as a single predictor, before and after controlling for the other predictor (in terms of the R2s, zero-order and partial correlations or total and indirect effects). Together with the other analyses (like constraining the weights in the models and evaluating the χ2 changes), these methods served the purposes discussed above well. However, they in fact do not tell us much about the significance of the mediation effect itself, making a conclusion about the presence of mediation quite problematic. From the statistical point of view, these methods are in fact modifications of the “causal steps approach” (Baron and Kenny, 1986, Hyman, 1955, Judd and Kenny, 1981), a method for testing mediation effects that has been criticized by a number of recent statistical-simulation studies for having the lowest power (Fritz and MacKinnon, 2007, MacKinnon et al., 2002). More sophisticated procedures, primarily bootstrapping, are strongly recommended for this purpose (e.g., Lockwood and MacKinnon, 1998, MacKinnon et al., 2004, Preacher and Hayes, 2004, Preacher and Hayes, 2008, Shrout and Bolger, 2002) since they allow the construction of confidence intervals for the mediating effect itself; this in turn makes it possible to judge whether this effect differs significantly from zero. Until now, to our knowledge, the significance of the mediation effect processing speed  intelligence  school achievement has not been tested.

Thus, the present study examines the relationships between processing speed, intelligence, and school achievement using several perspectives. First, processing speed and intelligence are analyzed as individual predictors of school achievement; their relative strength as predictors of school achievement is compared. Second, the unique contribution of each processing speed and intelligence, beyond and above the other predictor, is estimated. Third, the significance of the mediation effect processing speed  intelligence  school achievement is tested.

Three more remarks should be made here concerning the present study. First, the mediating role of intelligence in the association between processing speed and school achievement that is suggested by the causal mental speed hypothesis was analyzed based on cross-sectional data in this study. Therefore, all the limitations of studies with cross-sectional design for testing mediation effects must be borne in mind.

Second, teacher-assigned grades in school subjects were used to represent school achievement, as they provide a measure of a relevant real-life performance. Thus, lower relationships with cognitive variables were expected, as compared to those reported for achievement tests. However, this choice was assumed not to affect the relative strength of the observed effects, both direct and indirect, in any systematic way.

Third, a sample of Russian high-school students was analyzed. This fact does not seem to have any special impact on the outcomes of the cognitive tests since only commonly used intelligence tests (Raven's Advanced Progressive Matrices and the subtests of the Amthauer's Intelligence Structure Test) were administered. However, Russian educational program and grading system differ from those referred to by the studies in the literature. Some details on the grading system and a list of the subjects used in the analysis are described in the next section. Again, the impact of the country where the participants study, if any, was assumed to concern the absolute, but not the relative magnitudes of the effects of processing speed and intelligence on school achievement.

Section snippets

Participants

The participants were 184 Russian high school students. The study was conducted in nine public schools located in Moscow (in the Russian educational system, secondary education is mostly provided by public state schools). Participants' mean age was 16.00 (SD = .67); 38% were male. Processing speed tasks were administered individually through the computers; intelligence was tested in groups. All tests were held in the schools where the participants studied. School grades were collected from the

Results

Descriptive statistics for response times, intelligence scores, and school grades are presented in Table 1. To estimate reliabilities of processing speed measures, distance-weighted mean response times were calculated separately for odd and even trials in each task. Odd–even correlations adjusted by the Spearman–Brown formula were rsb = .89 and rsb = .95 for DT and RMFT, respectively.

Further analyses were conducted on unstandardized residual scores with age and sex variances removed. Correlations

Discussion

In the literature on intelligence, the following are well-established facts: (1) intelligence explains a significant portion of variability in school achievement; and (2) to some extent, intelligence is predicted by processing speed. An obvious theoretical model that might account for these regularities would imply causal relationships, where processing speed would be regarded as a basic factor underlying cognitive ability, which in turn influences scholastic performance. In the strict sense,

References (56)

  • S. Krumm et al.

    Reasoning and working memory as predictors of school grades

    Learning and Individual Difference

    (2008)
  • L. Lu et al.

    Predicting school achievement from cognitive and non-cognitive variables in a Chinese sample of elementary school children

    Intelligence

    (2011)
  • D. Luo et al.

    Elementary cognitive tasks and their roles in psychometric g

    Intelligence

    (1999)
  • D. Luo et al.

    The causal factor underlying the correlation between psychometric g and scholastic performance

    Intelligence

    (2003)
  • D. Luo et al.

    The criterion validity of tasks of basic cognitive processes

    Intelligence

    (2006)
  • F.V. Rijsdijk et al.

    The genetic basis of the relation between speed-of-information processing and IQ

    Behavioural Brain Research. Special Issue: Behavioural Neurogenetics

    (1998)
  • H. Rindermann et al.

    Processing speed, intelligence, creativity, and school performance: Testing of causal hypotheses using structural equation models

    Intelligence

    (2004)
  • R.D. Roberts et al.

    Individual differences in speed of mental processing and human cognitive abilities: Toward a taxonomic model

    Learning and Individual Differences

    (1999)
  • T.E. Rohde et al.

    Predicting academic achievement with cognitive ability

    Intelligence

    (2007)
  • L.D. Sheppard et al.

    Intelligence and speed of information-processing: A review of 50 years of research

    Personality and Individual Differences

    (2008)
  • M. Vock et al.

    Mental abilities and school achievement: A test of a mediation hypothesis

    Intelligence

    (2011)
  • T.P. Alloway

    Working memory, but not IQ, predict subsequent learning in children with learning difficulties

    European Journal of Psychological Assessment

    (2009)
  • L.A. Baker et al.

    The genetic correlation between intelligence and speed of information processing

    Behavior Genetics

    (1991)
  • R.M. Baron et al.

    The moderator–mediator distinction in social psychological research: Conceptual, strategic, and statistical considerations

    Journal of Personality and Social Psychology

    (1986)
  • M. Bartels et al.

    Heritability of educational achievement in 12-year-olds and the overlap with cognitive ability

    Twin Research

    (2002)
  • A. Binet

    New methods for the diagnosis of the intellectual level of subnormals

    L'Année Psychologique

    (1905)
  • Ch. Brand

    General intelligence and mental speed: Their relationship and development

  • N. Brody

    Intelligence

    (1992)
  • Cited by (0)

    View full text