Elsevier

Intelligence

Volume 36, Issue 2, March–April 2008, Pages 143-152
Intelligence

Effects of age and schooling on intellectual performance: Estimates obtained from analysis of continuous variation in age and length of schooling

https://doi.org/10.1016/j.intell.2007.03.006Get rights and content

Abstract

The effects of age and schooling on different aspects of intellectual performance, taking track of study into account, are investigated. The analyses were based on military enlistment test scores, obtained by 48,269 males, measuring Fluid ability (Gf), Crystallized intelligence (Gc), and General visualization (Gv) ability. A regression method, relying on simultaneous variation in age and length of schooling at the time of the testing, was used. The results showed that both age and schooling increase performance, with the exception of the test which primarily measures Gf, for which the age effect was negative. The effects of schooling were considerably stronger than the effects of age, and there was a pattern of differential schooling effects on different tests which matched the curricular emphasis of the programs. The estimates of the strength of the schooling effect obtained with this new method were generally in good agreement with the results obtained in previous research.

Introduction

The relative amount of influence of chronological age and formal schooling on the development of intellectual performance is a crucial problem in educational research, which has considerable theoretical, methodological and practical implications. A widely held view claims that schooling can increase intellectual performance, even if reviews of research by, for example, Ceci (1991), Herrnstein and Murray (1994), and Winship and Korenman (1997) have arrived at quite different conclusions about the strength of the effect.

Herrnstein and Murray (1994) concluded, on the basis of an empirical study and a review of the literature, that there is little evidence that differences in the amount of schooling account for much of intellectual variation. Their own study showed an increase of about 1 IQ point per year of schooling, which is lower than what has been obtained in other similar studies. However, Winship and Korenman (1997) conducted a reanalysis of these data, and after correction of some data problem and by use of alternative model specifications, they obtained an estimated effect of 2.7 IQ points per year of schooling. They also reviewed the research on effects of schooling on intelligence with a special emphasis on studies using an analysis of covariance design. These results showed that estimates varied quite a lot over studies, from 1 to 4 IQ points per year of education.

One of the reasons for the differences in results is the methodological problem on how to disentangle the school effect from the effect of chronological age on the growth of intellectual performance since these are typically confounded. Ceci (1991) identified in an influential review of 200 studies, eight different types of designs in this area, and concluded that the results from all the different approaches indicate that schooling exerts an effect on intellectual development, even though the estimates vary over studies, and all the studies suffer from different kinds of methodological limitations.

The main purpose of this study is to gain further knowledge of the effects of chronological age and schooling on the development of intellectual performance through use of more powerful methods and more appropriate data than have been used previously. The basic idea is to investigate effects of simultaneous variation in age and amount of schooling, which allows separation of the relative amount of influence of age and schooling. The analysis also investigates differential effects of schooling on different aspects of intellectual performance, taking track of education into account.

In one of the approaches identified by Ceci (1991) effects of variation in length and/or track of education on intellectual performance have been studied through comparison between different programs. Since individuals are both selected and self-selected into different educational programs, it is necessary to control for initial performance differences, which has been done by covariance-analytic techniques. Such studies require longitudinal data of good quality. One of the classical studies in this field was conducted by Härnqvist (1968). In this longitudinal study, Härnqvist tested a nationally representative 10% sample of the Swedish population of 13-year olds with a test battery including a verbal, a spatial and an inductive test. At the age of 18, the male subset of the sample took another test battery of similar composition at the enlistment to military services. In the analysis, the test results at age 13 were used to control for differences in entry characteristics to the different tracks of education. The study showed that students who had the most academically oriented education gained more in general intelligence compared to those with the least amount of academic education. Several similar studies have been carried out on Swedish data (e.g., Balke-Aurell, 1982, Gustafsson, in press, Husén and Tuijnman, 1991). There also are other such studies, several of which are Scandinavian (e.g., Lund & Thrane, 1983). All these studies have shown fairly strong effects of schooling on intellectual level, amounting to around 2.0–2.5 IQ points for each additional year of academic schooling.

It also has been demonstrated (e.g., Balke-Aurell, 1982, Gustafsson, in press) that it is necessary to take into account the fact that education in upper secondary school is not a homogeneous activity, and that the characteristics of different educational tracks are important for effects on different aspects of intellectual performance. Balke-Aurell (1982) concluded that spatial/technical ability factors develop in accordance with verbal and technical types of education, and, to a lesser extent, with type of occupation. Gustafsson (in press) also found differences in effects on intelligence for different tracks of education. The effects were minimal for vocational schooling, and among the academic tracks (Economics, Social Sciences, Natural Sciences and Technology) the lowest effect was observed for the Social Sciences track and the strongest for the Technology track. The mean result for the academic tracks corresponds to an improvement of about 2.5 IQ points per years of study. Furthermore, the results indicated that certain schooling experiences cause improvements both in general cognitive ability, and in specific abilities. The results showed that academic tracks with technical and science orientation cause at least as strong an improvement in General Visualization ability (Gv; Carroll, 1993) as in general ability, while for General Crystallized ability (Gc; Carroll, 1993) weaker positive effects were obtained for the academic tracks and some of the vocational tracks included in the study, compared to the effects on general cognitive ability.

However, these results showing a gain in intellectual performance as an effect of schooling raise a number of questions concerning validity and interpretation. Since experimental methods cannot be used for ethical and practical reasons, the schooling effects are not easily determined. In particular, the methodology used in these studies has been criticized for not providing a fully adequate control of initial differences (Brody, 1992), making the results vulnerable to self-selection bias. Brody has emphasized that it is not possible to control for all potential initial differences with statistical methods, which makes it uncertain whether the estimated effect is a result of schooling or of selection. Brody (1992) argued in criticism of the Härnqvist (1968) study that:

Individuals who chose or were assigned to an academic track might have gained in IQ even if they had been randomly assigned to a less rigorous academic education. Consider two individuals with the same IQ who elect to enter different educational tracks at the secondary school level. The student who chooses the academic track may like to read books more than the individual who chooses a less academic track. Differences in intellectual interest may be related to changes in IQ (p. 187).

Another problem concerning the validity and interpretation of obtained changes in IQ, is that even if the empirical results indicate that there is a change in IQ points as a function of amount of schooling, this need not reflect real changes in intelligence (Ceci, 1991). Ceci discussed alternative explanations of the empirically established effects of schooling on intelligence tests (1991, p. 771).

…schooling might influence IQ performance because the experience of being in school alters individuals' cognitive processes in a rather fundamental manner. This possibility can be contrasted with one that speculates that schooling does nothing to alter the efficiency of an individual's cognitive processes but merely supplies them with a reservoir of IQ-relevant knowledge and shapes their style of responding.

He considered that even though it could be argued that the ability for abstract thought increases as a result of schooling, the evidence for this is not compelling. Furthermore he emphasized that aptitude for abstract thought can be found even for non-schooled individuals.

Hansen, Heckman, and Mullen (2003) postulated that both schooling and scores on ability tests are generated by a common unobserved latent ability. They found that effects of schooling on test scores for a given level of ability are roughly linear across schooling levels, and that effects of schooling are slightly larger for those with lower ability. The effects of schooling on intelligence estimated in this study were in agreement with those demonstrated by Winship and Korenman (1997).

The approach, which by Ceci was characterized as the strongest of the eight designs for determining the effects of schooling, is the between-grades regression discontinuity design. In this design regression techniques are used to estimate the independent effects of age and schooling. The slope of the regression within grades on chronological age estimates the effect of age, and the effect of schooling is represented by the discontinuity between the regression lines for two adjacent grades. That is, the differences between the oldest and the youngest individuals in each grade estimate the net effect of a one-year difference in chronological age. The difference between the youngest individuals in any given grade level and the oldest ones in the lower adjacent grade level provides an estimate of the effect of one year of schooling. Cahan and Cohen (1989) reported a study with this design, and the results showed that the effect of schooling was about twice as strong as the effect of age. The magnitude of the influence of schooling on intellectual development was by Cahan and Cohen regarded as unexpectedly strong, but the finding has had strong impact. Other studies have replicated the results of the Cahan and Cohen study using a similar approach (e.g., Crone and Whitehurst, 1999, Stelzl et al., 1995). However, even though this design is strong, there are some problems with this approach as well.

One problem, which also was identified by Cahan and Cohen (1989), is the assumption that admission to school is based only on age, according to some arbitrary cut-off date. However, in reality the admission of some individuals is delayed or made earlier in time. Thus, they are over- or under-aged within their grade level. Moreover, delayed and early admission is not random with respect to age and intellectual development. Early admission is more likely to occur for bright individuals, who are born within some weeks after the cut-off date, whereas delay is more likely for intellectually less developed individuals who are born within some weeks before the cut-off date (e.g., Svensson, 1993). These selection effects imply that the regression on age will be underestimated, which means that the effect of schooling is overestimated. In recognition of this problem, Cahan and Cohen excluded those individuals who were over- or under-aged within their grade level, as well as those whose birthday fell within the month before or after the cut-off date. The regression equation obtained from these data was then extrapolated to cover the whole range of one year. However, even though these actions reduce the underestimation of the age effect, it is uncertain to what degree the problem remains. Another potential problem is the use of regression models which rely on the assumption of linearity of the within-grade regression. If the regression of intellectual performance on age is not perfectly linear, the effect of age is underestimated.

In an optimal design for separating effects of age and schooling there should be variation with respect to both length of schooling and chronological age, and the factors should vary independently of one another. To our knowledge, no study with such a design has been carried out. However, available data makes it possible to come close to such an optimal design. In Sweden, the tests used for enlistment to military services are generally carried out during the year the individual turns 18 year old. Since the testing sessions occur at different dates during the year, both length of schooling at different tracks in upper secondary school and the age at the time of the testing vary. Since age and length of schooling in these data vary more or less independently it is possible to estimate the relative influence of age and schooling on the intellectual performances measured by the military test battery. Because enlistment to military service in Sweden is compulsory for males only, the analyses in the present study will be restricted to this gender.

Section snippets

Participants

The study relies on data from “the Gothenburg educational longitudinal database” (GOLD), which contains information from different official registers about every individual in Sweden born between 1972 and 1987 who lived in Sweden at the age of 16. Register information such as scores from the military enlistment battery and registration on tracks from upper secondary school is available for all males in the population. Included in the study are those 48,269 males who: (a) were born in the years

Results

The presentation of the results is divided into two steps. In the first step, descriptive statistics are reported, both for the pooled group and for the five academic tracks. In the second step, results from the CC models comprising the pooled group are presented, followed by the results from the models applied to data from each of the five academic tracks.

Discussion and conclusion

In the present study a new method (the continuous age/continuous treatment method or the CC method) for estimating effects of schooling on intellectual ability has been applied. Among the many methods used for determining effects of schooling the regression discontinuity design, which takes advantage of adjacent grades and age variation within grades, is regarded to be the strongest one (Ceci, 1991). The CC method is similar to the regression discontinuity design, except that the estimates are

Acknowledgement

The Bank of Sweden Tercentenary Foundation has financially supported the research reported in this article. The work is a part of the research project GRASP (an acronym for “Effects of age and schooling on the development of intellectual performance”).

References (24)

  • J.-E. Gustafsson

    A unifying model for the structure of intellectual abilities

    Intelligence

    (1984)
  • I. Stelzl et al.

    The effect of schooling on the development of fluid and crystallized intelligence: a quasi-experimental study

    Intelligence

    (1995)
  • G. Balke-Aurell

    Changes in ability as related to educational and occupational experience

  • N. Brody

    Intelligence

    (1992)
  • S. Cahan et al.

    Age versus schooling effects on intelligence development

    Child Development

    (1989)
  • B. Carlstedt et al.

    Construct validity of the Swedish Enlistment Battery

    Scandinavian Journal of Psychology

    (1993)
  • J.B. Carroll

    Human cognitive abilities

    (1993)
  • S.J. Ceci

    How much does schooling influence general intelligence and its cognitive components? A reassessment of the evidence

    Developmental Psychology

    (1991)
  • S.J. Ceci

    Schooling and intelligence

  • D.A. Crone et al.

    Age and schooling effects on emergent literacy and early reading skills

    Journal of Educational Psychology

    (1999)
  • J.-E. Gustafsson

    Measurement from a hierarchical point of view

  • Gustafsson, J. -E. (in press). Schooling and intelligence: Effects of track of study on level and profile of cognitive...
  • Cited by (82)

    • The development of intelligence: Education and neuroscience

      2021, Encyclopedia of Behavioral Neuroscience: Second Edition
    • The future of intelligence: The central meaning-making unit of intelligence in the mind, the brain, and artificial intelligence

      2021, Intelligence
      Citation Excerpt :

      Even when obtained, increases are short-lived, fading out in about 3 years (Protzko, 2015). It is practically important to note, however, that g is highly predictive of school achievement, accounting for about 30% of variance in school grades (Demetriou, Kazi, Spanoudis, & Makris, 2019, 2020), and it changes because of it (Cliffordson & Gustafsson, 2008). The empirical facts above are generally accepted but the nature of g is strongly disputed for a century.

    View all citing articles on Scopus
    View full text