Skip to main content

Thank you for visiting You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.


How intelligence changes with age

An analysis of common genetic variants shows that hereditary factors that influence intelligence in childhood also affect it in old age. Such work could signal the end of the nature–nurture controversy. See Letter p.212

Francis Galton — Charles Darwin's half-cousin — argued 150 years ago that “there is no escape from the conclusion that nature prevails enormously over nurture”. Since then, the nature–nurture (or genetics–environment) controversy has never been more contentious than when it concerns human intelligence. A report by Deary and colleagues1 on page 212 of this issue, however, may mark the beginning of the end of this debate. Instead of estimating genetic influence on intelligence indirectly by using special groups such as twins and adoptees, the authors use DNA data from unrelated people.

A traditional approach to estimate the heritability of a trait, or phenotype, has been to compare groups of known genetic relatedness, such as identical twins (100% relatedness) and fraternal twins (roughly 50%). The strength of this approach is that it estimates the net effect of genetic influence without the need to know which genes are responsible. However, this absence of DNA sequence information is also its weakness.

So, following the sequencing of the human genome, researchers had great expectations from genome-wide association studies (GWAS). It was hoped that GWAS would identify enough associations between DNA sequence variants (typically, single nucleotide polymorphisms, or SNPs) and complex traits such as intelligence to account for most of the heritability of the traits. But such analyses, sometimes involving hundreds of thousands of individuals, have detected only a small portion of genetic influence, even for highly heritable traits such as height and weight. For instance, initial GWAS of intelligence2,3 have indicated contributions of many small genetic effects. This is because the genomic differences identified so far between individuals make only a small total contribution to the heritability of this trait — an issue that has been dubbed the missing heritability problem4.

Deary et al.1 use a variation of genome-wide complex-trait analysis (GCTA), a method that complements GWAS. In GCTA, researchers use DNA data for hundreds of thousands of SNPs from unrelated individuals to estimate genetic influence on a particular trait. Unlike the hypothesis-testing approach of GWAS, GCTA does not specify which DNA variants are associated with a measured trait. Instead, it is a parameter-estimation approach that relates similarity in SNPs to phenotypic similarity between pairs of individuals. The use of a large sample, together with pair-by-pair comparisons, allows amplification of the weak signal derived from the low genetic similarity between unrelated subjects. Heritability is estimated as the extent to which genetic similarity can account for phenotypic similarity.

GCTA has been applied to estimate heritability for traits such as height5 and weight6, psychiatric and other medical disorders7, and intelligence3. To estimate how genetic factors influence the stability of intelligence and how it changes with age, Deary et al.1 applied this approach to SNP data and intelligence-test scores from almost 2,000 unrelated people from Scotland. What is especially exciting about this report is that, in contrast to previous GCTA studies3,4,5,6,7, the authors extend their analysis to the multivariate case and obtain a noteworthy result. Essentially, the multivariate extension of GCTA evaluates relatedness between each pair of individuals for different traits. In Deary and colleagues' report, the different traits are intelligence assessed at two stages of life in the same people: in childhood (at age 11) and, half a century later, in old age. Specifically, they estimate genetic change and continuity as the extent to which similarity in SNPs between two individuals can account for similarity in change and continuity in their intelligence.

The authors find a substantial genetic correlation (0.62) between intelligence in childhood and in old age, which means that many of the same genetic elements are associated with this trait throughout life. The analysis also estimates the genetic influence on cognitive change across life: nearly a quarter of the variation in the change in cognitive scores that occurs throughout life could be explained by different genes being associated with this trait in childhood and later life. These findings are consistent with previous results8 from family-based genetic research, although no family-based studies have extended from childhood to old age.

Deary et al. show appropriate caution about their estimates. Their results are valuable because such data are rare, but the results come with large standard errors (a measure of how the data values spread around the mean) and are not statistically significant by conventional measures. This is because, in GCTA, a tiny signal is extracted from a lot of noise, and so samples in the tens of thousands — much larger than the one used in this report — are needed for accurate estimates.

Nonetheless, GCTA will stimulate research on the genetics of intelligence, because this method does not require special samples such as groups of twins or adoptees. Indeed, multivariate GCTA could be used to test findings from family-based research on intelligence. These findings include that the same genetic factors affect different cognitive abilities and disabilities, and that genetic propensities for intelligence correlate and interact with cognitively relevant experiences9.

The prerequisites for GCTA — very large samples in which huge numbers of SNPs have been analysed — seem daunting. But these are the same resources required for GWAS, and many such samples are already available for several traits, including intelligence. Another caveat of GCTA, however, is that it underestimates heritability because it is limited to SNPs that have been mapped on the genome and to DNA variants correlated with those SNPs (that is, variants that are in linkage disequilibrium with them). By contrast, traditional family-based genetic designs capture variation due to all causal variants in the genome.

Regardless of such caveats, GCTA may provide crucial clues for solving the missing heritability problem. It has been suggested10 that, to find genes associated with complex traits such as intelligence, researchers need to analyse rare genetic variants in addition to the common ones that are detected by available microarray tools. However, to the extent that GCTA estimates of heritability account for heritability estimates derived from family-based studies, this suggests that common SNPs can powerfully predict intelligence if sample sizes are sufficiently large. If true, this means that intelligence is similar to height in terms of genetic architecture and that — with similar sample sizes to those used for research on heritability of height — many associations between DNA and intelligence will be found.

So, although GCTA cannot quite mark the end of the nature–nurture controversy, it might be the beginning of the end. Similar to family-based genetic methods, this approach is limited to estimating genetic influence indirectly from genetic similarity between pairs of individuals, rather than directly from specific genes, which is the ultimate goal. But it is much more difficult to dispute GCTA results based on DNA data than it is to quibble about twin and adoptee studies.


  1. 1

    Deary, I. J. et al. Nature 482, 212–215 (2012).

    ADS  CAS  Article  Google Scholar 

  2. 2

    Davis, O. S. P. et al. Behav. Genet. 40, 759–767 (2010).

    Article  Google Scholar 

  3. 3

    Davies, G. et al. Mol. Psychiatry 16, 996–1005 (2011).

    CAS  Article  Google Scholar 

  4. 4

    Maher, B. Nature 456, 18–21 (2008).

    CAS  Article  Google Scholar 

  5. 5

    Yang, J. et al. Nature Genet. 42, 565–569 (2010).

    CAS  Article  Google Scholar 

  6. 6

    Yang, J. et al. Nature Genet. 43, 519–525 (2011).

    CAS  Article  Google Scholar 

  7. 7

    Lee, S. H., Wray, N. R., Goddard, M. E. & Visscher, P. M. Am. J. Hum. Genet. 88, 294–305 (2011).

    Article  Google Scholar 

  8. 8

    Lyons, M. J. et al. Psychol. Sci. 20, 1146–1152 (2009).

    Article  Google Scholar 

  9. 9

    Plomin, R., DeFries, J. C., McClearn, G. E. & McGuffin, P. Behavioral Genetics 5th edn (Worth, 2008).

    Google Scholar 

  10. 10

    Cirulli, E. T. & Goldstein, D. B. Nature Rev. Genet. 11, 415–425 (2010).

    CAS  Article  Google Scholar 

Download references

Author information



Corresponding author

Correspondence to Robert Plomin.

Rights and permissions

Reprints and Permissions

About this article

Cite this article

Plomin, R. How intelligence changes with age. Nature 482, 165–166 (2012).

Download citation

Further reading


By submitting a comment you agree to abide by our Terms and Community Guidelines. If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.


Quick links

Nature Briefing

Sign up for the Nature Briefing newsletter — what matters in science, free to your inbox daily.

Get the most important science stories of the day, free in your inbox. Sign up for Nature Briefing