Evidence from twin studies implies that economic and political traits have a significant heritable component. That is, some of the variation between people is attributable to genetic variation.
Despite this, there has been a failure to demonstrate that the heritability can be attributed to specific genes. Candidate gene studies, in which a single gene (or SNP) is examined for its potential influence on a trait, have long failed to identify effects beyond a fraction of one per cent. Further, many of the candidate gene results fail to be replicated in studies with new samples.
An alternative approach to genetic analysis is now starting to address this issue. Genomic-relatedness-matrix restricted maximum likelihood (GREML – the term used by the authors of the paper discussed below) is a technique that looks to examine how the variance in traits can be explained by all of the SNPs simultaneously. This approach has been used to examine height, intelligence, personality and several diseases, and has generally shown that half of the heritability estimated in twin studies can be attributed to the sampled SNPs.
A new paper released in PNAS seeks to apply this approach to economic and political phenotypes. The paper by Benjamin and colleagues shows that around half the heritability in economic and political behaviour observed in behavioural studies could be explained by the array of SNPs.
The authors used the results of recent surveys of subjects from the Swedish Twin Registry, who had their educational attainment, four economic preferences (risk, patience, fairness and trust) and five political preferences (immigration/crime, foreign policy, environmentalism, feminism and equality, and economic policy) measured. The GREML analysis found that for one economic preference, trust, the level of variance explained by the SNPs was statistically significant, with an estimate of narrow heritability of over 0.2. Two of the political preferences, economic policy and foreign policy, had narrow heritability that was statistically significant, with heritability estimates above 0.3 for each of these. The authors noted that as the estimates are noisy and GREML provides a lower bound, the results are consistent with low to moderate heritability for these traits.
Educational attainment was also found to have a statistically significant result, although the more precise measurement of educational attainment and the availability of this data across all subjects made that result more likely.
This result is corroboration of the evidence from twin studies and provides a basis for believing that molecular genetic data could be used to predict phenotypic traits. However, one interesting feature of the GREML method of analysis is that after conducting this analysis with one sample, the data obtained does not assist in predicting the traits for someone out of the sample. This technique shows the potential of molecular genetic data without directly realising those results.
As a comparison, the authors examined whether any individual SNPs might predict economic or political preferences, but found none that met the significance test standard of 5×10-8. Such a high level of significance is required to reflect the huge number of SNPs that are being tested.
The authors also conducted the standard comparison between monozygotic (identical) and dizygotic (fraternal) twins, which resulted in heritability estimates consistent with the existing literature, although with a much larger sample than typically used. Looking through the supplementary materials, the major surprise to me was that the twin analysis suggests that patience has low heritability, with a very low correlation between twins and almost no difference between monozygotic and dizygotic twins (in fact, for males, dizygotic twins were more similar).
The authors draw a few conclusions from their work, many which reflect the argument in a Journal of Economic Perspectives article from late last year. The first and most obvious is that we should treat all candidate gene studies with caution. Hopefully some journals that insist on publishing low sample size candidate gene studies will pay attention to this. Where they are going to be conducted, you need very large samples, and significantly larger than are being used in most studies being published.
Meanwhile, they are still hopeful that there can be a contribution from genetic research, particularly if the biological pathways between the gene and trait can be determined. This might include using genes as instrumental variables or as control variables in non-genetic empirical work. The use as instrumental variables does require, however, some understanding of the pathways through which the gene acts as it may have multiple roles (that is, it is pleiotropic). They also suggest that the focus be turned to SNPs for which there are known large effects and the results have been replicated.
On element of analyses of political and economic preferences that makes me slightly uncomfortable is the loose nature of these preferences. For one, the manner in which they are elicited from subjects can vary substantially, as can the nature of the measurement. Take the 2005 paper by Alford and colleagues on political preferences, which canvassed 28 political preferences. Many of the views are likely to change over time and be highly correlated with each other. And why stop at 28?
As a result, it may be preferable to take a step back and ensure that data on higher level traits are collected. I generally consider that IQ and the big five personality traits (openness, conscientiousness, agreeableness, extraversion and stability) are a good starting point and are likely to capture much of the variation in political and economic preferences. For example, preferences such as patience are likely to be reflected in IQ, while openness captures much of the liberal-conservative spectrum of political leaning. Starting from a basis such as this may also give greater scope for working back to the biological pathways.