In November 2015 Palgrave Macmillan published my book ‘The Welfare Trait: how state benefits affect personality’. In it I propose an overarching account of personality and welfare, arguing that welfare policies which increase the number of children born into disadvantaged households risk proliferating dysfunctional employment-resistant personality characteristics, due to the damaging effect on personality development of exposure to childhood disadvantage. Jonathan Portes recently took to social media to dismiss the book on the basis that one particular set of government data cited in it is “wrong”. The data in question appear on page 72 of the book and are also shown below in Table 1: they suggest that there is an association between reproduction and welfare usage (the higher the proportion of unemployed adults in a household, the greater the number of children - on average - that it contains).
I responded to Mr Portes’s criticism by explaining that the data in question are publically available on the website of the Office for National Statistics (ONS). Interested parties can therefore verify that the data presented in the book match the source data and are, consequently, correct (see dataset ref 002530 at the following weblink: http://www.ons.gov.uk/ons/about-ons/business-transparency/freedom-of-information/what-can-i-request/published-ad-hoc-data/labour/march-2014/index.html ). Those readers who have checked the weblink will see that the term “Households” refers to those where at least one occupant is aged 16-64 and at least one occupant is aged 0-15. This is an important point that I will return to later. Anyhow, following my rebuttal Mr Portes conceded that the data were, after all, “correct”, but that they were “irrelevant” to my thesis on the grounds that the comparison of employment status and reproduction shown in Table 1 is only for households with children. He went on to cite ONS data that includes childless households which he claimed disprove the association seen in Table 1. Although I disagree with Mr Portes’s conclusion for various reasons that are set out below, I nevertheless applaud his willingness to discuss this issue with me here on Dr James Thompson’s website: a vigorous debate can only help advance our understanding of possible links between personality, welfare usage and reproduction.
The first problem with Mr Portes’s argument is that it rests solely on census-style government data. This is a problem because these data, while benefiting from large sample sizes, suffer the drawback that they are plagued with confounds. For example, as mentioned by the commentator below, households are difficult to define. Additionally, reproduction may be prevented by physical problems that also hinder employment. This confound issue means that if we go fishing in government datasets it is easy to find seemingly impressive results that are actually spurious, because they are the product of confounds. As I mention on page 169 of The Welfare Trait, this phenomenon is best illustrated by visiting the website that mines US government datasets for significant correlations and shows just how misleading they can be (www.tylervigen.com). As an example, this scatterplot presents data from the website which suggests deaths from tangled bed sheets are linked to per capita cheese consumption: since there is no verified causal mechanism linking cheese-eating with tangling of bed sheets, this finding is spurious.
The same confound problem applies to the ONS data I present in Table 1 and I acknowledge this issue on page 72 of the book. But the difference between my position and that of Mr Portes is that the ONS data I cite converges with the findings of studies that have explored causality and have been published in peer-reviewed journals, which means they have been vetted for confounds. In particular the data shown in Table 1 converge with studies from both the UK and USA that show reproduction in welfare claimants rises with benefit generosity (Argys et al., 2000; Brewer et al., 2011). Furthermore, these studies tested causation by means of follow up interviews and found that claimants adjusted fertility by changing contraceptive usage, suggesting that the association between reproduction and unemployment shown in Table 1 is likely to be causal.
A second problem with Mr Portes’s argument is that he claims the focus on households with children that occurs in Table 1 is a weakness. I think it is actually a strength because it has the benefit of comparing like with like (i.e., all three categories contain households with children). This reduces some of the noise that is found in datasets that combine childless households with those that contain children. I should perhaps with hindsight have emphasised this point in the book, but it is implicit because, as has already been noted, the government dataset from which it is taken states that households refers to those where at least one occupant is aged 16-64 and at least one occupant is aged 0-15.
A third problem with Mr Portes’s argument is that the link between reproduction and unemployment suggested by the data in Table 1 tallies with a mosaic of other peer-reviewed findings – cited in the book – that show a tendency for individuals with certain personality characteristics (primarily low levels of conscientiousness and agreeableness) to have more children than average individuals. Given that longitudinal studies have revealed that these same personality characteristics also raise the risk of unemployment (hence I have dubbed it the “employment-resistant” personality profile), the link between reproduction and unemployment that appears in Table 1 is a plausible addition to this mosaic of findings. In conclusion, when all these different types of data are viewed as a whole, they tell a consistent story of a connection between personality, welfare policy and reproduction that means these data in Table 1 are, by any reasonable analysis, relevant to the thesis that I present in The Welfare Trait.