On inequality, health, scientific progress and political argument: A response to Dorling and Barford Ron Johnston a,Ã , Min-Hua Jen b , Kelvyn Jones a a School of Geographical Sciences, University of Bristol, University Rd, Bristol BS8 1SS, UK b Department of Primary Care and Social Medicine, Imperial College London, London W6 8RP, UK article info Article history: Received 13 June 2009 Accepted 13 June 2009 Dorling and Barford’s response to our note (Jen et al., 2009a) on the link between inequality and health asks whether there is not something better ‘than one academic paper merely cancelling out another’. We always assumed that science progressed in that way, that hypotheses were exposed to critical tests and where they were found wanting, rethinking was called for. To accuse us of ‘sly innuendo’ suggesting that ‘the other side have not been quite smart enough’ through ‘a little sad showing-off’ involving ‘a bit of a tragic waste of time’ is just a parody of science. All we did was suggest – through two analyses (one of simulated data and the other of survey data) and following the lead of another researcher (Gravelle, 1998) – that an argument widely rehearsed in the literature might be based on no more than a statistical artefact. We may be wrong – and if so our arguments should be put to critical tests rather than respondents getting involved in silly name-calling that does the goal of science and its contribution to progress no good at all. The critique raises an important empirical point regarding our paper, with its claim that our use of self-assessed health as a proxy for mortality rates is invalid, and that we should not generalise from the source we cited because it dealt with a different scale. To sustain their case Dorling and Barford present data for 15 countries – slightly different from our 12, but very probably not crucially so. Their graph of life expectancy against self-assessed good health rates for those 15 has only a weak fit between the two (an r of 0.19); they also suggest that if the two lower income OECD countries (Mexico and Turkey) are omitted there is a stronger, negative relationship (r ¼0.24) – which is the reverse of what was expected. (Wilkinson argues that his theory does not apply to Mexico and Turkey: it is valid for ‘rich nations’ only, but the boundary between the two is, necessarily, subjective. 1 ) With a small n it is always possible that the removal of one or two observations will lead to a substantially different regression outcome – as is clearly the case with those 15 countries because of heteroscedasticity. One could look at their graph and say that excluding Japan – justified because it has a very particular culture, a point that Wilkinson makes elsewhere (e.g. Wilkinson and Pickett, 2009) – one would get a strong positive relationship (r ¼ 0.5). And if you exclude Japan, Mexico and Turkey, you get a weak, insignificant negative relationship (r ¼0.16) – not surprisingly given that there is very little variation in life expectancy over the 12 remaining countries (from 77 to 82 only, with four values of 81 and five of 79). At best, therefore, the issue is unresolved. The data presented to counter our case is unconvincing. More importantly, Dorling and Barford do not address the fundamental reason why we used self-assessed health as our dependent variable. The core of our argument is that a proper, rigorous test of Wilkinson’s hypothesis regarding the link between income inequality and mortality must avoid the potential pitfalls of the ecological fallacy – assuming that a relationship at one scale (the aggregate) applies at another (the individual). You need to combine individual with contextual data, which is what we do using a relevant analytical procedure – multi-level modelling. Of course, with individual data you cannot get a measure of mortality unless you have panel data: mortality rates – or life ARTICLE IN PRESS Contents lists available at ScienceDirect journal homepage: www.elsevier.com/locate/healthplace Health & Place 1353-8292/$ - see front matter & 2009 Elsevier Ltd. All rights reserved. doi:10.1016/j.healthplace.2009.06.004 DOI of original article: 10.1016/j.healthplace.2009.06.005 Ã Corresponding author. E-mail address: r.johnston@bristol.ac.uk (R. Johnston). 1 For analyses of this relationship in a much wider range of countries, see Jen et al. (2009b, 2010). Health & Place 15 (2009) 1163–1165