26 October 2011

Does income inequality cause health and social problems? The challenge of evidence reviews

Just a few days ago on the Health Equity JISC Email Network Kate Rowlingson the author of the report ‘Does incomeinequality cause health and social problems?’ posted a critical response by Richard Wilkinson and Kate Pickett which shows the challenges of doing good reviews and how individual reviews while a great shortcut to reading the original papers need to be treated as carefully, and considered as critically, as original research papers.

I respect and commend that fact the Kate Rowlingson forwarded the critical response  to a wider audience and encourages debate.

The recent review was for the well respected UK charitable foundation the Joseph Rowntree Foundation on the links between income inequality and health and social problems.

Click here to download the report.
The report "provides an independent review of the evidence about the impact of inequality. Inequality grew dramatically in the 1980s and has remained at a high level ever since. But should high levels of inequality concern us? This report provides an independent review of the research, paying particular attention
to the evidence and arguments put forward in The Spirit Level by Richard Wilkinson and Kate Pickett, in which it was argued strongly that we should indeed be concerned about income inequality. This report reviews the points made in various critiques that have appeared since The Spirit Level was first published in 2009, alongside the evidence and debate in the broader peer-reviewed literature. The report examines:
  • whether or not there is a link between income inequality and health and social problems;

  • who might be most affected by income inequality; and

  • other possible impacts of income inequality, for example, on the economy"

Critical Response
Richard Wilkinson and Kate Pickett’s response to KarenRowlingson’s report for the Joseph Rowntree Foundation Report ‘Does incomeinequality cause health and social problems?’

" We are glad that an independent report from JRF has reviewedcritical responses to our work and found it to be robust.  The report recognizes that the critics havebeen "far more selective" in their presentation of data.  We are also glad that it cites thestatistical review by Noble (2010), which provides a detailed discussion oftechnical issues relating to The Spirit Level and strongly supports ourapproach.

However, although we very much welcome the generalconclusions of this report, it nevertheless contains a number of errors whichwill lead the reader to underestimate the power of income inequality to affectsocieties.  We regret that neither of uswere invited to participate more fully in the deliberations which led to thisreport.  We commented on a first draftand on the final report, but were told that there was not time for substantialchanges to the final report.  Althoughthe report gives our work a clean bill of health, it is marred by importantinaccuracies likely to mislead many readers.  We deal here only with the most important.

At several points it is said that the size of the effect ofincome inequality “on health and social problems” is small.  There are several mistakes here. First, thesestatements infer from studies of the effects of income inequality on health toall the other outcomes which have been shown to be related to inequality.  In other words a study suggesting that incomeinequality has a small effect on health is interpreted as meaning that incomeinequality also has a small effect on teenage births, violence, imprisonment,child wellbeing, trust etc..

Not only is the inference obviously unsound, but it is madein the face of the much stronger relationships shown for other outcomes, whichare not discussed. The relationships we have reported between inequality andhealth, though statistically significant, are weaker than those with otheroutcomes (see tables of correlation coefficients on p.15 of the JRFreport).  Not only are the relationshipsbetween income inequality and other outcomes much closer, but the magnitude ofdifference in outcomes between more and less equal societies is often vast:three-fold differences in population rates of mental illness, four-folddifferences in the proportion of people who feel they can trust each other,two-and-a-half –fold differences in rates at which pupils drop out of US highschools,  six to ten-fold differences inteenage pregnancy rates, almost ten-fold differences in the proportion of thepopulation in prison and vast differences in child-wellbeing, drug abuse, andsocial mobility. 

To call these differences “small” is hopelessly misleading –particularly when these findings are based entirely on data from the mostrespected sources and are not discussed or challenged in any way in thisreport.  Not only that, but researchworkers using data covering different societies have sometimes found evenbigger differences than we do – see for instance the 10-fold differences inhomicide reported both by Daly et al (2001) and, on separate data, by a groupat the World Bank (Fajnzylber et al 2002). Neither of these papers, nor a review of the literature which concludesthat the homicide and inequality link is robust (Hsieh al 1993), is referencedin the report.

The mistake is compounded by the statement on p.11 that thecorrelation coefficients between income inequality and homicides, educationalperformance, life expectancy and infant mortality “fall below the 0.5threshold”.  This statement seems toreflect nothing more than a bizarre confusion between correlation coefficientsand probabilities.  Not only are all therelationships statistically significant (p<0.05 ), but a correlationcoefficient of r = 0.5 means that  25 percent of the variance in one variable is accounted for by the other.  An explanation of even 25 per cent of thevariance in an important outcome is impressive. Some of the other correlation coefficients suggest twice as much of thevariance is explained by income inequality.

Lastly, the statement that the effects of income inequalityare small is based on a paper providing a meta-analysis of studies usingmultilevel models of health and income inequality which substantiallyunderestimated the overall effects of income inequality. Within these multilevelmodels the effects of individual income and/or education are controlled out.However, there is now widening agreement (see for instance M. Marmot’s TheStatus Syndrome) that individual income and education are related to healthsubstantially because they serve as markers of social status. To measure theeffects of inequality after controlling for individual status differences isclearly over-controlling.  We made thispoint clearly in our British Medical Journal editorial (2009) which accompaniedthe original meta-analysis.  Not torecognise this is analogous to thinking you can measure the effects of socialclass hierarchy while controlling for the effects of individual social class.

There are a couple of others points which we think are alsoimportant enough to need a comment.  

First, a point about the difference between the OECD income inequalitydata and the data we used from the UN (which was also given by the WorldBank).  Much the most importantdifference is that Japan is one of the most equal countries in the UN data butappears very much less equal in the OECD data. As we told JRF, part of the explanation is that the OECD data for Japanis (in contrast to that for other countries) based on income before tax.  However, thanks to a grant from the DaiwaAnglo-Japanese Foundation there has now been a thorough analysis of theoriginal Japanese income inequality data which confirms that Japan is indeedone of the more equal countries.  Areport on this work (Ballas D, Dorling D, Nakaya T, Tunstall H, Hanaoka K.Social cohesion in Britain and Japan: a comparative study of two islandeconomies) will be available shortly on The Equality Trust web site at: 

Second, the JRF report suggests that property crime isconspicuously absent from The Spirit Level. It was not included  because wewere unaware of internationally comparable data on property crime.  However, a literature search has now found anumber of papers published in peer reviewed academic journals which look atchanges in property crime and changes in inequality.  A review of these studies has just beensubmitted to a peer-reviewed journal.  Itconcludes that increases in inequality do indeed have a very substantial impacton crime (see: Rufrancos H, Pickett K, Wilkinson R, Income inequality andcrime: a review and explanation of the time-series evidence. Publicationpending)

Third, the JRF Report discusses the range of countries weinclude and ends by saying “Further research could be carried out on a widerrange of countries…”  Of course italready has.  In 2006 we reviewed 168analyses published in peer-reviewed journals of the relation between incomeinequality and health. These covered many different groups of countriesincluding developing countries (see: Wilkinson RG, Pickett KE. Incomeinequality and health: a review and explanation of the evidence. Social Scienceand Medicine 2006; 62: 1768-84.) 

There are a number of other significant errors in the JRFreport where crucial research in the peer-reviewed literature has been missedor issues have not been thought through sufficiently carefully.   We believe that two factors handicapped theproduction of this report: first, the literature is spread over journals ofepidemiology, public health, medicine, neurology, primatology and statisticswhich are often unfamiliar to those in social policy circles, and second, theJRF Advisory Group lacked proponents (but not opponents) of our thesis who knewthe literature well.

Ballas D, Dorling D, Nakaya T, Tunstall H, Hanaoka K. Socialcohesion in Britain and Japan: a comparative study of two island economies. Tobe made available shortly on The Equality Trust web site at:http://www.equalitytrust.org.uk/resources/other/response-to-questions
Daly M, Wilson M, Vasdev S. Income inequality and homiciderates in Canada and the United States. Canadian Journal of Criminology 2001;43: 219-36.
Fajnzylber P, Lederman D, Loayza N. Inequality and violent crime. The Journal of Lawand Economics 2002; 45 (1): 1-40.
Hsieh CC, Pugh MD. Poverty, income inequality, and violent crime: a meta-analysis of recentaggregate data studies. Criminal Justice Review 1993; 18: 182-202..
Marmot MG. The Status Syndrome. Bloomsbury 2004 Rufrancos H,Pickett K, Wilkinson R, Income inequality and crime: a review and explanationof the time-series evidence. (Publication pending) Wilkinson RG, Pickett KE. Incomeinequality and health: a review and explanation of the evidence. Social Scienceand Medicine 2006; 62: 1768-84."