A popular statistical technique used to analyse racial disparities risks masking more than it reveals, argues Alex Stevens
It is to the Prime Minister’s credit that she continues her interest in ethnic inequalities in the UK.
Her latest demand that organisations ‘explain or change’ inequalities in outcomes between ethnic groups follows a long line of reports, going back at least as far as the 1992 report by Roger Hood, Race and Sentencing, that suggested worse outcomes for Britons of black, Asian or minority ethnicity.
Mrs May’s call comes hot on the heels of David Lammy’s recommendation that if organisations cannot explain ethnic disparities, they must change the way they work in order to eliminate them.
It is way past time that a British Prime Minister took serious action to address enduring inequalities. And Lammy’s proposal offers a way of breaking the organisational inertia that has held back progress.
But by offering these organisations a choice – either to explain or to change – it gives them a potentially dangerous way out. It is much easier and cheaper for organisations to commission studies of ethnic disparities than to take effective action to reduce them.
Many such studies will use multivariate regression techniques. These are statistical procedures that promise to identify the ‘independent effect’ of one factor, while taking into account the predictive influence of other influences on the outcome. Such techniques commonly find that the apparent disparities between ethnic groups are statistically reduced when you take into account factors such as employment, education or location.
A good example is the study, ‘Racial stratification and multiple outcomes in police stop and search’, by Barack Ariel and Justice Tankebe. This shows that, in at least one police service area, people recorded as having a black ethnicity are more likely than white people to be arrested after a police stop and search. They are less likely to be given a cannabis warning or informal advice.
However, in a multivariate analysis the predictive effect of ethnicity is less than that of other factors, including gender of the suspect, whether they were previously known to the police, and whether the stop took place in an area considered a crime ‘hotspot’. Does this mean that this police service could say, ‘we have explained why more black people get arrested after stop and search than white, and so will take no further action’? Here is the danger of relying on multivariate statistics for the explanation of inequalities.
It will usually be possible to find factors that reduce the predictive influence of ethnicity on outcomes. In Ariel and Tankebe's study, for example, being male, being known to the police and being in a crime ‘hotspot’ seem to lessen the effect of being black. But in reality – in life as it is lived by people outside of statistical models – there is no independent effect of being black. It is experienced alongside all the overlapping inequalities that constitute race.
If underemployment, educational exclusion and residential segregation are concentrated on people of black ethnicities, then they will be more likely to live in crime ‘hotspots’. If the police focus their searches for drugs on young black men, and they do, then being male and previously known to the police is not ‘independent’ from the effect of ethnicity in police decision-making.
Another serious danger is that such explanations will dramatically contradict the lived experience of people who suffer from these inequalities. If someone sees that he and friends who share his ethnicity are more regularly targeted by the police than their white peers, it is no use telling him that this is independent of their ethnicity. It is experienced as a form of racism, as qualitative research with young black men shows.
If organisations are truly to explain disparities, they will need to go beyond multivariate analyses to engage in direct conversations with people of colour, to find out not only how these disparities are experienced, but also how they can be changed.
Credit for the points made in this piece should go to Ben Bowling and Coretta Phillips, who developed them in their 2002 book on Racism, Crime and Justice, themselves citing the work of Teresa Smith and Susan J Smith. See also the article by Anita Kalunta-Crumpton on ‘The importance of qualitative research in understanding the disproportionate black presence in crime figures in the United Kingdom’
Alex Stevens is Professor of Criminal Justice at the University of Kent