9.4 The Expectation Propagation algorithm
10.1.1 Tipping dynamics and self-segregation
In relation to peer preferences, we found that both white and Asian parentsusually pre- ferred to send their children to schools with a higher proportion of their own ethnicity on the rolls. We characterised such mutual and symmetric preferences for one’s own group as
horizontal peer preferences, in contrast tovertical peer preferences where the preferences of one group determine sorting. Further to this, when we allowed peer preferences to be non-linear, we found that preferences were consistent with an aversion to sending one’s child to a school overwhelmingly dominated by one’s own ethnic group. This pattern
of revealed preferences was robust across markets, and preference weightings for white and Asian families were of a similar magnitude, although those for Asian families were slightly more non-linear. We also expected to find that the specifics of each place made a difference, especially for Oldham whose recent history suggests much more consciousness of ethnic difference than the other two towns. However, what we found was, if anything, more consistent with an underlying geographic homogeneity to ethnic peer preferences, as if we had tapped into “deep” parameters of ethnic homophily. However, the similarity of estimates does not, on its own, provide firm evidence to support this assertion. It is possible that the three towns studied were not sufficiently dissimilar to reveal geographic diversity in ethnic peer preferences.
Finding that peer preferences tend to be reversed as one’s own ethnic group approaches dominance would seem to be a positive result. It suggests that parents do not want extreme ethnic segregation – that they would rather their children went to an ethnically- mixed school as long as their ethnic group constitutes the majority. This kind of ethnic preference sounds moderate, at least more moderate than an absolute aversion to ethnic mixing. However, models of tipping in sorting dynamics suggest that in the long run, even moderate preferences for ethnic in-groups can produce complete segregation. In Chapter 7 we populated the parameters of sorting models for Oldham with our esti- mated random utility parameters. Oldham is of special interest both because of its recent history of ethnic unrest, and because of the structural changes in the period 2010–2012 that merged several segregated schools into fewer, larger schools with ethnically-mixed populations, thereby creating a kind of natural experiment in mixing and ethnic prefer- ences. Waterhead Academy, in particular, received national media attention (eg. Nye, 2011) as part of a bold strategy to replace an almost-completely white school (Counthill School) and an almost-completely Asian school (Breeze Hill School) with an integrated academy. The school was used as a case study for theories of inter-group attitudes by Ramiah et al. (2015), who found that outside lessons pupils in the new school have tended to avoid contact with members of the other ethnic group. However, some (un- published) evidence has been reported, suggesting that through daily contact, pupils’ attitudes towards the other group were gradually improving (Nye, 2011).
ter 7 suggest, unfortunately, that the integration of the school population may not be sustainable given current parental preferences; both the equilibrium model and the com- putational model predicted increasing (re-)segregation at Waterhead over time. Whereas the equilibrium model predicted that the school’s population would become almost com- pletely segregated over time, the computational model implied that the proportions would become more uneven over many years, but would stop short of complete segregation. The problem, in Oldham and in the other towns we have studied, is that the preferences of each ethnic group for ethnic mix arenot realisticwith regards to the local context. Each group of parents would like asmall proportion of the other ethnic group in their child’s school; a proportion that does not reflect the actual proportion of the other group that is likely to wish to attend their local school given local demographics. This may be explained partly by people’s tendency to over-estimate the proportion of ethnic groups other than their own in a given setting (Alba et al., 2005). It is this mismatch between feasible intakes and desired intakes that sets up the sorting dynamics that result in additional post-residential sorting. In this respect, our emphasis on parental choices actually seems more pessimistic than a focus on constraints, since at least where institutional constraints, such as unfair admissions practices, are present they can be removed. Admissions policies, house prices, school capacity and other constraints undoubtedly do play a role in sorting. However, where preferences for sorting are mutual, rather than one-sided, constraints are not necessary for sorting to arise.
This begs the question of whether the only way to achieve ethnically-integrated schools is to give up on choice and impose mandatory desegregation in a manner similar to the US in the 1960’s and 1970’s. This need not involve busing children to distant schools; it would be sufficient to build schools near the boundaries of currently segregated areas, and to “gerrymander” the catchment areas of each school to achieve ethnic balance. However, the experience of the US suggests that, apart from being undesirable from a parental sovereignty point of view, such desegregation may not be sustainable as in the medium to long run, parents would just subsume school choice (and any ethnic peer preferences) within residential choice and move to new areas. So-called “white flight” is a well-documented phenomenon as a result of desegregation in the US (Clotfelter et al., 2006).
In the light of the seeming intractability of self-segregation, in the US there has been a recent interest in policies to promote diversity within a quasi-market system. For ex- ample the US Department of Justice has published guidance establishing that schools are permitted to consider ethnicity in admissions criteria to achieve a more diverse stu- dent body (Erdil and Kumano, 2014). However, as well as moral and legal issues, the idea of two-sided markets incorporating “ethnic quotas” creates difficult theoretical prob- lems requiring innovations in the mechanism design literature (Erdil and Kumano, 2014; Echenique and Yenmez, 2015).
It is unlikely that solutions to ethnic segregation involving the removal of choice, or even ethnic quotas on admissions, would be socially or politically acceptable in the UK. Nevertheless, the EU Referendum and other events in 2016 have brought issues around social cohesion, integration and multiculturalism to the fore again, as they were in 2001. Perhaps the only policy to tackle (self-)segregation in English schools that would be effective in the long run, would be to change parents’ perceptions about the benefits of attending an ethnically-diverse school. Ted Cantle, whose report in 2001 responded to the race riots in Oldham and elsewhere, proposes what he calls “salesmanship” (Asthana and Parveen, 2016) on the part of policy-makers, to promote the benefits of integration:
Politicians and policymakers need to encourage white British residents to remain in diverse areas; to choose, rather than avoid, diverse areas when they do relocate, encouraging similar choices with respect to placing pupils in diverse schools. (Cantle and Kaufmann, 2016)
However, changing people’s perceptions is not a job for which politicians, policy-makers or civil servants have an abundance of tools. Ironically, it may be that education, the site of the problem, is also the only way in which the problem can be solved.