The determinants of socio-economic segregation between schools
Stephen Gorard and John Fitz
School of Social Sciences
21 Senghennydd Road
Cardiff University
CF2 4YG
01222-875113
email:
Paper presented at the British Educational Research Association Conference, University of Sussex, at Brighton, 2-5 September, 1999.
Freedom is the freedom to say 'two plus two equals four'. From that, all else follows.
The 'politicians error'
Imagine a society in which no women but one per cent of the male population were MPs. We would all realise that this was a significant social fact, and that the gendered difference was a large one. Imagine another society in which 75% of men and 76% of women were in paid employment. While we might comment on the gendered difference here, and watch for signs of consistency or trend over time, we would not accord this second social fact the same significance as the first one. Obviously. So why am I saying this? It is worth reiterating this simple proposition occasionally lest we allow our beliefs and fears to overcome our common sense. We cannot allow commentators to declare that the difference in both examples is one percentage point and therefore the same. Differences in terms of percentage points are not proportionate (Gorard 1999a). In the first society the proportionate difference between men and women could be expressed as (1-0)/(1+0) or 100%. In the second society the proportionate difference between men and women could be expressed as (76-75)/(76+75) or 0.6 of one per cent. This is the method used in the Equal Opportunities Commission study of gendered attainment (Arnot et al. 1996), and it provides a calculational basis, and a numeric value, for our initial common sense that the first difference is large while the second is tiny. Variations on this method are used by urban geographers to measure residential ethnic segregation (e.g. Lieberson 1981), and by sociologists to measure social class mobility over time (e.g. Marshall et al. 1997). Despite the variations, all such methods are 'multiplicative', and can be derived from each other, and all give the same relative results (Gorard 1999b). Since the index wars of the 1950s and 1960s, it is only in education that we now see an 'additive' method (or more strictly difference between two percentages) in common use by journalists, politicians and even some academic researchers. Recent examples have included the absurd notion of growing gaps between sectors of education or groups of students, the only evidence for which is of the kind that we started with - the supposed equivalence of percentage point difference regardless of the size of the base figures from which they derive. See for example Mooney (1999), Russell (1999), or Woodhead (1999). When viewed proportionately the nature of these 'gaps' changes considerably (e.g. Gorard et al. 1999).
The story so far
As you will, I hope, have become increasingly aware, our ESRC-funded study of school-based segregation has already produced several surprising results. Our findings so far are in print or in press so I shall not go into great detail here (see Gorard and Fitz 1998a, 1998b, 1999 and Gorard 1998a). In summary, and (I repeat) to our great surprise, socio-economic segregation between the schools of England and Wales in terms of student intake has been decreasing since 1989. Our dataset is based on school-level data for all schools in England and Wales, encompassing eight million students in 23,000 schools over ten years. There is little possibility of sampling error or bias. Our methods are those described above, as used for 40 years by urban geographers and sociologists, and whichever variant of the method we use the results are the same. Segregation has declined significantly in every region, in both primary and secondary schools, and as assessed by student poverty, first language, ethnicity and statement of special educational need.
We are continuing our analysis with the alternative indicators at lower levels of aggregation (we have not, for example, analysed the primary sector at LEA or school level with any indicators, nor the secondary sector at school level using ethnicity and first language). Perhaps the most important next step for the numeric phase will be the addition of school outcome scores to the dataset, enabling us to consider further the relationship between segregation and educational 'performance' (as started in Gorard 1998b, 1998c, 1998d). Ironically, our expectation for these further analyses is now reversed and we would be both surprised and excited to find evidence of overall increasing segregation on any indicator. In fact, a part of us hopes to find such evidence since this is clearly what most commentators want (i.e. that schools are becoming more socially segregated over time). People might stop looking for the specks in our eyes, and we might even get to present our findings at BERA in a session with other researchers in the same field (even though our findings are dissonant to theirs).
We have found interesting variations in patterns of segregation at LEA, district and school level which are worthy of further investigation. Schools in the vast majority of LEAs in England and Wales (93 out of 122) are becoming less segregated over time (e.g. Islington, Southwark). These are generally areas of higher population density (perhaps where movement between schools is possible), in which there is little diversity (few selective, Welsh-medium, GM or fee-paying schools), which have faced school closures since 1989, and where indicators of poverty have increased substantially (leading perhaps to what we have termed 'equality of poverty'). Market theorists have therefore been shown to be wrong about the relationship between improvement, choice and diversity in terms of increasing social justice. It is also the case that areas with a high level of appeals against school allocation (a proxy measure of 'alert clients') show no consistent pattern of changes in segregation. Market theorists have therefore also been shown to be wrong about the relationship between active choice and social justice (we await with interest our test of the notion that areas with active choice will show greater improvement in school outcomes). Areas of high diversity or low population density have generally shown a far smaller decrease in segregation over time, perhaps because they are less susceptible to any policy change (e.g. Clwyd, Dorset). Intriguingly, the six areas with a significant increase in segregation over 10 years are very similar at first sight to those with a large change in the other direction (e.g. Haringey, Hammersmith and Fulham). The next major phase of our study is detailed fieldwork in the areas of interest identified by the first phase. This will involve a close examination of the school allocation process in areas currently on different trajectories of change. Hopefully the first results from this stage will be available next year.
But....
As already implied, the relatively good news that schools are becoming more mixed over time has met academic opposition (but general acclaim from schools). This opposition is often of a creative and ingenious nature, especially considering that this is the only study of its kind so far (but see emerging evidence from Bradley et al. 1999).
Some of the prior indirect evidence for an increase in segregation stems from studies of school choice which may have methodological or sampling problems (e.g. Gorard 1997). Some of the early results can be explained by our finding of a temporary general increase in segregation in many areas in 1990/91, which we have termed the 'starting-gun effect'. Some of the evidence presented for the segregation thesis is simply nonsense (e.g. Ambler 1997). Some of the evidence actually shows the opposite of what is claimed (e.g. Lauder et al. 1999). Some is based on Geographical Information Systems which we have already shown to be very poor indicators of school and catchment area composition (see Gorard 1998c). For example two schools we examined in South Wales have identical catchment areas. They therefore present the same data for GIS. One teaches through the medium of Welsh, and the other English. The first has 11% of children from families on Income Support. The second has 27%. GIS cannot distinguish between them
In retrospect, it is surprising how weak the evidence is that segregation in schools was increasing over time, and obvious that the major changes in educational policy since 1944 had at least the intention of providing greater social justice in schools. Some of the resistance to this idea may come from a personal investment by researchers in the segregation thesis, some may come from a confusion of 'decreasing' with 'low', and some seems to stem from a non-empirical belief in sociological predictions of the impact of markets. Some avowedly qualitative ('I don't do numbers') researchers simply may not understand measures of segregation.
Seven substantive empirical objections to our findings have emerged so far. Five of these have already been dealt with in existing published material (see in addition Gorard and Fitz 2000, Gorard 2000):
•Despite using data from six LEAs in Wales for the pilot study, this is not a local phenomenon. We have used all schools in England.
•It does not matter at what level of aggregation the analysis is performed (and we have used five levels from national to school). If segregation was increasing at any level, it would be picked up by our analysis at a higher level unless there was an equal and opposite trend towards desegregation. Although there are schools, years, and districts in which segregation increases, the overall trend is downwards.
•In the same way, the decrease in segregation is not being achieved at the expense of a small number of schools in a 'spiral of decline'. Analysis at the school level reveals that even in areas with an increase in segregation overall, this is spread between several schools. The most disadvantaged schools in each LEA are actually moving closer to their fair share of indicators of disadvantage. Even where these indicators are increasing over time in one school, they are not increasing as fast as the indicators in society as a whole (and this is one further reason why small-scale studies of segregation are unlikely to succeed).
•The finding is not a peculiarity of either take-up or eligibility for free schools meals (the most complete indicators in terms of record-keeping). We have now used five indicators of educational 'disadvantage', and each confirms the picture from the others.
The possibility has been raised that despite this triangulation, the indicators used are not sensitive enough to distinguish the segregation that must surely be taking place between schools (according to the crisis 'disciples'). Having been surprised that schools are becoming less segregated by poverty, class, ethnicity, first language and special need, these commentators now posit a set of differences that lie beneath the surface of these grosser indicators. Elmore and Fuller (1996), like Willms and Echols (1992) before them, present evidence that choosers (those who do not accept allocation to their nearest school) differ systematically from non-choosers. A common observation is that single-mothers from poor families in voucher schemes in the US or the (defunct) Assisted Places Scheme in Britain were more frequently better educated than mothers in equivalently poor families not using the schemes. A potential explanation is that they are part of a growing 'artificially poor' (after Edwards et al. 1989) who have become single through death or divorce. Even if this were true, and it may well be, this does not have to lead to segregation (unless it is a more stratifying process than that of allocation to residence). In support of their notion of segregation by stealth, both Elmore and Fuller (1996) and Witte (1998) cite secondary evidence from Wells (1996). This claim, based on a paper by Wells, has had a large impact and has propagated through the research literature as a social science fact. It is therefore of some interest to consider the nature of this evidence.
Wells refers to the results of an apparently large study by Lissitz, but this study is unpublished and is therefore not open to scrutiny. Her own primary evidence is based on 37 children (and the parents of some of these), and she compared the characteristics for families of Black students choosing to stay in a local city school, transfer to a school in the suburbs, or transfer but then either return to the city school or drop out. She concludes, for example, that 'city students who stay behind in the urban schools... tend to be more disadvantaged in terms of parental education and employment than the transfer students'. Her evidence for this statement in respect of parental employment is reproduced below (Table 1). The first noticeable thing is that by ignoring the drop outs and returnees, Wells now has only 24 cases on which to base her findings. The main figures are those observed, while the figures in brackets are the expected values for each cell, using the standard method for a cross-tabulation (see Siegel 1956). For example if the proportion of families in employment and the proportions using each type of school are held constant, she would expect 9.5 parents of those in City schools to be employed. She actually observed eight. The difference is not significant, statistically or in any other sense of the word. Yet this is the primary data on which the notion of sub-indicator polarisation rests, and which has been suggested to us in all seriousness as undermining our findings from 8 million students over ten years.
Table 1 - Parents of city and transfer students (figures from Wells 1996)
City / TransferParents unemployed / 4 (2.5) / 1 (2.5)
Parents employed / 8 (9.5) / 11 (9.5)
The purpose of the remainder of this paper is to discuss our initial response to the two latest objections to our findings. Despite the initial surprise at our results, we hope and believe that social scientists will begin to start discussing the issues they raise as much as the methods and data underlying them. Whilst always willing to discuss methodological points, we feel that the time has come to grapple with the more substantive implications of our study (and forget the fear that our results might be used by others to defend the use of markets in education). Hopefully, by dealing with the two most recent and elegant attempts to preserve the 'crisis account', we can help overcome the fear and so start our own rehabilitation process, for it is surely only once we have established what is actually happening in our schools that we can hope to proceed in a cumulative manner to an understanding of how and why. We are prepared to grow up in public on this issue. Are others?
Examinations bite back
The first and simplest objection to discuss is exemplified by the work of Gibson and Asthana (1999). Some commentators have observed an increasing polarisation of outcomes between the best and worst schools, and used this to argue back that therefore schools are likely to be becoming more polarised by intake as well. For example, Gibson and Asthana (1999) quite reasonably claim that if socio-economic segregation between schools was decreasing then one would expect the patterns of attainment linked to socio-economic factors to be converging also. In fact they go so far as to claim that 'if schools with poor GCSE results... [and associated SES measures] can be shown to be improving their performance and social composition relative to those with good GCSE results... then the polarisation thesis will have to be dismissed' (p.14). There is of course not much likelihood of this in their opinion since they claim in the same article that the actual results of schools are tending to widen the gap between the best and the worst (Table 2).
For example, Gibson and Asthana (1999) claim that the gap in terms of GCSE performance between the top 10% and the bottom 10% of English schools has grown significantly from 1992 to 1998. Table 2 shows the proportion of students attaining five or more GCSEs at grades C or above (the official Key Stage benchmark), for both the best and worst attaining schools in their sample from England. It is clear that the top 10% of schools has increased its benchmark by a larger number of percentage points than the bottom 10%. On the basis of these calculations they conclude that schools are becoming more socially segregated over time. In fact, they comment 'within local markets, the evidence is clear that high-performing schools both improve their GCSE performance fastest and draw to themselves the most socially-advantaged pupils' (in Budge 1999, p.3).
Table 2 - Changes in GCSE benchmark by decile
Decile / 1992 / 1994 / 1998 / Gain 94-98 / Gain 92-98Top 10% / 60.0 / 65.0 / 71.0 / 6.0 / 11.0
Nine / 48.2 / 54.0 / 59.1 / 5.2 / 10.9
Eight / 42.0 / 47.7 / 53.1 / 5.4 / 13.1
Seven / 37.1 / 42.5 / 47.2 / 4.7 / 10.1
Six / 32.8 / 37.5 / 41.6 / 4.1 / 8.8
Five / 28.2 / 32.9 / 37.2 / 4.3 / 9.0
Four / 23.8 / 28.4 / 32.2 / 3.8 / 8.4
Three / 19.3 / 23.7 / 26.8 / 3.1 / 7.5
Two / 14.7 / 18.2 / 21.4 / 3.1 / 6.7
Bottom 10% / 7.9 / 10.6 / 13.1 / 2.5 / 5.2
This is the moment to recall the example of the two societies with which this paper opened. Gibson and Asthana use differences in GCSE scores from 1992 and 1994 to 1998 for each decile of schools (ranked by GCSE results). These differences are expressed using the symbol '%'. For example a change from a benchmark of 60 to 71 is expressed by them as 11%. This appears to mislead them into thinking that the differences are expressed as percentages rather than as percentage points. If percentage points were a form of 'common currency', and in themselves proportionate figures, then perhaps no more complex analysis would be necessary. However, to accept this would lead to a paradox in achievement gap research (and lead to the conclusion that the differences in the two imaginary societies are identical), since the standard method used in reputable research calculates the change over time in proportion to the figures that are changing.