Notes on:  David Gillborn, Paul Warmington & Sean Demack (2018) QuantCrit: education, policy, ‘Big Data’ and principles for a critical race theory of statistics, Race Ethnicity and Education, 21:2, 158-179, DOI: 10.1080/13613324.2017.1377417 Dave Harris Qualitative research is often seen as subjective, but quantitative research is also socially constructed and contains hidden assumptions. These 'encode racist perspectives'. We need CRT to guide us in future use and analysis — 'QuantCrit' – which takes racism as central, although not easily quantified, suspects numbers as promoting deficit analyses, insists on the critical evaluation of categories, stresses the need for 'voice and insight and the 'experiential knowledge of marginalised groups', and explores 'the role in struggles for social justice' of quantitative analyses (158). As an example, the computer program to screen applicants for places at St George's Hospital medical's school unfairly discriminated against women and people with non-European sounding names, it was found by the CRE, and similar findings were est with an American program deciding whether criminal defendants were likely to reoffend or not. In both cases, the algorithms were not '"colourblind"' and mistakes were made in estimates of risk in both directions. Findings were greeted with amazement that computer calculations could make 'such gross and racially patterned errors', with such effects — the American program so that black defendants were '"77 percent more likely to be assigned higher risk scores"' (159). However, such patterns are predictable because human labour is involved together with biases. If anything, there is an added danger of assumed objectivity. We need to think critically about race in equity and how it is routinely embedded. The authors acknowledge their positionality, and their common ground in the use of CRT, based in their personal experiences and their concerns as 'educators/activists' (160) and their frustration with colourblind policy. They further understand that race 'may be "unreal" as a scientific category [but] it's "modes of existence" are real and have an [sic] innumerable material and social consequences', citing Leonardo 2005). This rules out race as a mere 'technology' of other relationships such as social class. [This article is a contribution to a special issue of the journal]. Different methods are appropriate for different aspects. Qualitative approaches are best to understand the nuances of the social processes that affect race in equity, but quantitative methods are better at charting 'the wider structures, within which individuals live their everyday experiences' and the 'structural barriers and inequalities that differently racialised groups must navigate'. At the same time, statistics are often used to camouflage or even legitimated racist inequities, which appear as neutral or objective and these must be examined. First it is important to look at how numbers disguise racism in education and defend white supremacy. Here, quantitative data can assume a special status as proof, 'quantitative hyperbole' associated with big data, where numbers apparently '"speak for themselves"' [and they reference Anderson 2008]. Secondly they show how there is potential to contribute to a project for greater equity drawing on classic work on CRT on the key principles to guide further quantitative research. There is often an assumption that quantitative material is different from and superior to qualitative data, that it reports the facts authoritatively, and neutrally and objectively. That quantitative data are used as a part of technology to reshape educational systems, '"policy as numbers"' (161), especially in neoliberalism and the audit society. Priorities are justified, successes and failures are defended, national testing programs, and school performance tables show how numbers are used to evaluate schools. The same goes for PISA, run by the OECD. There is rarer comment about the small samples, and the selective coverage of tests [usually only reading, maths and science] nor the difference in assessments [for Pisa]. Academics know these limits, but politicians cite them in their drive for '"accountability"' (162). 'Wider structural inequities that shape educational outcomes are ignored', such as spending per pupil. There is no discussion of the validity of the measures, the scope for cheating, including teachers altering students work, or schools removing low attaining students from the roll in advance of the test date [in England, according to Harding 2015]. In the US, similar fixing of student lists and other '"creative bookkeeping"' has occurred. UK government puts disclosure of numbers as a key part of its claims to establish transparency, which assumes the public can use statistics to understand and challenge especially public authorities. This assumes that the relevant and useful data will be made available and unaffected by the selective capacities of the authorities themselves. It assumes knowledge time and resources to access and analyse the data, and that the authorities will be affected by these analysis. At a deeper level, numbers are no more neutral or factual than any other form of data. They are socially constructed 'exactly the same way that interview data and survey terms are constructed' (163), through decisions about which issues should be research, which questions asked, how information is to be analysed and how findings should be written up [none of this applies to the favoured form of 'research' in CRT, of course, the personal story]. There is always selection of data and reporting of it. This was seen in the controversy about white British pupils and their apparent underrepresentation in universities, as reported in the. Press. The Inst for Fiscal Studies published government figures. The low rate for white students was the matter of interest for all news outlets. This was represented as news even though it was long existing [they cite a report from 1997 that white students were underrepresented in HTE compared to ethnic minorities]. There point is that access statistics gives 'an extremely partial, indeed biased view of race and higher education in Britain… [By]… Simply looking at who goes to university'. What is ignored is race inequities in the status of universities and the level of final degree achievement. If we look at Russell group universities, white and minoritised students seem to have 'roughly similar chances of attending' (164). This refers to a category of BME, however, and disaggregation reveals differences between, Indian and Black Caribbean students, for example, here White British students are 'almost 5 times more likely to gain access' than black Caribs. When we look at different classes of degree, white students are 'more likely to gain a first than any other group… Black students are least likely to be awarded first class degrees' [22% vs 9% — pretty old data I would have thought]. The debate in the media ignores these issues and thus helps paint 'white people as race victims in contemporary Britain' (165). It ignores past race inequities which have produced these are the differences — these are remarkable in that the group least likely to attend are most likely to do well [a weird assumption themselves here, that it is the same sort of white students in both cases]. Big data may now be too big for anyone except big tech, or for government [and there is a hint of the moral panic about AI]. If numbers speak for themselves, there is no need for theories of human behaviour, and numbers really will speak for themselves. The warnings about correlation and causation have been simply ignored. The analysis becomes something for machines, too difficult for human beings. But algorithms are not free from bias [well the old ones are not — maybe the really sophisticated ones will be?) All data is produced by human beings [same objection]. At the moment the role of theories in generating data are admitted [so are costs]. Can we develop QuantCrit? There has long been questioning the assumptions of mainstream science and rationality in CRT. CRT is now accepted as an important approach and has grown rapidly, especially in education [usual references]. It has spread into things like LatCrit, or DisCrit [I didn't think Gillborn entirely approved this]. Some people have already used quantitative data in CRT to examine the permanence of racism and the problems of liberalism [references 169] but more central tenets can be criticised in a more holistic view. They can build on some of the earlier criticisms made by Gillborn. They do not see this as an offshoot what toolkit. They do not think that numbers than ever capture the material impact of intersectional racism or affect change in policy more powerfully. Nevertheless, they can develop some first principles. The first one is the centrality of racism. Racism is 'complex, fluid and changing' and is not obvious (169), so simple statistical measures are unsuitable [and you can argue that just about any form of behaviour is racist]. Races not just a variable but a political statement. We need to look at power relations and how race is constructed by social relations [all the ambiguities of social construction are raised here, but they have been denied earlier with the remarks on class?]. So measurements can only be crude approximations. We are better in just criticising other people's statistical material. The second one is that numbers are not neutral, and data gathering reflects interests assumptions and perceptions 'of white elites' and this needs to be challenged. We can see examples in the growth of deficit theories, for example, by discussion that ignore racism is a central factor, and by arguments which lend '"objective" support to Eurocentric and white supremacist ideas' (170). We need to look behind the numbers to identify racist logics, for example the way race inequity has been normalised, for example the way there is an expectation of lower achievement by Black Caribbean students. This actually led to the idea of value added, but this had an unintended effect in condoning in effect the lack of achievement by Black students — they could not be expected to make the same achievements because of their social disadvantages. It was a 'normalisation of lower racialised attainment [in] official analyses' [in the USA I think] (171) [compare with Reilly on positive discrimination in university entrance]. It would be helpful in redefining the notion of underachievement as achievement falling below the forecast. If Black students do as badly as predicted, they would no longer be underachieving [an equally Conservative conclusion?]. Relative achievements would be recognised but this also threatens to 'enshrine the lower average achievements of some groups as normal, and even inevitable', involving '"disempowering fatalism" quotes'. The whole area shows that commentators can mobilise statistics to offer different definitions of educational achievement, that these are far from true or legitimate understanding [you still haven't solved the dilemma though, ya prats! You have rightly said that all stats are value laden, but some interpretation has to be offered of black kids underachievement. If we call it underachievement we enshrine low expectations, if we call it different achievement we normalise differences. It's not really a measurement problem is it?] The third one is that categories or groups are not natural: 'for "race" read "racism" [even when you use the term?] The problem is that contested terms are normalised and used in labelling and controlling. Race is associated with unequal outcomes and this indicates the operation of racism, and can lead to seeing race as a cause in its own right, as indicating some inherent efficiency. This can haunt research design. Many studies do not include race or ethnicity at all which can suggest that race and racism is unimportant. If race is included, the constructs can be treated differently and assumed to mean real things, biological or natural facts. Everything depends on how ideas are operationalised. We saw this with comparisons between White students and BME composites [white students is a composite too of course]. Sometimes analysts have compared white students with everyone else is an even worse composite. However the issue of categories is 'no easy matter' and 'using too many categories can be almost as bad' [!] (172). They came across one school using 70 separate ethnic categories, resulting in very small cell sizes. [The results were no significant differences between the ethnic groups, but the research team got all sophisticated and suspected that the sizes were too small to have statistical confidence]. Some categories seem to indicate a pre-existing fixed quality, especially at deficit, so identifying racial groups becomes immediate 'fodder for racist practices and beliefs': while a correlation between race and underachievement will mean to us CRT theorist that racism is present as a significant factor, to white racists it will mean that black people are less able to achieve. We can guard against this by replacing the term race with the couplet race/racism to avoid the threat of racist assumptions [they cite Omi and Winant here]. Much academic research sees race as something with a pre-existing validity explaining inequality, including the British government — hence contradictions like recognising disadvantage while increasing disciplinary regimes, or slipping between associations and causes. The fourth principle is that data cannot speak for itself. That is why we must place importance on experiential knowledge, with POC and other outsider groups [the usual ones — the usual problem is what if they contradict] [these experiential knowledge accounts are apparently not open for racialised assumptions to come into play, and they do make untested planes to objectivity, and describe the real world and so on. Regression analysis makes all sorts of assumptions about reality and claims to control for separate influences of numerous factors, but racism does not operate separately through factors [experiential knowledge cannot even see a range of factors beyond personal experience, admitted earlier]. The fifth one is social justice/equity orientation. Statistical research is not value free and politically neutral. CRT scholarship is on the side of the good guys supporting social justice goals and working to achieve equity 'by critiquing official analyses' and 'working with minoritised communities and activist groups' to provide more useful research. They should not abandon quantitative research but adopt a 'position of principled ambivalence'. This is the stance they take already with scientific claims about race [they like the science that denies it]. They should engage with statistics showing how race/racism is constantly made and legitimated but oppose any tendency to develop an unwelcome story. In conclusion, mostly quantitative data is used to challenge equity work, for example in defending the performance of governments or educators. This often involves selected decisions. There is an underlying belief in numbers as objective and factual, superior to testimony, with assumptions hidden. Qualitative data 'may be inherently better suited to exposing and opposing racist social processes' (175). Statistics are still useful though following their five principles, and they hope that quantitative data will feature more strongly in the service of equity goals, especially to criticise existing quantitative efforts.