Navigating ethical dilemma in education data: The story of the EQI

From the Centre for Data Ethics and Innovation (CDEI)

This blog explores practical and thorny questions that will resonate with many. Data driven technology seems to unlock opportunities. Yet it poses new types of risks, from bias and transparency challenges to issues of consent and data sovereignty.

Ministries collect and manage large quantities of administrative data as part of their everyday business. Contemporary data processing tools allow ministries to use that data to deliver services efficiently and effectively. Arguably, given the potential benefits these tools allow – benefits to both service users and to the ministries themselves – ministries are ethically obliged to take advantage of them when they can. 

However, it is clear that the use of these tools also pose significant ethical risks: Human bias, incomplete, or unrepresentative data may lead to biased outcomes; the complexity of some data processing tools makes it difficult to meet obligations of transparency; it is difficult (perhaps impossible) to obtain genuine consent for the use of data supplied to ministries, since service users have often have little choice about supplying it and data sharing arrangements may make it difficult for individuals to identify end-uses at the point of collection; some data processes exclude or limit human involvement; data processing tools are often more accurate than humans, but none are perfect and their errors may be more significant than those of their human counterparts given the scale of data they manage; in Aotearoa New Zealand agencies must meet specific Tiriti and data sovereignty obligations; and so on. This is an illustrative rather than complete list of the ethical risks of contemporary data processing.

In sum, the potential benefits of contemporary data processing tools are too significant to allow ministries to eschew their use, even if they could, but using them raises significant ethical risks.  Consequently, ministries must have appropriate systems in place to identify and manage the risks posed by processes they cannot avoid.

The development and implementation of the Ministry of Education’s Equity Index (EQI) is a useful case study.

Until 2023, the distribution of equity funding to NZ schools relied upon census data to indicate what proportion of a school’s population lived in low socio-economic or poorer communities.  The system divided schools into ten categories or deciles.  Lower decile schools received more equity funding to support their students’ learning needs.

There were significant problems with the Decile System including that restricting the system to ten bands created large steps at decile borders that did not reflect actual differences in equity needs; the system relied on only five-equally weighted socio-economic factors that did not recognise nuanced differences between the circumstances of different whanau, rangitahi, or Tamariki; and since they were based on census data, it was only updated after NZ’s five-yearly census.

In the face of these problems, it was decided to replace the Decile System with what was initially called a Predictive Risk Index which would estimate the number of children in each school and service at greater risk of educational underachievement due to socio-economic disadvantage. The resolve to abolish the decile System and the essential commitment to a more nuanced predictive model survived changes in Government but saw the redeveloping of the Risk Index as an 'Equity Index’.

The EQI is a more nuanced measure of socio-economic disadvantage than the decile system. It is based on the circumstances of individuals rather than of the neighbourhoods they live in and it uses 37 variables to produce a school-level measure of socio-economic disadvantage. The variables are weighted based on their statistical relationship to educational achievement. 

The variables are extracted from anonymised Government administrative data held in the IDI, the Integrated Data Infrastructure. The EQI uses an algorithm to identify correlations between these variables and educational achievement producing scores for individuals which are then aggregated according to school rolls. EQI outputs can be updated annually, rather than five yearly, allowing quicker responses to changes in the needs of schools and avoiding potentially large and so disruptive postcensus recalculations and shifts in decile funding.

Going back to our starting point, then, the EQI applies sophisticated data processing and predictive analytics to data collected by government agencies to deliver benefits to families and students. But it also posed the risks noted above. Its development and implementation required careful to ethical review to identify risks and mitigations.

An early step in that process for the EQI was consideration by the StatsNZ Data Ethics Advisory Group in November 2019. The group was mostly positive. The EQI, they said was, an improvement on the Decile System. The focus on equity, and the aim of more up-to-date and granular data was commended. A concern that there had been limited engagement with the public was partly addressed by subsequent consultation with a Sector Reference Group, including representatives from the Māori medium schooling system, which was said to have had significant impact on the design of the index, including changing how the index calculates academic achievement.

The Ministry of Education also commissioned an ethics review of the EQI from the author of this note, the Ministry’s ethics advisor. That report identified a number of ethical concerns, including the tendency of government data to be deficit data, collected when individuals have ‘negative’ engagements with government agencies; the fitness of IDI data for the EQI purposes; consent; transparency and explanation; Māori data sovereignty; and the possibility that EQI ratings might be misused and misrepresented in much the way that the Decile System scores had been, potentially stigmatising rangitahi and tamariki at schools. 

Each of these risks was accompanied by recommendations which were considered and, where possible, addressed before implementation either by way of mitigation or by an informed decision to accept a risk given the overall benefit of the EQI.