Aadhaar in Andhra: Chandrababu Naidu, Microsoft Have a Plan For Curbing School Dropouts

The massive data-driven governance project brings focus on the efficiency of government schools, data privacy and the role of Aadhaar in technology-driven development.

Aadhaar cards. Credit: PTI

Aadhaar cards. Credit: PTI

New Delhi/Bangalore: Imagine the following. Two weeks after a new school year starts, the teacher of a government school in a tiny Andhra Pradesh town receives a message from her mandal’s education officer, informing her that of the twenty students in her class, six are likely to drop out by the end of the year.

At a meeting later that week with the education officer and the school’s principal, the teacher gets to know that if two of her students don’t score above 75% in English, they may feel that their chances of getting a job after graduating are slim and thus are unlikely to spend more time in formalised schooling.

At the fingertips of this education officer, just a mouse click away, is a wealth of data that will allow her to make better decisions – courtesy of a faceless machine from Microsoft’s high-tech, machine learning lab.

Come this summer season, this futuristic scenario is likely become a reality for a little over 5 million students and 10,000 schools in Andhra Pradesh.

The Chandrababu Naidu-led administration plans to triangulate and collect information from a number of databases including the Aadhaar system, funnel that data into Microsoft’s machine learning platform, use the company’s software to flag at-risk children, and ultimately “monitor” a student as she moves throughout her academic journey.

The state government’s ambitious plan – for which The Wire spoke to representatives from Microsoft, various Andhra Pradesh government officials and activists who work in the field of education – represents not only Big Data-fuelled decision-making in action but also highlights the intimidating potential of the UID project.

The problem that Naidu and Microsoft are looking to solve is both simple on paper and dangerously tragic in reality: various organisations put the dropout rate of students who enter government schools in India from anywhere between 40% (before reaching 8th standard) to 64% (in the case of girl students). Andhra Pradesh, in particular, has a bad reputation on this front, with activists pointing out that seven out of every 10 girl students drop out before they reach 10th standard.

The consequences of this problem have been showing for sometime. Data from the recently released National Sample Survey shows that the literacy rate in both Andhra Pradesh and Telangana is the second lowest in the country after Bihar.  

A more effective grip

The need for any form of algorithmic regulation stems from a lack of control and Andhra Pradesh and its problem of school dropouts are no different. As one Andhra Pradesh government official put it, it’s difficult for the state government to have a pulse on the exact state of its schools and their overall efficiency.

“We get data from our schools, on dropouts, three months down the line. Everything has a time-lag and therefore it’s tough to take decisions and understand what works. A comprehensive look has not been taken at which school locations do better than one another when it comes to dropouts,” the official, who didn’t wish to be identified, says.

Enter Microsoft. Last year, when CEO Satya Nadella was in Hyderabad to meet Naidu, Microsoft and the state government signed a number of agreements to use the software giant’s machine learning software and apply them in the areas of education, agriculture and e-citizen services. According to Microsoft, these solutions would be “built and deployed to address specific problems within each of the fields to achieve better outcomes for the state.”

“What was offered was literally a solution. Azure, our cloud offering, comes with a bunch of high-value services which are effectively part of the overall Cortana Analytics Suite (CAS). CAS has a lot of tools and algorithms… a lot of data-crunching capabilities that can provide a lot of useful insight as long as you pump in the data and the right modelling,” Anil Bhansali, Managing Director of Microsoft India (R&D) told The Wire.

Building the data pipeline

Getting the right data has proven to be a trying, if surmountable obstacle, for Microsoft and the state government. The pilot project that the two parties ran earlier this year, which covered a little over 1,000 schools and used the data of 50,000 students, was restricted to tenth standard students.

Why tenth standard? Bhansali refers it to as one of a few inflexion points, because “that’s [10th standard] when you have one of your first standardised tests and there are a reasonable set of students who drop out from 10th to 11th.” Tenth standard is also, incidentally, where the state government has a great deal of digitised data on its students: the results of the tenth board examination are already online and the state’s education department also has access to the children’s hall-ticket data which has details on gender and subject wise grading.

Microsoft’s CAS works at two different levels: its most basic function serves as a form of business intelligence,which gives the Andhra Pradesh education department simple analytics such as the percentage of girl students that drop out versus boys; students from X region are more likely to pass eight standard than students from Y region and so on.

Its machine learning capabilities however also allow it to carry out more complex, predictive analytics. Given historical data and specific parameters that adequately measure the driving forces behind school dropouts, the software can start extracting patterns and start making predictions on which student is likely to drop out.

The three databases that the Andhra Pradesh government has tapped into, and which was fed into Microsoft’s system, are: the Unified District Information System for Education (U-DISE), educational assessment data that was taken from various sources, and socio-economic information from the UIDAI-Aadhaar system.

These three sources of data correspond to three attributes that the two parties believe contribute greatly towards dropping out: how well the student does academically, the school’s physical infrastructure and teacher skill and experience level, and the student’s socio-economic status.

The U-DISE database — which is a nation-wide database set up by the Ministry of Human Resources Development — contains school infrastructure information (how many functioning toilets etc) and data on teachers and their work experience. It’s a valuable and relatively easy resource to tap into.

Student assessment and enrolment data is perhaps the most difficult as when the project expands to include lower classes, a lot of this still needs to be digitised in a format that Microsoft’s software can consistently tap into. Last but not least, as discussed and analysed below, the Aadhaar database provides the missing piece (socio-economic information) as well as serving as a foundation for the Andhra Pradesh government to keep track of its students.

Click and see the dropout

Microsoft is going on the offensive this time. Credit: The Intercept/Getty Images

Microsoft is going on the offensive this time. Credit: The Intercept/Getty Images

While actual, targeted interventions didn’t happen during the pilot project, the Microsoft system now, according to Bhansali, has an over 90% prediction rate for determining who is at risk of dropping out. This number however may change once the company gets yearly feedback and see if its predictions match the students who drop out.

The end result is a rather simple web portal that comes with a database on student dropout predictions that can be accessed by each mandal’s education officer, who will use that data to hopefully intervene. “Literally the education officer at the mandal level or district level can choose the mandal and a specific school and what they can get is a list of students along with the probability of whether they will drop out,” says Bhansali.

“They [the Andhra Pradesh government] want every school principal to be able to see this model and see that anybody who is at risk of dropout, there is sort of customised counselling and help.”

Some of the initial insight from Microsoft’s software re-affirms long-standing notions of what causes school dropouts — girl students tend not to attend classes if toilets in a school do not work — but other factors such as how well students score in key subjects like Mathematics or English speak to educational theories of how outdated and irrelevant school syllabuses may be holding back students who can’t and don’t follow a traditional schooling route.

“Traditionally, we’ve focused a lot into infrastructure, which is of course important. But for students to continue, they also need to perform well. What we see from the data is of course that students from English-medium schools tend to drop out less than local language schools. But also that if students score less in Mathematics and English they feel that their chance of getting a white collar job after twelfth standard or getting into a good college is less and so they stop coming to school,” said one Andhra Pradesh education department official.

Algorithmic governance

The idea behind algorithmic governance or regulation is simple: it allows the government to do away with its previously inefficient and bureaucratic methods of deciding where to allocate resources, how to judge institutions based on their past and future performances, and how to assess the efficiency of a recent law or regulation.

Consequently, dropouts may end up being only the tip of the iceberg. As Bhansali points out, the same set of “very rich data” could be used to analyse a number of other issues, both within the field of education and outside it.

“Now, the set of data we have… we could and are using it to answer the question of who will drop out and who will not. But I could [also] in theory, and we are working with the education department… actually figure out which teachers are being more efficient,” he says.

In this context, the data pipelines that have been set up could be used to judge the efficiency of various government schools and divert resources and attention to weaker performers while rewarding and learning lessons from schools that do a better job at educating its students and curbing dropouts.

It could also be used to address the weaknesses of the average government school teacher. While the current project is already aimed at doing so, by helping predict dropouts, the machine learning software could also be used as an indicator of how well a teacher thinks her class is doing. For instance, in tenth and twelfth standard classes, which is when standardised tests take place, a teacher roughly knows what percentage of her class will do well, what percentage will fail and so on.

The Microsoft system could potentially be used therefore to keep track and determine the difference between predicted and actual outcomes. This would effectively introduce a system of accountability: If a teacher can’t provide an accurate assessment of her class’s success, either there are factors beyond the teacher’s control or the teacher requires greater training.

The Aadhaar-Andhra Nation

For the Andhra Pradesh government, the Aadhaar system is crucial to its project of predicting dropouts and using technology to increase the efficiency of its schools. At the most fundamental level, socioeconomic indicator data from the UID database is fed into Microsoft’s software.

Aadhaar data also helps in building a socioeconomic profile of the student. One person with direct knowledge of the matter told this correspondent that the state government could use the UID system to identify whether a student’s parents were part of the NREGA system; thus providing it with data on potential income and family status which in turn could help in predicting dropout risk. The Wire, however, couldn’t independently confirm this claim though.

This lays the ground-work for a system that, according to a Microsoft newsletter, can capture a “360-degree view of students, mapped close to 100 variables”.

An integral part of this is how the Aadhaar system is also being positioned as a method to track a student’s journey through the education system. The need for this became clear during the pilot project, when the government needed to pinpoint the number of dropouts between tenth and twelfth standard. To do this, the tenth standard results of all students had to be connected to their enrolment in the eleventh standard after which the exit points of the dropouts could be mapped.

This becomes an exhausting process if the government has to do it for each student within the state as they move from elementary to higher education.  Obviously, there needs to be greater collaboration between departments and ensure that school management information systems can link up, one government official admitted, while speaking with The Wire.

The most attractive solution, however, would be to use the Aadhaar platform as a unique identifier, thus allowing the government to follow a student throughout his education journey. There are a number of indicators that the Andhra Pradesh government is headed in that direction: The Sakshi Post has covered extensively the state government’s initiatives at making the Aadhaar card compulsory for students. A few months ago, it published a report that detailed the government’s efforts in collecting the Aadhaar details of nearly “41 lakh school students out of a total 60 lakh students.”

The reasons given for the state government’s aggressive Aadhaar push are similar to how Aadhaar has been sold elsewhere; that it would go a long way in eliminating double admissions and examination fraud. However, other measures that the education department is mulling — for instance the implementation of the Aadhaar number on the Class X marksheet — dovetail nicely with the project of curbing dropouts and being able to track the state’s students as they move from one grade to another.

Privacy privacy privacy

The elephant in the room — when it comes to any sort of data-driven governance and algorithmic regulation — is that of privacy. Embarking on projects such as this without proper safeguards, especially in light of the debate surrounding the Aadhaar system and the lack of privacy legislation in India, could be downright dangerous.  At the end of the day, terabytes of data are being funnelled towards to a Silicon Valley-based company whose community’s approach towards privacy has taken a beating since the Edward Snowden revelations.

When asked, Bhansali pointed out that unless the client asks for it, no personally identifiable information is used.

“We take it [privacy] very seriously. These data centres [for the project] are in India. When we work for the government, the data resides in India. From that point of view, nothing goes out of the country,” he says.

In addition to this, Microsoft also claims that the data is tied to the Andhra Pradesh government account and that the company itself doesn’t have the right to use it or repurpose it. “We don’t store any of the data ourselves.. It’s tied to the government’s account. The Andhra Pradesh government owns that data, that nobody else shares or uses. They decide whether the data needs to be shared,” Bhansali said.

Effectiveness and legacy

The Andhra Pradesh- Microsoft data-driven governance initiative represents a growing number of attempts at using citizen data to increase the efficiency of the Indian State; to make more informed decisions. Decisions that in the past were perhaps made by intuition, or a need to impress a certain constituency, but can now be taken with the help of cold-hard facts.

How successful will Naidu and Microsoft’s attempt be at curbing school dropouts? Anita Kumar of Plan India – NGO that fights for girl child education and against child labour in Andhra Pradesh, has one or two concerns but believes that overall the project definitely has the potential to help.

“Firstly, there are a few fundamental issues which plague school dropouts. Fundamental issues such as how far away the school is from a girl student’s home. While it is officially supposed to be one kilometre away, in tribal and forest regions this is rarely so, social issues also still persist,” says Kumar.

To be fair, however, algorithmic regulation isn’t geared to deal with these issues and Kumar agrees. Where she thinks it will be effective is in hopefully highlighting how a one-size-fits-all curriculum doesn’t work for a substantial number of government schools and increasing the accountability of teachers.

“What is missing in Andhra Pradesh especially, and generally throughout India, is a high quality of engagement with the child. That child-friendliness atmosphere is lacking in government schools. The teaching methodology is not participatory and we aren’t looking at how we engage the child and so on. These factors can’t be immediately quantified as well,” she says.

While the child-friendliness of government schools in Andhra Pradesh can’t be quantified, the data that does exist on problems such as corporal punishment in government schools isn’t pretty. In July 2015, a school in Telangana made news after a student died after being forced to kneel in the hot sun as punishment for not completing her homework.

A number of studies over the last five years consistently place Andhra Pradesh as one of the top three states in terms of corporal punishment at schools.

The initial results of a Young Lives India Study conducted across a ten-year period (with 2016 set to be the last round of surveys) in Andhra Pradesh showed that 92% of younger students reported witnessing some form of corporal punishment at school while 77% of younger students personally experienced some form of corporal punishment at schools.

“If students continue to be beaten and the teacher-student relationship doesn’t change, then dropouts will continue,” Kumar says.

Microsoft and Bhansali agree on this count. “Technology can’t solve all the problems,” says Bhansali. “It can help the organisation and allow us to conduct meaningful interventions. Can our software be fed data, analyse a problem, allow for more targeted intervention and then reduce the percentage of dropouts? That’s our aim.”