How to unleash the power of data to transform education policies

The Data Must Speak Positive Deviance Research identifies schools that perform better than others operating in similar contexts and explores which behaviors and practices make them perform better. The research also investigates how to apply these good practices in more schools.

June 08, 2023 by Alexis Le Nestour, UNICEF Office of Research – Innocenti, and Renaud Comba, UNICEF Office of Research – Innocenti
4 minutes read
Class 10 students at Shree Krishna Ratna School in Chautara, Ward 5, Sindhupalchowk District, Nepal. Credit: GPE / Kelley Lynch
Class 10 students at Shree Krishna Ratna School in Chautara, Ward 5, Sindhupalchowk District, Nepal.
Credit: GPE / Kelley Lynch

With initial technical support from the Data Must Speak research team and our partners, statisticians in ministries of education (MoEs) can build their skills to link data within their country’s Education Management Information System (EMIS) and regularly report trends to policymakers.

What have MoEs and the DMS team learnt from linking administrative datasets?

The Data Must Speak (DMS) research team at UNICEF Innocenti has worked closely with ministry of education partners to co-create and conduct in-depth analyses of their administrative data.

First, we linked school information over time that allowed us to answer important questions for learning trends, such as knowing how many girls and boys were enrolled in each grade across several consecutive years. This made it possible to track student cohorts and to know which schools were successful at both retaining and promoting students to the next grade.

The DMS team also linked school census data with exam data to determine which schools were most effective at preparing students for assessment.

Using this data, key educational inputs (e.g., textbooks, number of teachers) could then be linked with student performance to help us understand the current status of education, as well as the challenges and potential methods to improve school retention and learning.

In Togo, DMS research indicated female students were more likely to be promoted to the next grade and score higher in exams when their teacher was a woman. This is an important finding as recruiting more female teachers could help eliminate the country’s education gender gap.

In Madagascar, Nepal and Togo, DMS research analyzed the relationship between student-to-teacher ratio and academic performance, which informed ministries of education about the potential educational benefits of recruiting more teachers. Reports from all participating countries are published on our website.

The DMS co-creation approach also allows for the documentation of best practices and recommendations on how to improve overall EMIS data collection, cleaning, merging and analysis. In Côte d’Ivoire and Ghana, those recommendations sparked an initiative to create unique school identification documents facilitating linking school information over time.

The future of research with school datasets

Integrating other datasets within country EMISs could open new opportunities. Education sector monitoring could be improved by adding new layers of information in EMIS.

For instance, in Niger, the DMS team linked information on schools with local data on poverty to understand the contexts in which schools operate. This approach could be systematically applied in all countries to measure socio-economic inequalities in education as well as monitor progress.

Climate and natural disaster data will also become increasingly important to help schools adapt to climate change consequences.

How to use EMIS data to assess education policy impact

Rigorously measuring the impact of new policies is critical to improving education systems, but it is clear EMIS data is underused to achieve this. Did a policy on mass textbook delivery increase exam performance rates? Did the drop-out rate decrease when the ministry of education expanded the number of canteens?

DMS research shows that EMIS data can answer these questions and, by working closely with ministries of education, help us understand which schools or regions benefit from specific policies.

To measure policy outcomes, academic performance can be compared with schools and regions that have not benefitted from a policy both before and after it is implemented. Impact evaluations, such as randomized control trials (RCTs), that embed EMIS data can test the success of new education policies.

RCTs are considered the gold standard in program evaluation, but they are usually time consuming and costly. However, in some cases, it is possible to embed RCTs within administrative data collection and save on costs.

A World Bank study conducted in Pakistan embedded an RCT within routine data collection to evaluate the impact of school management committees on school performance. DMS research, just like in Pakistan, leveraged this integrated data to test new education policies.

Ideally, these types of analyses would be done systematically by ministries of education through education labs whenever new policies are implemented. The work that DMS and education experts are doing to integrate education datasets and analyze data can lay the foundation for future country education labs to grasp the transformative impact of data.

Often, lack of data is seen as barrier to creating evidence-based policy. However, when it comes to education, the majority of existing administrative data fills this gap. Valuing the potential of existing datasets is cost-effective and sustains both ownership and investment in national data systems.

The DMS research plans to continue co-creating our research with ministries of education and local academia to leverage data in innovative ways and inform education policies for greater impact.

This blog was adapted from the original piece published by UNICEF.

Related blogs


Wow! This is indeed a great experience from afar. I hope to implement it in my school.

Leave a comment

Your email address will not be published. All fields are required.

The content of this field is kept private and will not be shown publicly.

Plain text

  • Global and entity tokens are replaced with their values. Browse available tokens.
  • No HTML tags allowed.
  • Lines and paragraphs break automatically.
  • Web page addresses and email addresses turn into links automatically.