Every year, Statistics Canada (StatCan) conducts hundreds of surveys to better understand the economic and social realities of Canadians. By analyzing the vast amounts of data these surveys generate, StatCan helps inform government decisions on issues such as health care, affordable housing and unemployment.
Now, a new Canadian Statistical Sciences Institute (CANSSI)-funded Collaborative Research Team (CRT), co-led by professors David Haziza (uOttawa), Zeinab Mashreghi (University of Winnipeg), and Changbao Wu (University of Waterloo), is exploring how national statistical offices like StatCan can better leverage machine learning to analyze large, complex datasets. The team has received an impressive $210,000 in funding and their project will unfold over three years (April 1, 2026 to March 31, 2029).
What’s so unique about a CANSSI CRT?
According to Professor Haziza, CANSSI teams are diverse from multiple angles. A CRT should include both senior and junior researchers from large and small universities to provide ample opportunity for training and mentorship.
“If you work in a smaller university where there is no graduate program, a CRT gives you the chance to co-supervise a graduate student with other members of the team,” says Professor Haziza.
CRTs should also be geographically diverse. For example, Professor Haziza’s CRT brings together researchers from institutions across several Canadian provinces—including the University of Waterloo, Université Laval, McGill University, and the University of Winnipeg—as well as an international collaborator from the University of Oklahoma. Statistics Canada and the Bank of Canada are also project partners, with researchers from both organizations participating in the CRT.
“A CANSSI CRT is really about creating a diverse research ecosystem. By bringing together people with complementary areas of expertise, we can build on one another’s work and develop stronger, more innovative solutions.”
More than a research project, CANSSI CRTs facilitate collaborative networks that foster lasting partnerships and mentor the next generation of statisticians.
“A CANSSI CRT is really about creating a diverse research ecosystem. By bringing together people with complementary areas of expertise, we can build on one another’s work and develop stronger, more innovative solutions.”
David Haziza
— Professor, Department of Mathematics and Statistics
Machine learning meets survey sampling
Professor Haziza’s CRT focuses on how national statistical offices can better integrate machine learning into survey sampling practices.
Survey sampling involves selecting a small group of people from a larger population to draw conclusions about the whole population. Researchers like Professor Haziza then develop methods to make these estimates more precise, reduce bias and measure uncertainty.
On the other hand, machine learning is a collection of algorithms that learn patterns from data. Professor Haziza describes it as “a class of flexible tools” that can uncover relationships that may be too complex for traditional statistical methods. Machine learning is already being used for statistical tasks, such as treating missing data and generating predictions in the context of small area estimation.
But better predictions do not necessarily translate into more accurate estimates of population-level indicators, such as average income or the unemployment rate.
“If we just needed predictions, then the problem would be rather simple,” says Professor Haziza.
The research on statistical inference from machine learning predictions is still incomplete. Policymakers using Statistics Canada data need to know not only what the estimates are, but also how reliable they are, before using them to inform decisions.
Taking action
To address this problem, the CRT is developing the foundational frameworks needed to reliably use machine learning for survey sampling. Professor Haziza and his collaborators are studying how machine-learning methods affect estimation, variance estimation and confidence intervals.
Over the next several years, the team hopes to translate those findings into practical tools, including user-friendly software that could be used by national statistical offices.
By ensuring machine learning is not only powerful, but also trustworthy, the team is helping to shape the future of statistical science and strengthen the evidence that guides decisions affecting millions of Canadians.