Collaboration with the Colombia Statistics Office

academia engagement

The Leverhulme Centre for Demographic Science is partnering with Colombia’s Stats Office to rethink how population counts are produced between censuses. By combining Bayesian modelling, administrative records, and satellite data, the project is pioneering new methods to generate accurate annual population estimates.

Edith Darin true
2023-07-01

The Leverhulme Centre for Demographic Science is building a partnership with the National Administrative Department of Statistics of Colombia (DANE) to work together to address the challenge of outdated population count in Colombia, paving the way for other national statistics offices to follow.

Through our partnership with DANE, we are addressing the following question: Can we predict the demographics that a national census would observe if it were conducted today?

This project is designed to build the foundation to answer this question by achieving four objectives:

  1. Build knowledge at DANE to implement Bayesian statistical modelling,

  2. Install necessary computing infrastructure at DANE,

  3. Implement a pilot project as a proof-of-concept to “predict the census”, and

  4. Foster new relationships for broader collaborations between Oxford and DANE.

Our ultimate aim is to produce annual population estimates for official use by the Colombian government and to share these approaches with a network of national statistics offices globally through our partnership with the United Nations Population Fund (UNFPA).

We conducted conduct a one week capacity-building workshop at DANE in Bogotá, Colombia in partnership with UNFPA. We taught fundamentals of Bayesian statistics to representatives from 5 national statistics offices in Latin America, targeting methods to produce annual small area population estimates using the 2018 Colombian census, annual administrative registers, geospatial data, and social media data. We then invited DANE researchers to visit Oxford for a one week knowledge exchange workshop intended to spark additional collaborations and expand the partnership between our institutions to include a broader range of topics and diversity of researchers.

Building on this visit, we worked on two different research projects.

The first one consist on using confidential Colombian census data as a benchmark for testing satellite-imagery derived population. The results have been published and have shown that detailed building footprint data significantly improves accuracy of population model, and that Bayesian models are more effective than machine learning when the input data is biased.

The second research strand consists in modelling intercensal population by leveraging the signals from administrative records to update the 2018 census. I have accessed to anonymized administrative data for 50 million Colombians in 52 administrative databases. I am currently working on a state-space model, where the process model is based on the cohort component framework for population projecton that are nowcasted through observation models based on the administrative records.

Citation

For attribution, please cite this work as

Darin (2023, July 1). Meet Edith: Collaboration with the Colombia Statistics Office. Retrieved from https://edarin.github.io/thatsme/posts/2023-07-01-collaboration-with-the-colombia-statistics-office/

BibTeX citation

@misc{darin2023collaboration,
  author = {Darin, Edith},
  title = {Meet Edith: Collaboration with the Colombia Statistics Office},
  url = {https://edarin.github.io/thatsme/posts/2023-07-01-collaboration-with-the-colombia-statistics-office/},
  year = {2023}
}