Justine Schott

In 2017, Muso launched a large, three-year randomized controlled trial (RCT) to test the impact of a proactive community case management (ProCCM) intervention compared to passive CHWs, which is Mali’s current model.

Unique identification numbers (IDs) were assigned to each person enumerated.

However, Muso learned that in practice, study participants may have multiple IDs, due to natural migration and marriage patterns as well as considerable population migration amidst increasing insecurity in the region.

Muso identified a need to perform probabilistic record linkage across census survey years in order to create a Master Person Index (MPI).

Check out the Linkedin post here.

Features of the record linkage:

• Used python recordlinkage package to efficiently calculate probabilities using various methodologies, customized for each feature
• Used blocking techniques to create model efficiently
• Created final probability of match using Expectation-Maximization (ECM), which weights probabilities using maximum likelihood estimators
• Determined threshold for acceptance by reviewing final probabilities

There are 3 jupyter notebooks stored in a private GitHub repository:

• 01_data_transformation.ipynb: Creates data transformations to ensure that the model will run on standardized data.
• 02_recordlinkage.ipynb: Runs record linkage model on two standardized datasets, for example year 0 and year 1.
• 03_final_datasets.ipynb: Creates final datasets that can be imported into a database or used for the next year of record linkage.