Selected Country Experiences​
Rwanda and Togo used different methodologies for the PES matching. Rwanda carried out matching at head-quarters once all PES data had been collected. In contrast, Togo carried out matching in the field as PES enumeration was ongoing.
Rwanda conducted its PES one month after the main census, sampling 180 EAs across all districts, representing a 0.74% sampling fraction. The PES used a short questionnaire focusing on demographic details and household movement relative to census night. Enumeration was done digitally using a CAPI application, with real-time monitoring via a dashboard. A new team of 210 enumerators were recruited to ensure independence, supported by district team leaders and a central monitoring team. Matching between census and PES data was performed using a Python-based algorithm employing deterministic matchkeys that allowed for minor data discrepancies. These matchkeys used combinations of names, age, sex, and household identifiers, with techniques like Levenshtein distance and Soundex adaptations for local language nuances. Despite the absence of a PES pilot, the matching algorithm was refined during early data collection. The final strategy combined deterministic, associative, probabilistic, and clerical matching, with over 90% of records matched automatically. Clerical review and field reconciliation addressed unmatched cases. The use of digital tools and real-time monitoring significantly reduced the matching duration from six months in 2012 to under a month in 2022. The matched data were then used for dual system estimation to assess population size and coverage, and to evaluate content errors in the census.
Togo conducted its fifth General Population and Housing Census (RGPH-5) in November 2022, followed by a Post-Census Survey (EPC) in February 2023 to assess data coverage and quality. The EPC was fully digital, using a rigorous field-based data matching system. A sample of 160 enumeration zones (ZDs) was selected from over 10,000 mapped zones. Survey agents used Android tablets with a custom application to collect household data, which included demographic characteristics. Each field team had a team leader, a data matcher with a laptop and web-based matching tool, and a survey agent. Data were transferred via Bluetooth and internet to enable near real-time matching. The matching process focused only on residents present during both the census and EPC, excluding movers. Matching was based on geo-location, household size, and individual characteristics, with an algorithm assigning weights to similarity scores. The web-based tool allowed on-site matching without internet, displaying census and EPC data side-by-side for efficient comparison. Teams also used household lists from the main census to resolve discrepancies. The system categorised matches and generated summary tables with omission and coverage rates by zone. This approach enabled efficient, localised matching and real-time data validation, though it maybe raises questions about the independence of the PES process.