Missing data and data in context

Completeness analyses

Cleaning continues! The focus of this week was understanding the completeness of the data and in particular whether missing data would be a huge challenge for this project. To estimate whether women who need to give birth are receiving adequate coverage of facility delivery, you need sufficient information for both the numerator and the denominator - meaning quality information on facility delivery and equally on the total number of women giving birth. This is the bread and butter of every research project as the extent and type of missingness can make the project unfeasible. I met with a former Harvard statistics researcher who previously interned at the Ethiopian Public Health Institute as well and received extremely helpful guidance on various techniques and common pitfalls around navigating missingness.  I am grateful for the opportunity to continue strengthening my technical coding and statistics skills in this process especially as I incorporate those skills in a new context.

Report on data missigness

Fortunately, after conducting exploratory analyses to determine data completion, I found that once aggregated to the district level, for the most part we had reports across all months in 2020 (the year of focus) but with some obvious and expected heterogeneity across regions. I worked on an internal brief summarizing these findings and shared it with our internal team for discussion.

In addition to aggregate missigness summaries, understanding patterns of missingness is important to identify specific problematic districts or months

2020 was also selected as the year of focus in our research because experts in the ministry informed us that it was the latest year with the most comprehensive data. Disruptions after 2020 with COVID and the conflict in Ethiopia tragically affected public health facilities as expected and particularly conflict-stricken areas like Tigray have since reduced or even stopped reporting to the DHIS2 system limiting the feasibility of using 2023-2024 data. A common theme for me throughout these reflections has been that as a researcher and data analyst, context is key. Data informs policy but it does not exist in a vacuum. The country context determines the quality and reliability of data and it also informs the interpretation of data. That is why I am appreciative of the opportunity to connect with talented Ethiopian colleagues and gain a better understanding of perspectives on the ground.