Cleaning national survey data without losing the story
At KNBS I saw how quality assurance shapes every downstream statistic. A dropped decimal or a mis-coded response ripples into policy. Good data cleaning is less about tidiness and more about protecting the meaning inside each row.
During my internship at the Kenya National Bureau of Statistics I expected data cleaning to be mechanical — deduplicate, standardise, move on. Instead I learned it is one of the most consequential acts in the entire statistical chain.
Every row is a person
A mis-coded response is not a formatting nuisance; it is a household misrepresented in a figure that will inform planning and budgets. A dropped decimal doesn't just fail a validation check — it can quietly distort a national estimate. Once you internalise that, quality assurance stops feeling like chores and starts feeling like stewardship.
I began treating each cleaning decision as a small ethical one: am I removing noise, or am I erasing signal? Documenting why a value was changed became as important as the change itself, because reproducibility is how you protect meaning when your work outlives your memory of it.
Tidiness in service of truth
Clean data should be tidy, yes — but tidiness is the means, not the end. The end is a dataset where every row still tells the truth it was collected to tell.
Cleaning national survey data without losing the story
Interested in this work? Catherine is open to public health & data roles worldwide.
Email Catherine