Reading a dataset critically
- Practice
- Ethics
Most analysis goes wrong before any code runs. It goes wrong in the quiet assumption that the file in front of you describes the thing you care about.
Three questions before the first chart
Who collected this, and why? Data is collected to serve a purpose, and that purpose shapes what got recorded. A registry built for billing will measure what can be billed for. It will be silent about everything else.
Who is missing? Absence rarely announces itself. People who could not reach the office, could not fill the form, or were not counted do not appear as null values — they simply are not there. In a fragile context, the people missing from a dataset are often the ones the analysis most needs to describe.
What question was it built to answer? A dataset can be excellent for its original purpose and misleading for yours.
A worked habit
Before summarising, write two sentences in plain language:
- What this data can support a claim about.
- What it cannot.
If the second sentence is hard to write, you do not yet understand the data well enough to chart it.
Why this matters here
Reporting on Syria often stitches together sources collected under different conditions, by different actors, for different reasons. Combining them is sometimes the only option — but the combination should be stated, not hidden behind a clean-looking average.
Being explicit about limits is not a weakness in an analysis. It is most of what makes one trustworthy.