Starting an Inquiry From the Numbers Up

Data-led investigations tend to go wrong at the very beginning, long before any spreadsheet is opened. The mistake is asking a question the available data cannot actually answer, then bending the analysis to produce an answer anyway. A more durable approach starts by testing the question itself against what is realistically measurable.

Suppose the starting hunch is that a particular district has seen an unusual rise in a category of incidents. Before assembling any figures, it is worth asking what "unusual" would even mean here — compared to which baseline, over what period, adjusted for population or not. A question phrased this loosely invites cherry-picking, because almost any dataset can be sliced to support a vague claim. A tighter question, phrased before the data is even opened, makes the eventual analysis honest by design.

Recording definitions as you go

Once a workable question exists, the next discipline is recording exactly what each source measures, as you gather it, rather than trusting memory later. A statistic labelled "reported cases" is not the same as "confirmed cases," and a survey using one sampling frame is not directly comparable to another using a different one. Keeping a running log of definitions, dates, and publication versions turns into the backbone of the eventual methodology note — and it is far easier to build this log in real time than to reconstruct it after the fact.

The ordinary explanation should be ruled out first, not last.

This leads to perhaps the most important habit in the whole process: checking a pattern against mundane explanations before reaching for unusual ones. A spike in a figure might reflect a genuine shift, or it might reflect a change in reporting practice, a new data collection method, or simple population growth. Working through the ordinary possibilities first, and ruling them out with evidence rather than assumption, is what separates a defensible finding from a coincidence dressed up as a story.

Seeking comment on the actual figures

Before publication, it matters to approach the bodies concerned with the specific figures in question, not a general request for comment. Asking "can you explain this particular number from this particular table" produces far more useful responses than an open-ended query, and it gives the subject of the inquiry a fair and precise opportunity to respond.

Finally, publishing the underlying working — the sources, the calculations, the definitions used — lets readers check the claim themselves. This is not a formality. It is often what determines whether an investigation is trusted or dismissed.