Imagine you're handed a spreadsheet showing that reported burglaries in a city fell by 20% over five years. It's tempting to conclude the streets got safer. But before you celebrate, ask a simple question: where did these numbers come from? Who wrote them down, and what did they gain or lose by doing so?
Most of us treat data like a photograph of reality, a neutral snapshot of how things are. In practice, it's closer to a memoir. Every dataset has an author, a purpose, and a set of circumstances that shaped what ended up on the page. Reading that backstory is one of the most underrated skills in analysis, and it doesn't require any software at all.
Data Is Made, Not Found
We often talk about data as if it were a natural resource. We 'mine' it, 'collect' it, 'gather' it. Those words suggest it's lying around like stones on a beach. But data doesn't exist until someone decides something is worth recording, defines how to record it, and builds a system to capture it.
Consider hospital records. A diagnosis code isn't a pure description of what's wrong with a patient. It's a choice made inside a system designed largely for billing and administration. Two hospitals treating identical cases might code them differently depending on their software, their staff training, or which codes get reimbursed. The illness is real. The record of it is a human translation.
This is true even of data that feels automatic. An app's 'daily active users' depends on what someone decided counts as active. Is it opening the app, tapping a button, or staying for thirty seconds? Every dataset is a set of answers to questions you didn't get to ask. The statistician John Tukey encouraged analysts to explore data with curiosity before drawing conclusions. Part of that exploration is remembering that the columns themselves were designed, and someone else did the designing.
TakeawayData is a record of decisions as much as a record of events. Before asking what the numbers say, ask what someone decided to count.
Following the Incentives
Once you accept that people make data, the next question follows naturally: what did those people want? Whoever records information is rarely neutral. They have targets to hit, managers to satisfy, and workloads to manage. Those pressures leave fingerprints.
Crime statistics are a classic example. Recorded crime depends on what victims choose to report and how officers classify it. When police departments face pressure to show falling crime, there have been well-documented cases of serious offences being logged as minor ones, such as a theft recorded as 'lost property.' The reverse happens too. A new crackdown on a particular offence can make that crime look like it's surging, simply because officers are now actively looking for it.
The same dynamic shows up in ordinary workplaces. A sales team rewarded for the number of customer calls will log more calls, including brief ones that barely count. A support team judged on how fast it closes tickets may close them quickly and open new ones later. None of this requires dishonesty. People respond sensibly to the systems around them, and the data quietly absorbs those responses. The idea is often summed up as Goodhart's law: when a measure becomes a target, it stops being a good measure.
TakeawayWhenever you see a number, ask who benefits if it goes up and who benefits if it goes down. The answer often explains more than the number itself.
Questions for Any Dataset
You don't need to be a professional investigator to read a dataset's biography. You just need a handful of questions to ask before you start calculating averages or drawing charts. Think of them as a background check.
Who made it? An agency, a company, a volunteer, a sensor configured by an engineer? For what purpose? Data collected for billing, compliance, or marketing will be shaped by those goals, not yours. What was left out? Every dataset has silent gaps: people who didn't respond, events nobody logged, categories that didn't exist on the form. What changed over time? A new definition, a software update, or a policy shift can create a jump in the numbers that has nothing to do with the real world.
These questions won't always have tidy answers. Sometimes the documentation is thin or missing entirely. But even noticing that you can't answer them is valuable, because it tells you how cautiously to hold your conclusions. A dataset with a well-understood history can support a strong claim. A dataset with a mysterious past should support a modest one. If you can, talk to someone who worked with the data directly. A five-minute conversation often reveals quirks no spreadsheet ever will.
TakeawayThe strength of your conclusion should match how well you understand where the data came from. An unknown origin calls for a humble claim.
Back to that falling burglary rate. Maybe the city really did get safer. Or maybe reporting habits changed, definitions shifted, or someone had a reason to prefer smaller numbers. The figure alone can't tell you which. Its biography can.
Next time you open a spreadsheet or read a headline statistic, pause before analysing. Ask who made the data, why, what's missing, and what changed. It's a small habit that turns a number reader into an evidence hunter.