Start with the path back to the source
A number can look precise even when its origins are unclear. Before deciding what a figure means, ask where it came from. Was it recorded during a transaction, reported by people answering a survey, estimated from a model, or copied from another publication? Each route can produce useful information, but each answers a slightly different question. A sales record describes recorded purchases; a survey about purchasing habits describes what respondents remember and choose to report.
Try to trace the figure back to the organization or person that produced it, rather than stopping at the article or graphic where you found it. Look for a description of the original dataset and its collection method. If a chart cites a report, and the report cites a survey, the survey is the place to check how the number was made. When that trail is unavailable, treat the figure as harder to evaluate, not automatically false.
Find out exactly what is being counted
Every measure depends on a definition. A count of “customers” might mean individual people, accounts, households, or purchases. “Active users” could mean people who opened an app once in a month or people who used a particular feature each week. Neither definition is inherently wrong, but swapping one for another changes the story. Look for a sentence that identifies the unit being counted and the rule for including it.
Then check the time period and location. A weekly total is not directly comparable with a monthly total, and a figure for one region may not describe a wider population. Percentages need special care: find the denominator, or the group that forms the whole. If 30 percent of survey respondents chose an option, that does not necessarily mean 30 percent of all customers would choose it. The meaning depends on who was asked and who answered.
Consider how the data was collected
Collection methods shape what data can show. A survey may capture opinions that cannot be observed directly, but its wording and response options can influence answers. An administrative record, such as a service log, may document actions consistently while missing activity that happens outside the system. A model can combine several inputs to estimate a value that cannot be counted directly; in that case, it helps to know which assumptions drive the estimate.
Ask whether the source covers the people or events relevant to your question. A voluntary survey may attract people with stronger opinions. A store’s purchase records will not include purchases made elsewhere. Missing information matters most when the absent group could differ from the included group. You do not need to identify every possible gap. Start by naming one plausible group or type of activity the method might miss, then consider whether its absence could change your conclusion.
Make comparisons on matching terms
Two numbers belong side by side only when their definitions and collection methods are sufficiently alike. Suppose one source reports the number of support requests received, while another reports the number of people who contacted support. A person can make several requests, so the figures measure different things. Even if both sources use the label “support volume,” they should not be treated as interchangeable without clarification.
Changes over time need the same scrutiny. An apparent increase might reflect a real shift, but it might also follow a new reporting form, a wider coverage area, or a revised definition. Check whether the source explains any change in how it collected or classified data. If you cannot establish that the measures match, describe the figures separately rather than calculating a growth rate or presenting a trend as settled.
Use a short source check before acting
A practical check can be brief: Who produced the number? What is the unit, time period, location, and denominator? How was the information gathered? Who or what might be missing? Are any figures being compared measured in the same way? These questions work for a workplace dashboard, a consumer survey, or a chart shared in conversation. They help you identify what the number can support without demanding perfect data.
If an important answer is missing, state the uncertainty in ordinary language. For example, “This figure covers responses from surveyed customers, not necessarily all customers” is more informative than either ignoring the limit or dismissing the data entirely. Good data literacy is not about distrusting every number. It is about keeping a number connected to the choices that produced it, so the conclusion does not reach further than the evidence allows.
