3 Understanding Uncertainty
Learn how incomplete evidence, estimates, intervals, and differences across contexts affect what conclusions can be drawn and how uncertainty should be communicated.
What means
describes how much the available evidence leaves room for more than one plausible conclusion. It does not automatically make evidence poor or useless: a careful with an honest account of its limits can be more informative than a precise-looking number without an explanation.
When interpreting a result, ask three questions: What is uncertain? How large or important is that ? What does it mean for the conclusion or decision?
Check evidence and its limits
Missing information can arise when people do not respond, measurements are missed, records are unavailable, or a study covers only part of the population or period of interest. Gaps can affect both the amount and the representativeness of evidence. If those missing differ in relevant ways from those observed, the result may be biased, not merely less precise.
Errors can also remain when every intended observation is present. Questions may be misunderstood, instruments may measure imperfectly, records may be processed inconsistently, or the data may not match the question. Distinguish —the from observing a sample rather than the whole population—from other sources of error, including , inaccurate answers, and processing mistakes.
Assess whether the evidence is fit for its intended use by checking:
Completeness: What information is missing, and could the gaps systematically affect the result?
Accuracy and consistency: Were measurements and procedures reliable and applied similarly?
Relevance: Do the data address the question and population of interest?
Timeliness and context: When and under what conditions were the data collected?
Data quality depends partly on the question and context, not just on the amount of data.
Interpret estimates and intervals
An is calculated from data to describe or approximate something not directly known, such as a population average. Reporting several decimal places does not make an exact. Its interpretation depends on how the data were collected, how the was calculated, and which sources of are represented.
A describes the estimated sampling of an . A gives a range generated by a procedure designed to capture the unknown population value at a stated rate over repeated samples, under the method’s assumptions. For example, when assumptions hold, a 95% confidence-interval procedure captures the true value in about of repeated samples. This does not mean that, after one particular interval is calculated, there is a probability that the fixed true value lies within it. Wider intervals generally indicate less precision.
Suppose a survey estimates that of residents support a proposal, with a from to . The alone might suggest a narrow majority, but the interval is also compatible with support below . The result does not establish that support is definitely above half.
An interval may not capture every important limitation. A narrow interval can accompany biased or unrepresentative data if it quantifies sampling without accounting for , measurement problems, or other systematic errors. Check what the measure includes and what it leaves out.
Consider across contexts
Results can differ across places, groups, time periods, or conditions. This may be real—for example, an intervention may work differently in different settings—or it may reflect limited data and chance. Check whether the evidence covers the context where you want to apply it. An average across many settings may hide meaningful differences, while a result from one setting may not transfer to another.
Comparisons should account for in each . If estimates or their ranges overlap, the evidence may not clearly distinguish them; overlap alone is not a definitive test of whether a difference exists. Consider the size and practical importance of the difference, the study design, and whether the comparison was planned and measured consistently.
Communicate clearly
A useful report states the central , the measure and what it represents, the main data limitations, and the conditions under which the result applies. Identify assumptions when conclusions depend on them. Use precise probability language only when it has a defined basis; otherwise, describe evidence quality and areas of agreement in plain language.
Show in a chart when it could change the interpretation of a chart or comparison, and explain what the displayed range represents. When would not affect the conclusion, avoid unnecessary visual clutter, but make relevant details available in notes or accompanying data.