From Interval Endpoints to a Statement About a Population
A confidence interval is useful only if its numerical endpoints are connected to the question that motivated the sampling. An interval such as \((0.3577, 0.4723)\) does not, by itself, tell a reader who or what was studied, which characteristic was counted, or how confident the method is. A careful interpretation supplies all three.
For a population with a binary characteristic, let \(p\) denote the proportion of the population having that characteristic. For example, if the characteristic is supporting a proposal, each member either supports it or does not. A sample produces an estimate \(\hat{p}\), and a confidence interval uses the sample information to give a range of plausible values for \(p\). The interval is about the population proportion, not a range containing a specified proportion of individual people.
A Reliable Interpretation Has Three Parts
A complete statement identifies the population to which the estimate applies, the characteristic whose proportion is being estimated, and the confidence level associated with the interval procedure. A useful sentence pattern is:
The interval endpoints must remain attached to the exact population and characteristic used to define \(p\). If the sample was drawn from households in one county, an interval does not automatically describe all households in the country. If the characteristic is having a particular opinion, the result does not estimate the proportion who will act on that opinion.
The endpoints are often written as decimals because proportions range from 0 to 1. To state them as percentages, multiply each endpoint by 100. For example, \(0.24\) corresponds to \(24\%\). This conversion changes the scale of the numbers, not the population or the meaning of the interval.
What the Confidence Level Adds
The confidence level describes the long-run performance of the interval-producing method. If we repeatedly took random samples under the same conditions and constructed an interval in the same way each time, the stated proportion of those intervals would capture the fixed population proportion \(p\). The particular interval from one sample either covers \(p\) or does not; its endpoints do not make \(p\) a changing quantity.
This repeated-sampling description explains why a contextual interpretation includes both the endpoints and the confidence level. The endpoints report the range produced by this sample. The confidence level describes how often the method would produce intervals containing the true proportion over repeated sampling.
What Must Be True for the Interval to Support the Statement?
The interpretation is only as sound as the interval itself. A standard one-proportion \(z\) interval relies on a random sample from the population of interest, independent observations (or an adequate independence approximation), and enough observations of both outcomes for the normal approximation to be reasonable. For sampling without replacement, the sample is commonly required to be no more than 10% of the population. The success-failure check examines whether the sample has enough observations in each outcome category.
These conditions do not alter the grammar of the interpretation, but they matter to whether the confidence procedure is justified. In the examples, the sampling details are given so that the target population and the basic conditions can be checked. Assume each interval is a standard one-proportion \(z\) interval calculated from the described sample.
Worked Example: Support for a City Proposal
In a hypothetical city with more than 2,000 registered voters, a simple random sample of 200 voters finds that 83 support a proposed transit plan and 117 do not. A 90% confidence interval is reported as \((0.3577, 0.4723)\). Interpret it in context.
Check the setting. The outcome is binary: a voter either supports the plan or does not. The sample is random, and 200 is less than 10% of a population exceeding 2,000. The sample has 83 supporters and 117 nonsupporters, so both outcome counts are adequate for the usual normal approximation.
The sample proportion is \(\hat{p}=83/200=0.415\). To see how the reported endpoints relate to the estimate, the standard one-proportion interval has the form
For a 90% interval, use \(z^*\approx1.645\). Here the estimated standard error is approximately \(\sqrt{0.415(0.585)/200}\approx0.03484\), so the margin of error is \(1.645(0.03484)\approx0.05731\). The endpoints, using the displayed rounded values, are \(0.415-0.0573118=0.3576882\) and \(0.415+0.0573118=0.4723118\), which round to \(0.3577\) and \(0.4723\).
Convert the endpoints to percentages: \(0.3577(100)=35.77\%\) and \(0.4723(100)=47.23\%\). We are 90% confident that between 35.77% and 47.23% of all registered voters in this city support the proposed transit plan.
The population is all registered voters in this city, not only the 200 sampled voters. The characteristic is supporting the plan, not having heard of it or saying they might vote for it.
Worked Example: Households With Solar Panels
In a hypothetical county with 3,000 households, a random sample of 120 households finds that 31 have rooftop solar panels. A 95% interval is reported as \((0.1800, 0.3367)\). State its interpretation.
Check the setting. Each sampled household either has rooftop solar panels or does not; the counts are 31 and \(120-31=89\). The sample is 4% of the 3,000 households, which is below 10%, and both outcome counts are sufficient for the usual normal approximation.
The sample proportion is \(\hat{p}=31/120\approx0.258333\). The estimated standard error is \(\sqrt{(31/120)(89/120)/120}\approx0.03996\). With \(z^*\approx1.96\), the margin of error is approximately \(1.96(0.03996)=0.07832\). Using the unrounded sample proportion and standard error gives endpoints of about \(0.1800\) and \(0.3367\).
Convert each endpoint to a percentage: \(0.1800(100)=18.00\%\) and \(0.3367(100)=33.67\%\). We are 95% confident that between 18.00% and 33.67% of all households in this county have rooftop solar panels.
These are percentage bounds for the population proportion. The statement is not that the interval estimates the percentage of sampled households: the sample percentage is \(31/120\approx25.83\%\), a single estimate lying within the interval.
Why the Population and Characteristic Matter
A proportion is defined by a population and a rule for classifying each member. Changing either one changes the parameter being estimated. “The proportion with a follow-up visit” is incomplete unless the population of visits, patients, or records is specified, and the rule for what counts as a follow-up is clear.
A helpful way to justify the standard confidence wording is to write down the target before interpreting the endpoints. The parameter \(p\) is fixed for the defined population and characteristic. The interval procedure is random because a different random sample could produce different endpoints. The confidence level describes the procedure across repeated samples; the observed interval then supplies the numerical range for \(p\).
This distinction also prevents a common overreach: an interval based on a sample from one defined group does not automatically support a claim about a broader group. Generalizing beyond the sampled population requires a sampling design that represents that broader group, not merely a confidence interval.
Worked Example: Follow-Up After a Hospital Discharge
In a hypothetical hospital, there were 6,000 eligible discharge records during a year. A random sample of 250 records includes 205 with a follow-up visit within seven days and 45 without one. Software reports a 95% confidence interval of \((0.7724, 0.8676)\). Interpret it.
Check the setting. The unit is a discharge record, and the binary characteristic is having a follow-up visit within seven days. The sample is \(250/6{,}000\approx4.17\%\) of the records, below 10%, and the two outcome counts, 205 and 45, are both adequate for the usual normal approximation.
The sample proportion is \(\hat{p}=205/250=0.820\). The estimated standard error is \(\sqrt{0.820(0.180)/250}\approx0.02430\). The 95% margin of error is approximately \(1.96(0.02430)=0.04763\), giving endpoints of about \(0.820-0.04763=0.7724\) and \(0.820+0.04763=0.8676\), subject to rounding.
On a percentage scale, the lower endpoint is \(0.7724(100)=77.24\%\), and the upper endpoint is \(0.8676(100)=86.76\%\). We are 95% confident that between 77.24% and 86.76% of all eligible discharge records at this hospital during the specified year had a follow-up visit within seven days.
The conclusion refers to discharge records in that hospital and year. It does not state the follow-up rate for every hospital, for all patients in general, or for discharges in a different year.
A Common Mistake: Leaving Out the Target
A weak interpretation might say, “We are 95% confident the proportion is between 18% and 34%.” The numbers and confidence level are present, but the sentence leaves “the proportion” undefined. A reader cannot tell which households, voters, records, or characteristic the claim concerns.
Correct the statement by naming the population and characteristic: “We are 95% confident that between 18% and 34% of all households in the county have rooftop solar panels.” The endpoints now answer a specific population question. Before finalizing an interpretation, check that the named population matches the sampling frame and that the characteristic is the same binary outcome used to calculate the interval.
Check Your Understanding
For each question, make sure your answer identifies the population, the characteristic, and any relevant confidence level.
- A 90% interval for the proportion of students at one university who commute by bus is \((0.12, 0.21)\). Write a contextual interpretation in percentages.
- A 95% interval is reported as \((0.403, 0.517)\) for the proportion of sampled city residents who favor a park renovation. What information about the target population should be included in the interpretation?
- Explain why a 95% confidence interval does not mean that 95% of individual members of the population have the characteristic.
- A report gives a proportion interval from a sample of households in one county. What would need to be justified before using it to make a claim about all households in the state?
- Convert the endpoints \(0.286\) and \(0.341\) into percentages, and state what information must accompany them for a complete contextual interpretation.