Report Interval Estimates as Answers

From an Interval to an Answer

A confidence interval is often the main answer to a question comparing two groups. But reporting two endpoints alone leaves important questions unanswered: what population quantity do they estimate, which groups are being compared, and what does the range tell us? A useful conclusion answers all three. It names the parameter, gives the estimated range in the context and units of the question, and states the limits of what the interval establishes.

For example, an interval for a difference in average recovery time is not an interval containing most patients’ recovery times. It is an estimated range for a population mean difference. Similarly, an interval for a difference in proportions concerns the population proportions, not the outcomes of particular individuals. Keeping the parameter clear prevents an otherwise correct numerical interval from being misinterpreted.

Definition: A confidence interval is a range of values, calculated from sample data by a specified procedure, intended to estimate an unknown population parameter. To report one as an answer, identify the parameter, give the interval endpoints in context and with units, state the confidence level, and interpret the range as an estimate of that parameter.

What the Confidence Level Means

The confidence level describes how a procedure performs over repeated samples, not the probability that a fixed parameter lies in one particular interval. If a method has 95% confidence coverage under its assumptions, then across many repetitions of the sampling process, approximately 95% of the intervals produced by that method would contain the true population parameter. After a particular sample has been observed, its interval is fixed; the parameter is also fixed, though usually unknown.

This repeated-sampling interpretation explains why the interval is often described as a range of plausible parameter values. Values well outside the interval are not supported by the data at the interval’s stated confidence level under the method used. But an interval does not assign a probability to the parameter’s being at any one value, and it does not by itself tell us whether an effect is important in practice.

Key interpretation: A 95% confidence interval is produced by a method that would capture the true population parameter in about 95% of repeated samples, when the method’s assumptions hold. For one observed interval, report it as an estimated range for the parameter; do not say there is a 95% probability that the parameter is in that interval.

The connection between a confidence interval and repeated sampling can be seen by inverting a standardized estimate. For example, suppose an estimate \(\hat{\theta}\), with standard error \(SE\), has an approximately \(t\) sampling distribution when centered at the true parameter \(\theta\). For a two-sided 95% interval, the central \(t\) range contains approximately 95% of that standardized distribution. Rearranging that range to isolate \(\theta\) gives the interval:

P\left(-t^* \leq \frac{\hat{\theta}-\theta}{SE} \leq t^*\right) \approx 0.95 \quad\Longrightarrow\quad \hat{\theta}-t^*SE \leq \theta \leq \hat{\theta}+t^*SE.

Here \(t^*\) is the critical value for the relevant degrees of freedom. The approximation depends on the sampling model and its assumptions. The algebra also shows why the interval is centered on the estimate and why its width reflects the standard error. The interpretation, however, always concerns \(\theta\), the population parameter—not the estimate itself and not individual observations.

Worked Examples: State the Parameter and the Range

Worked Example: Difference in Two Population Means

Suppose a hypothetical study compares the time, in minutes, to complete a task for two independently sampled groups. Group A has \(n_A=10\), \(\bar{x}_A=12.4\) minutes, and \(s_A=\sqrt{3.9}\) minutes. Group B has \(n_B=10\), \(\bar{x}_B=10.8\) minutes, and \(s_B=\sqrt{2.494}\) minutes. Assume the samples are random and independent, and that plots show no severe skewness or outliers in either group. With these small samples, those distribution checks matter for using a two-sample \(t\) interval.

The target parameter is \(\mu_A-\mu_B\), the difference in the population mean task-completion times, with Group A first. The sample estimate is \(12.4-10.8=1.6\) minutes. Using the Welch two-sample \(t\) interval, the estimated standard error is

SE=\sqrt{\frac{s_A^2}{n_A}+\frac{s_B^2}{n_B}} =\sqrt{\frac{3.9}{10}+\frac{2.494}{10}} =\sqrt{0.6394} \approx 0.7996\text{ minutes}.

The Welch degrees of freedom are calculated from the two estimated variance contributions:

df= \frac{(0.39+0.2494)^2} {\dfrac{0.39^2}{9}+\dfrac{0.2494^2}{9}} \approx 17.17.

For a 95% interval, the corresponding critical value is approximately \(t^*=2.108\). Therefore,

(\bar{x}_A-\bar{x}_B)\pm t^*SE =1.6\pm 2.108(0.7996) =1.6\pm 1.686 \approx (-0.086,\ 3.286)\text{ minutes}.

A suitable conclusion is: For the populations represented by these two samples, the 95% confidence interval for the difference in mean task-completion time, Group A minus Group B, is approximately −0.086 to 3.286 minutes. The interval includes zero, so these data do not pin down whether the population mean is greater in Group A or Group B. That does not establish that the population means are equal; a range of differences on either side of zero remains compatible with the estimate.

Worked Example: Difference in Two Population Proportions

In a hypothetical survey, 78 of 120 sampled people in Group A and 55 of 110 sampled people in Group B report using a particular service. Suppose each group is a random sample from its target population, the two samples are independent, and all four success and failure counts are sufficiently large for the normal approximation: 78 and 42 in Group A, and 55 and 55 in Group B.

The sample proportions are \(\hat{p}_A=78/120=0.65\) and \(\hat{p}_B=55/110=0.50\). The parameter is \(p_A-p_B\), the difference between the population proportions. Using the unpooled standard error for a two-proportion confidence interval,

SE=\sqrt{\frac{\hat{p}_A(1-\hat{p}_A)}{n_A} +\frac{\hat{p}_B(1-\hat{p}_B)}{n_B}} =\sqrt{\frac{0.65(0.35)}{120}+\frac{0.50(0.50)}{110}} =\sqrt{0.004169} \approx 0.06456.

The 95% normal critical value is \(1.96\), and the estimated difference is \(0.65-0.50=0.15\). Thus,

0.15\pm1.96(0.06456) =0.15\pm0.1265 \approx (0.0235,\ 0.2765).

A contextual report is: The 95% confidence interval for the difference in the population proportions using the service, Group A minus Group B, is approximately 0.0235 to 0.2765, or 2.35 to 27.65 percentage points. The interval estimates a population difference in proportions. It is not a range for the percentage of service users in either group separately, nor does it imply that membership in one group caused service use.

Because the interval is entirely above zero, it supports a positive difference for the parameter as defined. It still does not tell us whether that difference matters for a practical decision. That judgment depends on context, consequences, and what size of difference would be meaningful.

Worked Example: Mean Change in Paired Measurements

A hypothetical study records a measurement in minutes for each of 16 people before and after a program. The data are paired because both measurements come from the same person. Define each difference as after minus before. Suppose the differences have sample mean \(\bar{x}_d=-3.2\) minutes and sample standard deviation \(s_d=4.0\) minutes. Assume the 16 people are independent of one another and the distribution of paired differences has no severe skewness or outliers.

The population parameter is \(\mu_d\), the population mean within-person change in minutes, defined as after minus before. A one-sample \(t\) interval on the differences has \(df=n-1=15\), and its standard error is

SE=\frac{s_d}{\sqrt{n}}=\frac{4.0}{\sqrt{16}}=1.0\text{ minute}.

Using \(t^*=2.131\) for a 95% interval with 15 degrees of freedom,

\bar{x}_d\pm t^*SE =-3.2\pm2.131(1.0) =(-5.331,\ -1.069)\text{ minutes}.

A suitable conclusion is: The 95% confidence interval for the population mean change, after minus before, is approximately −5.331 to −1.069 minutes. Under this definition of change, the estimated range is entirely negative: it corresponds to a mean after measurement between about 1.069 and 5.331 minutes lower than the mean before measurement. This interval estimates an average change; it does not say that every person’s measurement decreased by that amount.

What an Interval Does and Does Not Establish

A confidence interval gives a range of values for a specified population parameter, with uncertainty summarized by the interval’s width and confidence level. A narrow interval indicates a more precise estimate than a wide interval, in the units of the parameter. Neither width nor confidence level alone determines whether the estimated difference is important: that requires a meaningful benchmark from the application.

For a difference parameter, zero represents no difference between the two population quantities. If a two-sided confidence interval excludes zero, it indicates that zero is not among the values in that interval. Under the corresponding assumptions and procedure, this is evidence of a nonzero difference at the matching two-sided significance level. If the interval includes zero, the data do not rule out zero using that interval—but they do not prove that the groups are identical. The interval’s other values matter too: a very small difference and a large difference may both remain plausible.

An interval also cannot repair weaknesses in the design. If groups were not randomly sampled, the population to which the result generalizes may be limited. If assignment was not randomized, an observed group difference does not by itself establish a causal effect. A precise interval can still be precisely about the wrong target, or about an association rather than a cause.

A Reliable Reporting Sequence

A concise interval conclusion can be built in a few deliberate steps. First name the population parameter, including the groups and the quantity being compared. Then state the confidence level and endpoints with units. Finally, describe what the range indicates and note any important limit on interpretation.

Name the parameter.
Specify whether the interval estimates a population mean, a population proportion, or a difference between population quantities. For a paired analysis, identify the population mean difference and how each difference was defined.

Give the range in context.
State the confidence level, both endpoints, the units, and the populations or groups named in the question. Keep the subtraction order consistent with the defined difference parameter.

Explain its implication and limits.
Say what the endpoints suggest about the population parameter. Do not turn the interval into a probability about a fixed parameter, a range for individual observations, or a causal conclusion unsupported by the study design.

A useful final check is to read the conclusion without looking at the calculations. Can a reader identify what quantity is estimated, whose population it concerns, what the endpoints mean, and what the interval does not show? If not, add the missing context rather than adding more decimal places.

Common Mistakes

Giving endpoints without the parameter. “The interval is 2 to 6” does not say whether the numbers describe a mean, proportion, or difference. Name the population quantity and give its units.
Claiming a probability for the fixed parameter. “There is a 95% probability that the true difference lies in this observed interval” is not the repeated-sampling interpretation. State that the method produces intervals that capture the parameter in about 95% of repeated samples, under its assumptions.
Treating the interval as a range of individual outcomes. A confidence interval for a mean or mean difference estimates an average population quantity. It does not predict that each person’s value will fall between the endpoints.
Calling an interval that includes zero proof of no difference. Such an interval leaves zero among the values compatible with the estimate at that confidence level; it does not establish equality.
Turning a difference into a causal claim. An interval estimates the parameter defined by the study. Causal wording needs a design that supports causal interpretation; an interval alone does not supply that design.
Reporting a numerical difference without context. “The estimate is 0.15” is difficult to interpret. Say whether it is a proportion difference, a mean difference, or another quantity, and state the units or percentage points.

Key takeaway: Report a confidence interval as an estimated range for a clearly named population parameter. Include the confidence level, context, and units; then explain what the interval supports without turning it into certainty, an individual prediction, or an unsupported causal claim.

Check Your Understanding

For each question, focus on the parameter and on what a careful interval conclusion can claim.

A 95% interval for a population mean difference in waiting time is −1.2 to 4.8 minutes. What should a contextual conclusion name, and why would “95% chance the true difference is in the interval” be incorrect?
A confidence interval for a difference in population proportions is 0.04 to 0.18. What does its position relative to zero indicate, and what does it not establish about practical importance?
A paired study defines change as after minus before and reports a mean-change interval from −6 to −2 points. Explain the sign in words and identify the population parameter being estimated.
Why is a confidence interval for a population mean not a range expected to contain most individual measurements?
A study reports a narrow interval for an association between group membership and an outcome. What additional issue must be considered before describing the group as causing the outcome?