Calculating an Interval from a Reported Percentage

From a Reported Percentage to an Interval

A report may say that 58% of surveyed adults supported a proposal, but an interval calculation needs more than a percentage alone. It needs the sample size as well. With the denominator, the percentage can be converted into a sample proportion; if the percentage was rounded, however, it may correspond to more than one possible count. This tutorial shows how to recover the information that is available, calculate the interval, and recognize when the reported precision limits the answer.

The method here is the one-proportion z interval. It estimates a population proportion \(p\) using the sample proportion \(\hat{p}\). A percentage is not itself a count: for example, 60% of 200 corresponds to 120, but 60% of 203 does not correspond to an integer count. Reports also commonly round percentages, so multiplying the displayed value by the sample size may produce only an approximate count.

Definition: If \(x\) of \(n\) sampled individuals have the outcome of interest, then the sample proportion is \(\hat{p}=x/n\). A reported percentage \(R\%\) represents the proportion \(R/100\) if it is exact. If it is rounded, the underlying proportion lies in a range determined by the report’s rounding precision.

Recovering the Count and Proportion

First identify the denominator \(n\), meaning the number of sampled individuals on whom the percentage is based. Then divide the reported percentage by 100 to express it as a proportion. Multiplying that proportion by \(n\) gives a candidate count \(x\). Because a count must be an integer, an exact percentage and denominator should produce an integer; a rounded percentage may produce a non-integer candidate, or several plausible integer counts.

Suppose a report gives \(R\%\), rounded to the nearest \(d\) percentage points. Ignoring a possible tie exactly at a rounding boundary, the underlying percentage is between \(R-d/2\) and \(R+d/2\). Dividing those limits by 100 gives a range for the underlying proportion. Multiplying the range by \(n\) gives a range for the count, from which we keep only integers. This is the direct way to check whether a reported percentage uniquely identifies the sample proportion.

Key distinction: A percentage without its denominator does not determine the sample size, the count, or the standard error. Even when \(n\) is known, a rounded percentage may not uniquely determine \(x\). Do not treat a displayed percentage as more precise than it is.

Worked Example: Recovering a Count From a Percentage Rounded to One Decimal Place

In a hypothetical random sample of 250 adults, a report states that 58.0% support a proposal, rounded to the nearest 0.1 percentage point. Recover the count and calculate a 95% one-proportion z interval.

The displayed percentage corresponds to proportions from approximately 0.5795 to 0.5805. Multiplying these limits by 250 gives a count from 144.875 to 145.125. The only possible integer count is \(x=145\), so \(\hat{p}=145/250=0.580\).

For the calculation, assume the 250 adults were selected using a random sampling process and that the population contains more than 2,500 adults, so the sample is less than 10% of the population. The observed counts are 145 supporting and \(250-145=105\) not supporting, both at least 10. These checks support use of the one-proportion z interval.

For a 95% interval, the standard normal critical value is \(z^*=1.96\). The interval is the sample proportion plus or minus \(z^*\) times its estimated standard error:

\hat{p}\ \pm\ z^*\sqrt{\frac{\hat{p}(1-\hat{p})}{n}}

Substitute the recovered values:

0.580 \pm 1.96\sqrt{\frac{0.580(1-0.580)}{250}} =0.580\pm1.96\sqrt{0.0009744} =0.580\pm0.06118

Thus the interval is approximately \((0.5188,\ 0.6412)\), or 51.88% to 64.12%. The report’s rounding precision identified exactly one possible count here, so the resulting interval is unambiguous to the precision shown.

Calculating the Interval

Once the count and denominator are known, calculate \(\hat{p}=x/n\), check the conditions, and substitute into the interval formula. For the normal approximation used by this interval, the sample should come from a suitable random process, observations should be independent (or the sampling fraction should be small enough to justify the usual approximation), and the sample should contain at least 10 observed successes and 10 observed failures. In this setting, “success” means having the particular outcome being counted; it does not imply that the outcome is desirable.

If the report permits more than one integer count, do not silently choose one. Calculate the interval for each plausible count, or state that the reported information is insufficient to specify one exact interval. The following example illustrates a rounded report that does identify a unique count.

Worked Example: Checking a Report Rounded to One Decimal Place

A hypothetical random sample of 1,000 eligible voters is reported as having 43.7% support for a measure, rounded to the nearest 0.1 percentage point. The sampled population contains more than 10,000 eligible voters. Find the possible count and calculate a 95% interval.

Rounding to the nearest 0.1 percentage point means the underlying percentage is approximately from 43.65% to 43.75%. In proportion form, this is 0.4365 to 0.4375. Multiplying by 1,000 gives a count range from 436.5 to 437.5. The only integer count in this range is \(x=437\), so \(\hat{p}=437/1000=0.437\).

There are 437 observed supporters and 563 non-supporters, so both counts exceed 10. The stated population size makes the sample less than 10% of the population, and the random sample supports the sampling and independence conditions. Use \(z^*=1.96\) for a 95% interval:

0.437 \pm 1.96\sqrt{\frac{0.437(1-0.437)}{1000}} =0.437\pm1.96\sqrt{0.000246031} =0.437\pm0.03074

The interval is approximately \((0.4063,\ 0.4677)\), or 40.63% to 46.77%. The point estimate and interval are based on the uniquely recoverable count of 437, not on treating the printed “43.7%” as infinitely precise.

When Rounding Leaves Several Possible Counts

A percentage rounded to a whole percentage point can conceal more information than one rounded to a tenth. Whether that matters depends on the denominator. With a large enough denominator, several integer counts can fall within the range that rounds to the same displayed percentage. Since the interval uses \(\hat{p}\), different plausible counts can produce slightly different intervals.

Worked Example: Several Counts Behind a Whole-Percentage Report

A hypothetical random sample of 350 households is reported as having 62% use a particular service, rounded to the nearest whole percentage point. Assume the households were randomly sampled from a population larger than 3,500 households. Determine the possible counts and show how the resulting 95% intervals differ.

A displayed 62% rounded to the nearest whole percentage point corresponds approximately to proportions from 0.615 to 0.625. Multiplying by 350 gives counts from 215.25 to 218.75. The possible integer counts are therefore 216, 217, and 218. The report does not identify which count occurred.

For all three candidates, the success and failure counts exceed 10: the smallest success count is 216, and the smallest failure count is \(350-218=132\). The random sampling and population-size statements support the other conditions. Using \(z^*=1.96\), calculate the interval for each candidate:

\begin{aligned} x=216:\quad \hat{p}&=\frac{216}{350}=0.6171,\\ 0.6171\pm1.96\sqrt{\frac{0.6171(1-0.6171)}{350}} &\approx 0.6171\pm0.05093 =(0.5662,\ 0.6681);\\[4pt] x=217:\quad \hat{p}&=\frac{217}{350}=0.6200,\\ 0.6200\pm1.96\sqrt{\frac{0.6200(1-0.6200)}{350}} &\approx 0.6200\pm0.05085 =(0.5691,\ 0.6709);\\[4pt] x=218:\quad \hat{p}&=\frac{218}{350}=0.6229,\\ 0.6229\pm1.96\sqrt{\frac{0.6229(1-0.6229)}{350}} &\approx 0.6229\pm0.05078 =(0.5721,\ 0.6736). \end{aligned}

Each possible interval is close to the others, but they are not identical. The displayed percentage and denominator alone do not justify reporting one of them as the uniquely calculated interval. A careful analysis would seek the unrounded count or explain that the interval depends on which of the three counts was observed.

A Short Justification for the Rounding Check

The count check follows directly from the definition of a rounded percentage. If the displayed percentage is \(R\) and the report rounds to the nearest \(d\) percentage points, the unrounded percentage must lie within about \(d/2\) percentage points of \(R\). In proportion units, that range is centered at \(R/100\) and has half-width \(d/200\). Since a proportion from a sample of size \(n\) equals \(x/n\), multiplying the range by \(n\) gives the corresponding possible values of \(x\). Only integer values can be actual counts.

\frac{R-d/2}{100} \ \lesssim\ \frac{x}{n}\ \lesssim\ \frac{R+d/2}{100} \quad\Longrightarrow\quad n\frac{R-d/2}{100} \ \lesssim\ x\ \lesssim\ n\frac{R+d/2}{100}

The approximate boundary notation acknowledges that exact halfway cases depend on the reporting convention for ties. In ordinary applications, the important step is to retain the full range implied by the stated precision and identify every integer count in it. If no integer count is possible, the reported percentage, denominator, and assumed rounding rule are inconsistent; do not force a count by rounding the product.

After recovering each possible count, the interval calculation is standard: use that count to obtain \(\hat{p}\), then use \(\sqrt{\hat{p}(1-\hat{p})/n}\) as the estimated standard error in the one-proportion z interval. The uncertainty due to sampling is represented by the interval formula; uncertainty about which count was hidden by rounding is a separate information limitation.

Common Mistakes and Practical Checks

Using a percentage without a denominator. The percentage alone cannot determine the number of sampled individuals or the standard error. Find the sample size before calculating an interval.
Rounding the candidate count automatically. If \(n(R/100)\) is not an integer, rounding it to the nearest whole number may be plausible but is not proof that it was the observed count. Check all integer counts compatible with the report’s precision.
Ignoring the precision stated in the report. A whole-number percentage and a percentage reported to one decimal place imply different ranges of possible values. Use the actual displayed precision.
Proceeding when no integer count is compatible. If the allowable count range contains no integer, revisit the denominator, the meaning of the reported percentage, or the assumed rounding convention.
Checking only the percentage, not the conditions. Even a perfectly recovered count does not make the z interval appropriate automatically. Check random sampling, independence, and the success-failure condition.

Key takeaway: A reported percentage can be converted into an interval only when the denominator and the report’s precision are understood. Recover all compatible integer counts first; calculate a separate one-proportion z interval for each if the count is not unique.

Check Your Understanding

For each question, attend to both the denominator and the precision of the reported percentage.

A survey of 500 people reports 36.0% with an outcome, rounded to the nearest 0.1 percentage point. What integer count or counts are compatible with the report?
Why can’t a reported value of 72% by itself determine a one-proportion confidence interval?
A sample of 240 is reported as 51% with an outcome, rounded to the nearest whole percentage point. List the integer counts that could produce this report.
What two observed counts should be checked for the success-failure condition when calculating an interval from \(x\) successes out of \(n\)?
If a rounded percentage and its stated denominator imply no integer count, what is a better response than rounding the calculated count?