From Binary Outcomes to a Sample Proportion

From Individual Outcomes to a Proportion

Many questions about a population have a yes-or-no form: Does a patient experience a specified side effect? Does a product pass inspection? Does a registered voter support a proposal? For each individual, the outcome is recorded in one of two categories. Once the categories and the population of interest are clearly defined, the number in one category can be summarized as a proportion.

The proportion in the population is the target quantity. A sample proportion is calculated from observed data and serves as an estimate of that target. The distinction matters: the sample proportion is available once the sample has been counted, while the population proportion is generally unknown.

The success category must be stated before the count is interpreted. For example, in a quality-control setting, “passes inspection” might be defined as success. Counting the failures instead would produce a different sample proportion: the proportion of observed items that fail. Either category can be designated as success, but the designation must remain consistent between the data, the calculation, and the population quantity being estimated.

Counting the Right Outcomes and Trials

In \(\hat{p}=x/n\), the numerator \(x\) is the number of observed successes, and the denominator \(n\) is the number of eligible trials or individuals whose outcomes are included. The denominator is not automatically the number originally contacted, enrolled, or listed. It must match the set of outcomes used to form the numerator and the question being answered.

A “trial” here means one eligible opportunity to record the binary outcome; it does not imply a laboratory experiment. If the question concerns the proportion of returned questionnaires that report a particular answer, the denominator is the number of questionnaires with an answer to that question. If it concerns the proportion of all sampled people with that answer, unanswered questionnaires do not provide observed binary outcomes for that item. How missing outcomes should be handled depends on the question and data collection; they should not silently be counted as failures.

When each eligible individual contributes one outcome, the calculation can also be written as the average of zero-one indicators. Let \(Y_i=1\) if individual \(i\) has the designated success outcome and \(Y_i=0\) otherwise. Then the number of successes is \(x=\sum_{i=1}^{n}Y_i\), so

This identity explains why a proportion is an average of binary values: every success contributes 1, every failure contributes 0, and their average is the fraction of successes.

What the Sample Proportion Estimates

The population proportion \(p\) refers to a specified population and a specified success outcome. For example, \(p\) might be the fraction of all items produced in a defined production period that meet an inspection rule. The sample proportion \(\hat{p}\) refers to the observed sample from that population. It is computed from data, so it is known after the sample outcomes have been recorded. The value of \(p\) is ordinarily not directly known, because observing every member of the population is often impractical.

The notation helps keep these quantities separate. The symbol \(p\) refers to the population parameter, while \(\hat{p}\) refers to the sample statistic. The hat signals that the sample value is being used as an estimate of the population quantity. Although \(\hat{p}\) is used to estimate \(p\), the two symbols do not mean the same thing, and their values need not be equal.

Whether a sample proportion is informative about \(p\) depends on how the sample was obtained and how the outcome was measured. A carefully defined calculation cannot fix a sample that fails to represent the target population. For now, the main task is to identify the population quantity the sample is meant to estimate, state the observed sample and success definition, and calculate the corresponding ratio. Questions about how sample proportions vary and how uncertainty is quantified are separate topics.

Why the Denominator Needs Care

A proportion is only meaningful relative to its denominator. A reported percentage without a clear denominator can be ambiguous, even when the arithmetic is correct. The denominator should include the individuals eligible for the question and represented by the counted outcomes. State it explicitly before calculating.

For instance, consider a hypothetical online questionnaire sent to 200 people. Of the 160 who return it, 120 answer “yes” to a particular item. If the question is “What fraction of respondents who answered this item said yes?”, then the observed sample proportion is \(120/160=0.75\), provided all 160 respondents answered the item. If only 150 answered the item, then that question’s denominator is 150, not 160 or 200. The fraction of everyone contacted who both responded and answered yes would instead use 200 as the denominator, but that is a different quantity and should be described differently.

Thus, a correct calculation starts with a clearly worded question. Specify the population or group, the binary outcome, the success category, and which individuals count as eligible trials. Then use the number of successes over the number of those trials.

A Short Justification: Why the Average Equals the Proportion

Suppose the observed outcomes are represented by \(Y_1,\ldots,Y_n\), where each \(Y_i\) equals 1 for success and 0 for failure. Adding these values counts successes: each success adds 1, and each failure adds 0. Therefore \(\sum_{i=1}^{n}Y_i=x\). Dividing that total by the number of observations gives the arithmetic average:

For example, if the outcomes are success, failure, success, failure, failure, their indicators are \(1,0,1,0,0\). Their sum is 2 successes, and their average is \(2/5=0.40\), exactly the fraction of the five outcomes that are successes. This is a direct algebraic identity; it does not require a probability approximation or a confidence interval.

Common Mistakes

• Reversing the count and the total. The number of successes goes in the numerator and the number of eligible trials goes in the denominator: \(\hat{p}=x/n\), not \(n/x\).
• Using “success” to mean a good outcome. Success is simply the category being counted. It can be “has a symptom,” “fails inspection,” or any other consistently defined category.
• Treating \(\hat{p}\) as the known population value. Saying “58% of those surveyed said yes” describes the sample. Saying “58% of all registered voters support the measure” makes a claim about \(p\) that the sample calculation alone does not establish.
• Choosing a denominator without matching it to the question. A contacted group, a respondent group, and a group with a recorded answer may have different sizes. State which group the proportion describes before dividing.
• Counting unanswered outcomes as failures without justification. An unanswered item is not automatically a “no.” It may not be an observed binary outcome at all.
• Reporting a decimal without context. A value such as 0.25 is a proportion, but readers need to know which outcome and which group it describes. State it in words and, when helpful, as a percentage.