Conservative Sample Size Planning

Why Use a Conservative Planning Value?

A study needs enough observations to estimate a population proportion with useful precision. But the sample size calculation depends on the proportion itself, and before collecting data that value may be unknown. For example, a team planning a poll might want its estimate to be within 4 percentage points of the population proportion, but have no defensible estimate of how common the response will be.

For this situation, use \(p=0.5\) in the planning calculation. This value does not assert that half the population has the characteristic of interest. Instead, it is a conservative choice: among all possible proportions, it produces the largest variance and therefore the largest required sample size under the usual one-proportion normal-approximation margin-of-error formula.

Definition: Conservative sample size planning for a one-proportion estimate means choosing a sample size that meets a specified margin-of-error target using the largest possible value of the planning variance \(p(1-p)\). When no prior estimate is available, this maximum occurs at \(p=0.5\).

The Planning Formula

Suppose each sampled individual has a binary outcome, such as “supports the proposal” or “does not support the proposal.” The population proportion with the outcome is \(p\), and \(\hat p\) is the sample proportion. For a planned confidence level, let \(z^*\) be the corresponding positive standard-normal critical value, and let \(E\) be the desired margin of error, expressed as a proportion rather than as a percentage number.

For planning with the one-proportion normal-approximation interval, the anticipated margin of error is

E_{\text{planned}}=z^*\sqrt{\frac{p(1-p)}{n}}.

This expression uses the unknown population proportion \(p\) to describe the variability the study is designed to handle. Setting the planned margin of error no larger than the target \(E\), then solving for \(n\), gives the sample-size requirement for a specified planning value \(p\).

z^*\sqrt{\frac{p(1-p)}{n}}\le E \quad\Longrightarrow\quad n\ge \frac{(z^*)^2p(1-p)}{E^2}.

Because \(n\) must be a whole number of observations, the calculated requirement is rounded up to the next integer. When no prior estimate is available, substitute \(p=0.5\). The resulting conservative formula is

n_{\text{planned}} = \left\lceil\frac{(z^*)^2(0.5)(0.5)}{E^2}\right\rceil = \left\lceil\frac{(z^*)^2}{4E^2}\right\rceil.

The ceiling symbols mean to take the smallest whole number at least as large as the quantity inside them. The result is a planning target under the stated normal-approximation framework, not a guarantee that every confidence interval computed from the eventual data will have exactly the requested width.

Why \(p=0.5\) Gives the Largest Requirement

The key quantity in the formula is \(p(1-p)\), which measures the variability of a binary outcome. Its maximum can be seen directly by completing the square:

p(1-p)=p-p^2=\frac14-\left(p-\frac12\right)^2\le\frac14.

The squared term is never negative, so \(p(1-p)\) cannot exceed \(1/4\). Equality occurs when \(p=1/2\). Since \(z^*\) and \(E\) are fixed and positive, the required sample size \((z^*)^2p(1-p)/E^2\) increases as \(p(1-p)\) increases. Therefore, using \(p=0.5\) gives the largest required sample size among all possible population proportions.

This is the precise sense in which the calculation is conservative. If the true \(p\) is closer to 0 or 1, then \(p(1-p)\) is smaller, and the usual formula would require fewer observations to reach the same target margin of error. A smaller actual variance does not make \(p=0.5\) a better guess about the population; it only explains why the \(p=0.5\) plan is cautious.

Assumptions Behind the Plan

A sample-size formula inherits assumptions from the interval method it is designed to support. Before applying this calculation, the study should be planned around a clearly defined binary outcome and a random sample or another sampling design that justifies inference to the target population. Observations must be independent; for sampling without replacement, the sample should ordinarily be no more than 10% of the population when that population is finite.

The margin-of-error formula is based on the normal approximation for a sample proportion. In an actual study, the success-failure condition is checked using the observed counts: there should be enough observations with the outcome and enough without it. A planned \(p=0.5\) does not prove that these counts will be adequate if the true proportion is near a boundary. Thus, conservative variance planning does not remove the need to assess the interval conditions once data are collected.

Finally, the formula concerns sampling variability. It does not correct for nonresponse, measurement error, coverage problems, or a sampling process that systematically misses part of the target population. A large sample can estimate the wrong target very precisely if the study design is biased.

Worked Example: Planning a 95% Poll Within 4 Percentage Points

A hypothetical polling organization wants a 95% confidence interval for the proportion of eligible voters who support a local measure. It has no prior estimate of support and wants a margin of error no greater than 4 percentage points. How many voters should it plan to sample?

The outcome is binary: each sampled voter either supports the measure or does not. Assume the organization can obtain a random sample, that responses are independent (or that the sample is no more than 10% of the relevant voter population if sampling without replacement), and that the normal-approximation planning framework is appropriate. Since no prior proportion estimate is available, use the conservative value \(p=0.5\). For 95% confidence, use \(z^*=1.96\). Four percentage points is \(E=0.04\) on the proportion scale.

Substitute into the conservative formula:

n\ge\frac{(1.96)^2}{4(0.04)^2} =\frac{3.8416}{0.0064} =600.25.

The organization must plan for at least 601 voters, rounding up to a whole observation. This is the conservative sample-size target for the stated confidence level and margin of error under the planning assumptions; it is not a claim that exactly half of voters support the measure.

Worked Example: A More Demanding Confidence Level and Precision

A hypothetical community survey wants to estimate the proportion of residents who favor a proposed transit change. With no prior estimate, the team requests 90% confidence and a margin of error of 5 percentage points. Assume a binary response, an appropriate random sampling plan with independent observations, and a population large enough for the intended sample to satisfy the independence requirement. Use the normal-approximation planning formula.

For 90% confidence, take \(z^*=1.645\). The target margin is \(E=0.05\). With no prior estimate, use \(p=0.5\):

n\ge\frac{(1.645)^2}{4(0.05)^2} =\frac{2.706025}{0.01} =270.6025.

The planned sample size is therefore 271 residents. The value \(p=0.5\) is being used to protect against the largest variance, not because the team expects exactly half of residents to favor the change. The assumptions about random sampling and independence concern how the data will be collected; the calculation alone cannot establish that they will hold.

Worked Example: A Narrower Target Margin

A hypothetical health survey aims to estimate the proportion of adults who received a particular vaccine. It has no prior estimate and wants a 95% confidence interval with a margin of error of 2.5 percentage points. Suppose the survey can select a random sample and the planned sample is small relative to the adult population, so independence is reasonable. The response is binary, and the calculation is intended for the usual normal-approximation method.

Use \(z^*=1.96\), \(E=0.025\), and the conservative \(p=0.5\):

n\ge\frac{(1.96)^2}{4(0.025)^2} =\frac{3.8416}{0.0025} =1536.64.

Rounding up gives a planning target of 1,537 adults. The target margin is expressed in proportion units: \(0.025\) corresponds to 2.5 percentage points. This example also illustrates why a demanding precision target can require a large sample even when the conservative formula is straightforward.

How to Interpret the Planning Result

The result should be reported as a design decision, with its conditions made clear: for example, “Plan to sample at least 601 voters to target a 95% confidence interval with a 4-percentage-point margin of error, using the conservative \(p=0.5\) planning value.” This does not say that the eventual interval will have that exact margin of error. The realized interval uses the observed sample proportion, and its actual width can differ from the planned target.

Conservative planning is especially useful when no credible estimate of \(p\) exists and undersizing the study would be costly. It also makes the dependence on precision transparent: in the formula, the required sample size is proportional to \(1/E^2\). Halving the desired margin of error therefore multiplies the required sample size by four, all else equal. The confidence level also matters through \(z^*\): a larger critical value increases the required sample size.

Common Mistakes

Treating \(0.5\) as a forecast. The reasoning “we do not know the proportion, so we predict that half the population has the outcome” confuses a planning device with an estimate. Use \(0.5\) because it maximizes \(p(1-p)\), not because it is a claim about the population.
Entering a percentage number instead of a proportion. A 4-percentage-point margin is \(E=0.04\), not \(E=4\). The formula requires the proportion-scale value.
Using \(p(1-p)\) without checking its maximum. A value such as \(p=0.2\) gives a smaller required sample size than \(p=0.5\). When there is no basis for choosing a different planning value, \(0.5\) gives the conservative requirement.
Assuming a large planned sample fixes every design problem. The calculation addresses sampling variability under its model. It does not make a biased sample representative or repair nonresponse and measurement problems.
Rounding down. A calculated requirement such as 600.25 is not met by 600 observations. The whole-number target must be at least as large as the calculated value.

Key takeaway: With no prior estimate of a population proportion, set the planning value to \(p=0.5\). Because \(p(1-p)\le 0.25\), this gives the largest sample-size requirement under the usual one-proportion normal-approximation margin-of-error formula. State the confidence level, target margin, sampling assumptions, and the conservative nature of the choice.

Check Your Understanding

Try these questions before reviewing the explanations in the tutorial.

Why does \(p=0.5\) give a conservative sample-size calculation when no prior estimate is available?
A study wants a 95% confidence interval with a 3-percentage-point margin of error and has no prior estimate. What values of \(z^*\), \(E\), and \(p\) should be used in the planning formula?
Explain why using \(E=5\) for a 5-percentage-point margin gives an incorrect sample-size calculation.
Does a sample-size plan using \(p=0.5\) guarantee that the success-failure condition will hold in the collected data? Explain.
Name one important source of error that conservative sample-size planning does not correct.