Use confidence intervals without pretending small campaigns are precise

A small creator campaign can leave a wide range of plausible results, even when you count every survey answer correctly. In the hypothetical example below, 12 positive answers out of 40 give a 30% observed rate and a 95% Wilson confidence interval of about 18% to 45%. That interval describes sampling uncertainty under stated assumptions. It does not measure selection bias, prove campaign impact, or predict the next campaign.

For influencer campaign sample size, first ask what the sample contains. Forty people, forty posts, and forty creators answer different questions.

Define the result before choosing an interval

Suppose you want to estimate the share of a defined audience that recognizes your brand after a creator campaign. Write down:

• Population: the people whose recognition rate you want to estimate.
• Unit: one eligible person, counted once.
• Outcome: a yes or no answer to the same recognition question.
• Timing: one declared survey window after the campaign.
• Selection: how each person entered the survey and who did not respond.

Those choices determine whether the calculation below fits. A count of comments does not become a sample of independent people merely because it is large. Likewise, surveying customers answers a question about customers; extending that answer to everyone who saw a creator requires further evidence.

Modash's survey of influencer marketers describes several ways teams measure awareness campaigns. Choosing a measure is only the first step. To define the audience and questionnaire, use a declared brand-awareness survey method before attaching an uncertainty range.

Calculate one hypothetical proportion

Assume a large, defined population. You randomly select 40 eligible people, every selected person responds, and each contributes one independent yes/no answer. Treat the sampling fraction as negligible. Twelve answer yes and 28 answer no.

This is a teaching dataset, not a result from a Virev customer. It deliberately excludes weighting, repeat respondents, and recruitment through a voluntary creator poll.

Use the Wilson score interval documented by NIST. It avoids the impossible negative lower limits that a common symmetric normal interval can produce.

The calculation uses a shifted center and a half-width. The interval need not sit symmetrically around the observed 30%.

d = 1 + z*z/n
center = (p + z*z/(2*n)) / d
half_width = z * sqrt(p*(1-p)/n + z*z/(4*n*n)) / d
lower = center - half_width
upper = center + half_width

Substituting the hypothetical inputs gives a center of approximately 0.3175 and a half-width of 0.1368. The limits are 0.1807 and 0.4543, or 18.1% to 45.4%. Report whole percentages in the headline; keep the extra digits for calculation checks.

You can reproduce the result with Python's standard library:

from math import sqrt

x, n, z = 12, 40, 1.96
p = x / n
d = 1 + z*z / n
center = (p + z*z / (2*n)) / d
half_width = z * sqrt(p*(1-p)/n + z*z/(4*n*n)) / d
print(round(100*(center-half_width), 1))  # 18.1
print(round(100*(center+half_width), 1))  # 45.4

The observed sample rate remains 30%. The shifted center is part of the interval calculation, not a replacement for that observed rate.

Read the range without adding claims

A 95% confidence level describes the interval procedure's repeated-sampling coverage. NIST explains this distinction: confidence belongs to the method across repeated samples. Wilson coverage is approximate because binary counts are discrete.

Avoid saying there is a 95% chance that this particular fixed population rate lies inside these bounds. Also avoid treating the interval as a forecast for the next campaign, which may involve different creators, audiences, and conditions.

For a hypothetical planning threshold of 40% recognition, the interval crosses the threshold. This sample does not resolve whether the underlying recognition rate exceeds it. A larger, well-designed sample could reduce sampling uncertainty. Whether collecting it is worth the cost depends on the decision you would change.

The interval also says nothing by itself about how much recognition the campaign caused. A post-campaign rate lacks the comparison needed to estimate that effect. For that separate question, plan a measurement of incremental impact.

Keep selection bias outside the sampling interval

Now change one fact in the hypothetical example. The same 40 answers came from people who volunteered after seeing a creator's survey link.

The arithmetic still returns 18.1% to 45.4%. Its interpretation changes. People who already like the brand may be more willing to respond. People who never saw the link cannot enter through it. The formula contains no adjustment for either selection process.

AAPOR's guidance on sampling error limits conventional sampling margins of error to probability-based surveys. It also explains that questionnaire and other survey errors fall outside those margins. A bigger opt-in sample can produce a narrower calculated interval while retaining selection bias.

For an ordinary voluntary poll, report the count and respondent rate. Describe recruitment and avoid presenting this Wilson calculation as a population margin of error. Any model-based inference needs its assumptions and limitations stated separately. AAPOR's survey practices distinguish random sampling from social-media opt-in recruitment and call for care when analyzing nonprobability samples.

Use this decision table before adding an interval:

AAPOR notes that sample design and weighting can change sampling error. Raw response count alone therefore cannot establish precision. For missing responses, use a report that keeps unknown values visible.

Put the limits beside the result

This completed reporting template fits the random-sample hypothetical:

We recorded 12 yes answers among 40 randomly selected eligible respondents, an observed recognition rate of 30%. The approximate 95% Wilson confidence interval is 18% to 45%, assuming independent responses and the stated sampling design. The interval covers sampling uncertainty only. It excludes question wording, measurement error, and other possible bias. We cannot estimate campaign-caused lift from this post-campaign survey alone. The range crosses our hypothetical 40% planning threshold, so this sample does not settle that decision.

For the voluntary-poll version, replace the interval sentence with: "These results describe the 40 volunteers who responded. We have not estimated a population margin of error."

Before sending your next campaign report, write its population, selection method, numerator, and denominator beside the percentage. Then decide whether an interval is justified and which decision its width leaves unresolved.