How this instrument works
Sampling error, in this context, is the standard error of a sample proportion — a measure of how much p-hat would bounce around if you drew a fresh sample of the same size from the same population, over and over. It answers a precision question, not an accuracy question: it says nothing about whether your sampling method itself was biased, only how much random sample-to-sample variation to expect given your sample size and the proportion you observed.
The formula, SE = sqrt(p-hat(1-p-hat)/n), depends on two things: how close p-hat sits to 0.5, and how large n is. The product p-hat(1-p-hat) is largest — and so is SE — when p-hat is exactly 0.5, and it shrinks as p-hat moves toward 0 or 1, because a near-unanimous result has less room to swing between samples. Growing n always shrinks SE, but through a square root, so quadrupling your sample size only halves the sampling error, not divides it by four.
This is the proportion-specific counterpart to this site's standard error instrument, which computes the standard error of a mean (SE = s/sqrt(n)) for numeric data instead. Both describe the precision of an estimate rather than the spread of raw data, but they use different formulas because a proportion's own variability, p(1-p), is determined entirely by the proportion itself rather than needing a separately measured standard deviation. Multiply this sampling error by a z critical value and you get the margin of error reported alongside a poll percentage.
- Enter your observed proportion into Sample proportion (p-hat) — get this from this site's p-hat instrument if you only have raw counts of successes and sample size.
- Enter your total sample count into Sample size (n).
- Read Sampling error (standard error of p-hat) — the typical amount p-hat would shift if you resampled at this size.
- To turn this into a reported margin of error, multiply the result by your chosen confidence level's z critical value, such as 1.96 for 95% confidence.
- p-hat must be between 0 and 1 — enter it as a decimal (0.3, not 30) or use the p-hat instrument first to compute it from raw counts.
Worked example — p-hat = 0.3, n = 150
Enter 0.3 into Sample proportion (p-hat) and 150 into Sample size (n) — the same 45-out-of-150 survey from this site's p-hat instrument. The instrument computes 0.3 x 0.7 = 0.21, divides by 150 to get 0.0014, then takes the square root: Sampling error reads 0.037417, or about 0.0374.
To build a 95% margin of error from this, multiply by z = 1.96: 1.96 x 0.037417 is about 0.0733, or roughly 7.3 percentage points. Reported as a poll result, that's '30% ± 7.3 points at 95% confidence' — a fairly wide margin, reflecting the moderate sample size of 150.
Questions
Why does sampling error peak when p-hat is 0.5?
Because the term p-hat(1-p-hat) is largest exactly at p-hat = 0.5 (giving 0.25) and shrinks toward 0 as p-hat approaches either extreme. Intuitively, a near-unanimous result, with p-hat near 0 or 1, has little room to vary between samples, while a roughly even split has the most room to swing — so a 50/50 poll result carries more sampling error than a 95/5 one at the identical sample size.
How is this different from the standard error calculator on this site?
That instrument computes the standard error of a mean, SE = s/sqrt(n), for numeric data using a separately measured standard deviation s. This instrument computes the standard error of a proportion, using p-hat(1-p-hat) in place of a standard deviation, because a proportion's variability is fully determined by the proportion value itself — no separate spread measurement is needed.
How do I turn sampling error into a margin of error for my poll?
Multiply this sampling error by the z critical value for your desired confidence level — 1.96 for 95% confidence is the most common choice. In the worked example, 1.96 x 0.037417 is about 0.0733, so a poll reporting 30% with this sample size would typically state a margin of error of roughly ±7.3 percentage points.
Does a bigger sample size always shrink sampling error?
Yes, but slowly, because n sits under a square root. Quadrupling your sample size, say from 150 to 600, only halves the sampling error rather than dividing it by four — which is why national polls often need thousands of respondents to push sampling error down to just a percentage point or two.
Can I calculate sampling error if I only have raw counts, not p-hat yet?
Yes — compute p-hat first. This site's p-hat instrument takes your number of successes and total sample size and returns p-hat = x/n, which you can then feed directly into this instrument's Sample proportion (p-hat) field.