Accuracy vs precision in statistics

By Jude Wallis · Published

Accuracy is how close measurements or estimates are to the true value. Precision is how close repeated results are to each other. They are independent: a scale that always reads 4 grams too high is precise and inaccurate. In statistics, accuracy means low bias and precision means low variability.

Accuracy versus precision is general statistics background and is not a named topic in the Fall 2026 AP Statistics framework. The closest course material is the comparison of bias and variability, covered on this site at bias vs variability, along with the width of a confidence interval as a measure of how precise an estimate is.

Accuracy vs precision: the short answer

Repeat a measurement or an estimate several times and you can ask two separate questions about the results. Accuracy asks whether they are centered on the true value. Precision asks whether they agree with each other. The first is about where the results land on average. The second is about how tightly they cluster.

Neither one implies the other. Results can cluster tightly around the wrong value, which is precise and inaccurate. Results can scatter widely around the right value, which is accurate on average and imprecise. A good method needs both.

In the language of statistics, a lack of accuracy is bias, and a lack of precision is variability. The comparison of those two terms as the AP course uses them is in bias vs variability.

What accuracy means

A measuring process or an estimator is accurate when its results are centered on the true value. The gap between the center of the results and the truth is the bias. A bathroom scale that reads 2 pounds heavy for everyone, a survey that reaches only people with landline phones, and a ruler printed with slightly short centimeters all produce results that are off in a consistent direction.

Two features of accuracy are worth remembering. First, it cannot be judged from the results alone. A set of readings shows how much they vary, and nothing in them reveals how far their center is from the truth. To assess accuracy you need a known reference value, such as a calibration weight, or an argument about how the data were collected. Second, repetition does not improve it. Averaging more readings from a biased scale, or surveying more people with a biased method, gives a steadier answer that is still wrong by the same amount. That point is developed in does a bigger sample fix bias.

What precision means

A process is precise when repeated results are close to one another. Precision is measured by variability: the standard deviation of repeated measurements, or for an estimate, its standard error and the width of the confidence interval built from it. A smaller standard deviation or a narrower interval means more precision.

Unlike accuracy, precision can be read directly from the data, and it does improve with repetition. The mean of many readings varies less than a single reading, and the standard error of a sample mean or sample proportion shrinks in proportion to 1n\frac{1}{\sqrt{n}}. Quadrupling the sample size halves it.

On this site, the precision glossary entry uses the word in its confidence interval sense: a narrower interval is a more precise estimate. That entry makes the same warning from the other side. A narrow interval tells you the estimate is stable. It does not tell you the estimate is centered on the truth.

The differences side by side

FeatureAccuracyPrecision
Question it answersAre the results centered on the true value?Are repeated results close to each other?
Statistical term for lacking itBiasVariability
Measured byCenter of the results minus the true valueStandard deviation, standard error, or interval width
Visible in the data aloneNo, a reference value is neededYes
Improved by more repetitions or a larger sampleNoYes
Improved byA better instrument, calibration, random samplingLarger samples, more consistent procedure
Target analogyShots centered on the bullseyeShots tightly grouped

The four combinations all occur. A tight group on the bullseye is accurate and precise. A tight group off to one side is precise and inaccurate. A wide scatter centered on the bullseye is accurate on average and imprecise. A wide scatter off to one side is neither.

Which one matters when

Both matter, and they call for different fixes, so the first step is to work out which one a problem is about.

If results are consistent and wrong, the problem is accuracy. Look at the instrument and the data collection method. For a measurement, calibrate against a known standard. For a survey, the remedy is random selection from the whole population, since a method such as voluntary response over-represents some groups however many people respond.

If results are right on average and too scattered to be useful, the problem is precision. Take more measurements and average them, or increase the sample size. The sample size calculator shows how large a sample a given margin of error requires.

A known, constant bias is often the easier problem. A scale that reads exactly 4 grams high can be corrected by subtracting 4. Random scatter cannot be subtracted away, only reduced by averaging.

The classic mix-up and how to avoid it

The most damaging error is taking precision as evidence of accuracy. A result reported to many decimal places, or a poll with a very small margin of error, looks trustworthy. The decimal places and the margin describe precision only. A margin of error accounts for the variability of random sampling and includes no allowance for bias, so a huge biased sample yields a narrow interval around the wrong value. The second worked example shows a poll of 10,000 that misses the truth by 12 points with a margin of error under 1 point.

The second error is using the two words interchangeably. In everyday speech they overlap. In statistics and in science, saying that a measurement is precise makes no claim about whether it is correct.

The third error is expecting a larger sample to repair accuracy. Sample size acts on variability. It leaves bias untouched. Ask how the data were produced before asking how many observations there are.

Two scales and a 100-gram standard

A reference weight known to be exactly 100.0 grams is weighed five times on each of two scales. Scale A reads 103.9, 104.0, 104.1, 104.0, 104.0. Scale B reads 97.5, 102.0, 99.0, 103.5, 98.0. Find the mean and standard deviation of each set of readings and describe each scale's accuracy and precision.

  1. Scale A mean: 103.9+104.0+104.1+104.0+104.0=520.0103.9 + 104.0 + 104.1 + 104.0 + 104.0 = 520.0, so xˉ=520.05=104.0\bar{x} = \frac{520.0}{5} = 104.0 grams.

  2. Scale A standard deviation: the deviations from 104.0 are −0.1,0,0.1,0,0-0.1, 0, 0.1, 0, 0, with squares summing to 0.01+0.01=0.020.01 + 0.01 = 0.02. So s=0.024=0.005≈0.07s = \sqrt{\frac{0.02}{4}} = \sqrt{0.005} \approx 0.07 grams.

  3. Scale B mean: 97.5+102.0+99.0+103.5+98.0=500.097.5 + 102.0 + 99.0 + 103.5 + 98.0 = 500.0, so xˉ=500.05=100.0\bar{x} = \frac{500.0}{5} = 100.0 grams.

  4. Scale B standard deviation: the deviations from 100.0 are −2.5,2.0,−1.0,3.5,−2.0-2.5, 2.0, -1.0, 3.5, -2.0, with squares 6.25,4,1,12.25,46.25, 4, 1, 12.25, 4 summing to 27.5. So s=27.54=6.875≈2.62s = \sqrt{\frac{27.5}{4}} = \sqrt{6.875} \approx 2.62 grams.

  5. Accuracy: scale A's readings center on 104.0, which is 104.0−100.0=4.0104.0 - 100.0 = 4.0 grams above the true value. Scale B's readings center on 100.0, the true value.

  6. Precision: scale A's readings vary by about 0.07 grams. Scale B's vary by about 2.62 grams, roughly 37 times as much.

Scale A is precise and inaccurate: its readings have a standard deviation of only about 0.07 grams and all sit about 4.0 grams too high. Scale B is accurate on average and imprecise: its readings center on 100.0 grams, with a standard deviation of about 2.62 grams, so any single reading may be off by several grams. Scale A's error could be corrected by subtracting 4.0 from every reading; scale B needs many readings averaged.

A large biased poll against a small random one

Before a vote in which the true support for a measure turns out to be 50%, two polls are taken. A website poll that anyone can answer gets 10,000 responses with 62% in favor. A random sample of 400 voters finds 52% in favor. Compute a 95% margin of error for each as if both were random samples, and compare the intervals with the true value.

  1. Website poll standard error: 0.62×0.3810000=0.00002356≈0.00485\sqrt{\frac{0.62 \times 0.38}{10000}} = \sqrt{0.00002356} \approx 0.00485.

  2. Website poll margin of error: 1.96×0.00485≈0.00951.96 \times 0.00485 \approx 0.0095, about 1 percentage point. The interval is 0.62±0.00950.62 \pm 0.0095, from 0.6105 to 0.6295.

  3. Random sample standard error: 0.52×0.48400=0.000624≈0.02498\sqrt{\frac{0.52 \times 0.48}{400}} = \sqrt{0.000624} \approx 0.02498.

  4. Random sample margin of error: 1.96×0.02498≈0.04901.96 \times 0.02498 \approx 0.0490, about 4.9 percentage points. The interval is 0.52±0.04900.52 \pm 0.0490, from 0.4710 to 0.5690.

  5. Compare precision: the website poll's interval is about 0.019 wide and the random sample's is about 0.098 wide, roughly five times wider. The website poll is far more precise.

  6. Compare with the truth: 0.50 lies inside 0.4710 to 0.5690 and far outside 0.6105 to 0.6295. The website poll's estimate is off by 0.62−0.50=0.120.62 - 0.50 = 0.12, more than twelve times its computed margin of error.

  7. Explain: the website respondents chose to take part, so they were a voluntary response sample, and the margin of error formula does not apply to them. It measures random sampling variability only and cannot detect the bias.

The website poll gives 0.62 with a computed margin of about 0.0095; the random sample gives 0.52 with a margin of about 0.0490. The website poll is more precise and badly inaccurate: its narrow interval, 0.6105 to 0.6295, misses the true 0.50 by 12 points. The random sample is less precise and accurate: its interval, 0.4710 to 0.5690, contains 0.50. Sample size bought precision, and random selection is what protected accuracy.

Frequently asked questions

Can a measurement be precise but not accurate?

Yes. Precise means repeated results agree with each other, and they can agree on the wrong value. A scale that reads 104.0 grams every time for a 100.0 gram weight is very precise and is off by 4.0 grams on every reading.

Does a larger sample make an estimate more accurate?

A larger random sample makes an estimate more precise, because the standard error shrinks as the sample grows. It does not remove bias. If the sampling method favors some outcomes, a larger sample gives a tighter estimate of the wrong value.

Is accuracy vs precision on the AP Statistics exam?

The pair of words is general statistics and science background and is not a named topic in the course. The ideas behind it are tested under other names: bias and variability of an estimator, the problems that make a sample unrepresentative, and the width of a confidence interval.