Skip to main content

Parameter vs Statistic: Differences, Symbols, Examples

Parameter vs statistic: a parameter describes a whole population, a statistic one sample. In 10,000 samples of 100, 94.8% of 95% intervals held the true mean.

Parameter vs Statistic: Differences, Symbols, Examples

A parameter is a number that describes a whole population, such as the mean income of every household in a country. A statistic is a number calculated from a sample of that population, such as the mean income of 1,000 surveyed households. The parameter is fixed and usually unknown; the statistic is known and changes from sample to sample. This guide to parameter vs statistic shows both with published figures and a simulation.

Key Takeaways

  • A parameter describes every member of a population. A statistic describes the sample you measured and estimates the parameter.
  • Greek letters usually stand for parameters (μ, σ, ρ, β) and Roman letters for statistics (x̄ or M, s or SD, r, b). The proportion p is the exception.
  • The same number is either one depending on who was measured: 50% of a whole seminar is a parameter, 50% of a sample of voters is a statistic.
  • In our simulated population of 40,000 students, mean weekly study time is 13.90 hours. Across 10,000 samples at each size, samples of 25 gave means from 8.32 to 19.64 hours and samples of 400 stayed between 12.50 and 15.33.
  • Each fourfold increase in sample size cut the spread of the sample means about in half, and 93.8% to 95.1% of the 95% confidence intervals contained the parameter.

What is the difference between a parameter and a statistic?

A parameter summarizes every member of a population; a statistic summarizes only the members you sampled. Because the whole population is rarely measured, the parameter is usually unknown and the statistic stands in for it as an estimate. A parameter has one value for a given population. A statistic takes a different value in every sample, which is why it is reported with a measure of its uncertainty.

OpenStax’s Introductory Statistics 2e defines the pair in two sentences: “A statistic is a number that represents a property of the sample” and “A parameter is a numerical characteristic of the whole population that can be estimated by a statistic.” The population is the full “collection of persons, things, or objects under study”, and a sample is “a portion (or subset)” of it.

ParameterStatistic
What it describesA whole populationA sample drawn from that population
Does it vary?No: one value for a given population at a given timeYes: every sample gives a somewhat different value
Usual notationGreek letters: μ, σ, ρ, βRoman letters: x̄, s, r, b
ExampleMean weekly study hours of all 40,000 students at a universityMean weekly study hours of the 100 students who answered a survey

The first letters are a memory aid: parameter goes with population, statistic with sample.

OpenStax calls organizing and summarizing data “descriptive statistics” and the formal methods for drawing conclusions “inferential statistics”. A sample mean is descriptive while it only summarizes the sample, and inferential once it is used to estimate the population mean.

The vocabulary has been in print since at least 1922. In a paper on the foundations of statistics, R. A. Fisher complained of “a purely verbal confusion”: it was customary, he wrote, to apply the same name (mean, standard deviation, correlation coefficient) “both to the true value which we should like to know, but can only estimate, and to the particular value at which we happen to arrive by our methods of estimation”. Estimation, in his terms, means calculating from a sample “statistical derivates, or as we shall call them statistics, which are designed to estimate the values of the parameters of the hypothetical population.”

Parameter vs statistic symbols

By convention, Greek letters stand for population parameters and Roman letters for the sample statistics that estimate them: μ and x̄ for the mean, σ and s for the standard deviation, ρ and r for a correlation, β and b for a regression slope. The proportion breaks the pattern, with p for the parameter and p-hat for the statistic. N counts the population and n the sample.

QuantityPopulation parameterSample statistic
Meanμ (mu)x̄ (x-bar); M in APA-style papers
Standard deviationσ (sigma)s; SD in APA-style papers
Varianceσ²s²
Proportionpp̂ (p-hat); “P prime” in some textbooks
Correlationρ (rho)r
Regression slopeβ₁b₁
Number of membersNn

The Greek and Roman pairs follow OpenStax. Its introductory text says: “The lower case letter s represents the sample standard deviation and the Greek letter σ (sigma, lower case) represents the population standard deviation.” The same book’s correlation chapter lists ρ as the “population correlation coefficient (unknown)” and r as the “sample correlation coefficient (known; calculated from sample data)”. For regression, Introductory Business Statistics 2e names the rule: “the convention of Greek letters for the population parameters and Roman letters for their estimates.”

Three conventions vary, so check the textbook or journal you are writing for.

  • The sample proportion. OpenStax’s introductory text writes it as p with a prime mark (read “P prime”), while the NIST/SEMATECH e-Handbook of Statistical Methods writes p̂. Both estimate the population proportion p.
  • N and n. OpenStax uses N for “the number of items in the population” and n for the sample, but the NIST handbook writes N for the sample size in its formulas.
  • β. The same letter is the probability of a Type II error in hypothesis testing, so say which β you mean.

How do you tell whether a number is a parameter or a statistic?

Ask three questions. Who is the population the claim is about? Was every member of that population measured? If the study were repeated with different people, would the number come out different? Everyone measured and the same number on a repeat means a parameter. Only some measured and a different number on a repeat means a statistic.

Most parameter vs statistic exercises turn on the first question, because the same group of people is a population in one study and a sample in another. OpenStax’s own example is one math class: its average score is a statistic “if we consider one math class to be a sample of the population of all math classes”. If that class is the only group you want to describe, the same average is a parameter.

In the table below, the first four cases are invented for illustration; the rest are published numbers.

NumberPopulationEveryone measured?Verdict
50% of the 28 students in a seminar passed a quiz, and the seminar is all you want to describeThe 28 studentsYesParameter
50% of 1,000 surveyed voters support a ballot measureAll voters in the stateNoStatistic
Mean GPA of all enrolled students, from the registrar’s recordsAll enrolled studentsYesParameter
Mean GPA of the 200 students who answered a surveyAll enrolled studentsNoStatistic
A coin lands heads in 996 of 2,000 tosses (0.498)Every toss the coin could makeNoStatistic; the parameter is the coin’s probability of heads
331,449,281 U.S. residents on April 1, 2020All U.S. residents on that dayYes, by designParameter (a census count)
U.S. unemployment rate of 4.1% in October 2026The U.S. labor force that monthNo: about 60,000 householdsStatistic

Two shortcuts fail this test. A percentage is not automatically a statistic: the first two rows carry the same 50%, and only one comes from a sample. A large sample does not turn a statistic into a parameter either: the unemployment rate rests on about 60,000 households and is still an estimate. The coin row, from OpenStax, has a population that cannot be listed at all: every toss the coin could ever make.

What are real examples of a parameter and a statistic?

The 2020 Census count of the U.S. resident population, 331,449,281, is a parameter in the textbook sense: it comes from an attempt to count everyone. The U.S. unemployment rate, 4.1% in October 2026, is a statistic: the Bureau of Labor Statistics estimates it each month from a sample of about 60,000 households and says how far sampling could put it off.

A parameter: the 2020 Census. On April 26, 2021, the Census Bureau announced that “the resident population of the United States on April 1, 2020, was 331,449,281”. The release gives the figure to the last person and attaches no margin of error, because nobody was sampled.

A census still errs in another way. The Bureau checked its count with a Post-Enumeration Survey, and the results of March 10, 2022 put the national “net coverage error” at −0.24%, or 782,000 people, with a standard error of 0.25%, “not statistically different from zero”. Note what kind of number −0.24% is. The Post-Enumeration Survey is a sample survey, so its estimate of the census’s error is itself a statistic, published with a standard error.

A statistic: the monthly unemployment rate. The Employment Situation release for October 2026, published on November 6, 2026, put the unemployment rate at 4.1% and the number of unemployed people at 7.1 million. Nobody counted 7.1 million people. BLS’s explainer “How the Government Measures Unemployment” (last modified in October 2015) says “it is impractical to count every unemployed person each month”, so the figures come from the Current Population Survey, a sample of “about 60,000 eligible households”.

The same page states the price of sampling: “A sample is not a total count”, and “the chances are 90 out of 100 that the monthly estimate of unemployment from the sample is within about 300,000 of the figure obtainable from a total census.” The technical note in the October 2026 release calls what the estimates aim at “the true population values”, which is the parameter by another name.

A worked example: one population, many samples

We built a population we could see in full: 40,000 students at an invented university, each with a number of weekly study hours. Its mean is 13.90 hours, a parameter we know because we hold every record. Across 10,000 random samples of 25 students, the sample means ran from 8.32 to 19.64 hours; across 10,000 samples of 400, from 12.50 to 15.33. The parameter never moved. The statistics did.

The population is invented and simulated in Python (seed 20261121). We made study hours right-skewed: the median is 12.80 hours, the mean 13.90 and the range 0.5 to 67.9. Of the 40,000 students, 16,033 have a paid job. So three parameters are known exactly: μ = 13.90 hours, σ = 6.91 hours and p = 40.1%.

Step 1: one sample at each size. We drew one simple random sample of 25 students, one of 100 and one of 400. The last two rows, the standard error and the 95% confidence interval, are explained in the next section; for now, read each interval as that sample’s range of plausible values for μ.

Population (parameter)Sample of 25Sample of 100Sample of 400
Mean weekly study hoursμ = 13.90x̄ = 11.88x̄ = 13.99x̄ = 14.66
Standard deviation (hours)σ = 6.91s = 4.61s = 6.11s = 7.16
Share with a paid jobp = 40.1%52.0% (13 of 25)38.0% (38 of 100)40.5% (162 of 400)
Standard error of the mean, s ÷ √n (hours)None: nothing was sampled0.920.610.36
95% confidence interval for μNot needed[9.98, 13.79][12.78, 15.21][13.96, 15.36]

No sample statistic equals its parameter. The sample of 25 put the mean 2.02 hours too low and the share with a job at 52.0% against a true 40.1%. A real study has one sample column and never the population column.

This run was also unlucky, and we kept it as drawn. The sample of 400 landed 0.76 hours from μ, further off than the sample of 100 (0.09 hours), and two of the three intervals miss 13.90. A larger sample size makes a statistic closer on average, not every time.

Step 2: 10,000 samples at each size. We repeated the draw 10,000 times per sample size.

Across 10,000 samplesn = 25n = 100n = 400
Mean of the sample means (μ = 13.899)13.90313.89613.895
Standard deviation of the sample means (hours)1.4030.6890.343
σ ÷ √n1.3830.6910.346
Middle 95% of the sample means (hours)11.25 to 16.7512.58 to 15.2713.23 to 14.57
95% confidence intervals that contain μ93.8%94.8%95.1%
Middle 95% of the sample shares with a job20.0% to 60.0%30.0% to 50.0%35.2% to 45.0%

Three things stand out.

  1. The statistics center on the parameter. At every sample size the 10,000 sample means average 13.90. Single samples miss, but not in one direction.
  2. The spread follows σ ÷ √n. The standard deviation of the sample means is the standard error, and it stays within 0.03 hours of the formula. Four times the sample cuts it in half, which is why precision gets expensive.
  3. The intervals keep their promise in the long run. About 95% contain μ. At n = 25 the share is 93.8%, a little short because study hours are skewed and the sample is small; the same test on a bell-shaped population built to the same mean and standard deviation gave 94.8%.

How does a statistic estimate a parameter?

A sample mean or proportion from a random sample equals the parameter plus sampling error, a chance gap that averages out across samples and shrinks as the sample grows. The standard error measures the typical size of that gap, and a confidence interval turns it into a range of plausible values for the parameter. Bias is a different problem: it pushes every sample the same way, and a bigger sample does not remove it.

Standard error. OpenStax describes the standard error of the mean, σ ÷ √n, as “a description of how far (on average) that the sample mean will be from the population mean in repeated simple random samples of size n.”

Confidence interval. The NIST handbook defines a 95% confidence level by repetition: if the same population is sampled many times, “the resulting intervals would bracket the true population parameter in approximately 95 % of the cases.” The 95% belongs to the procedure. NIST’s section on confidence limits for the mean adds: “The interval computed from a given sample either contains the true mean or it does not.”

Random error and bias. All of this assumes the sample was chosen by chance, as in random sampling. When some members of the population are more likely to end up in the sample than the design allows for, and they differ on what you are measuring, the statistic is off in one direction however many people you add: our guide to sampling bias covers a poll that had nearly 2.4 million ballots and still missed by 19.6 points. A better design lowers the standard error without adding people, as the stratified sampling guide shows.

The n − 1 divisor. OpenStax says a population variance divides the squared deviations by N and a sample variance by n − 1, because that “gives a better estimate of the population variance.” The simulation shows it. The population variance σ² is 47.78. Across the 10,000 samples of 25, the variance computed with n − 1 averaged 47.99; computed with n, it averaged 46.07, which is 96.4% of σ².

Hypothesis tests. A hypothesis is a statement about a parameter, tested with a statistic: in OpenStax’s correlation chapter the null hypothesis is ρ = 0, and the decision rests on r and n. Our hypothesis examples show the wording.

How do you write parameters and statistics in a paper?

Report what you measured as a statistic: name it, give the sample size and add a confidence interval. Use italic Roman symbols such as M, SD, r and n for sample values, and keep Greek letters for the population values you are estimating or testing. Word your conclusions about the population as estimates, not as facts read off the sample.

APA’s numbers and statistics guide sets the typography: “Use italics for letters used as statistical symbols or algebraic variables”, but “use standard (nonitalic) type for Greek letters.” Symbols go with a mathematical operator (“M = 7.7”), running text uses the word (“the means were”), and M, SD, N and n need no definition. APA’s reporting standards (JARS-Quant) ask for “effect-size estimates and confidence intervals on those estimates that correspond to each inferential test conducted, when possible”.

Instead ofWriteWhy
“Students study 14 hours a week.”“Students in the sample studied a mean of 13.99 hours a week.”The first sentence states a population value nobody measured
“μ = 13.99”“M = 13.99”μ is the parameter; the sample mean is M or x̄
“The population mean is 13.99 hours.”“We estimate the population mean at 13.99 hours, 95% CI [12.78, 15.21].”An estimate travels with its interval

The passage below was written for this guide as an illustration. It reports the sample of 100 from the worked example.

Results (illustration). The 100 students in the sample reported studying a mean of 13.99 hours a week (SD = 6.11), 95% CI [12.78, 15.21]. Thirty-eight of them (38%) had a paid job.

Discussion (illustration). If the sample represents the student body, students at this university study about 14 hours a week on average, and the interval puts the population mean between roughly 13 and 15 hours.

The results sentences describe the sample. The discussion sentence makes a claim about the population, names its condition and keeps the interval in view. Here we know that μ = 13.90 sits inside the interval. A real study never gets to check.

If you measured the whole population, such as every student in one seminar, say so and report the figures as they are: there is no sampling error when nothing was sampled. The order and format of t statistics, p values and effect sizes are in our guide on how to write a results section.

Where Papero fits

Papero is an AI writing workspace for students and researchers. It is a writing and citation tool, not a statistics package: it does not calculate your statistics for you. It helps with the sentences around your numbers.

Paste a DOI to cite the source of a definition or a method, search more than 250 million papers through OpenAlex in Find papers, and run Review’s Claims check to find sentences that need a citation. The research paper writing page shows the process from an IMRaD outline to a Word manuscript.

Frequently Asked Questions

Is 50% a parameter or a statistic?

It depends on who was measured. If 50% describes every member of the population, such as all 28 students in a seminar, 50% is a parameter. If 50% comes from a sample, such as 1,000 surveyed voters standing in for all voters in a state, 50% is a statistic, and another sample would give a slightly different percentage.

What is an example of a parameter?

The mean amount that all first-year students at a college spend on school supplies is a parameter, because it describes the whole population. A published example is the 2020 Census count of the U.S. resident population, 331,449,281 on April 1, 2020, which comes from an attempt to count every resident.

What is an example of a statistic?

The mean amount spent on school supplies by 100 randomly surveyed first-year students is a statistic, because it describes a sample. A published example is the U.S. unemployment rate of 4.1% for October 2026, which the Bureau of Labor Statistics estimated from a monthly survey of about 60,000 households.

Is the mean a parameter or a statistic?

The mean is either one, depending on the data behind it. The mean of an entire population is a parameter, written μ. The mean of a sample is a statistic, written x̄ or, in APA-style papers, M. The same holds for the standard deviation: σ for a population, s or SD for a sample.

Can a parameter change?

Yes, but only when the population itself changes. A parameter does not vary from sample to sample: for a defined population at a defined time it has one value. The U.S. resident population, for example, was 308,745,538 in the 2010 Census and 331,449,281 in the 2020 Census.

Sources

Papero Team

Papero

The Papero team builds an AI writing assistant for students, scholars and researchers, and writes about academic writing from what the product does.

Follow along on Instagram at @paperoai for campaign breakdowns.

Your next paper, with real citations.

Describe your paper and get an outline in a minute. The Free plan needs no card.

Write your paper with the sources attached.