Sampling variability is the natural sample-to-sample difference that appears when we take repeated random samples from the same population.
Quick Answer
Sampling variability is the natural sample-to-sample difference that appears when we take repeated random samples from the same population. Even good random samples do not all produce identical statistics. Expect it whenever a number comes from a sample: two honest polls of 600 students can report 58% and 62% without either being wrong. Knowing how big this chance wobble usually is tells you whether a difference between samples means something or is just noise. It shrinks as samples get bigger, in proportion to n1, so four times the sample size halves the wobble.
Why This Matters
Sampling variability is why polls come with a margin of error and why a drop from 64% to 56% in a small survey may mean nothing. Without the idea, people read every change in a sample as a real change in the population, accuse honest surveys of bias, or trust one sample as the exact truth. With it, you can ask the right question: is this difference bigger than chance alone would make?
Intuitive Explanation
Picture a giant jar of thousands of marbles, 60% of them red. Scoop out 600 and count the reds: maybe 58%. Pour them back, stir, scoop again: 61%. Again: 59.5%. The jar never changed, yet every scoop gives a slightly different answer, because each scoop catches a different handful. That scoop-to-scoop wobble is sampling variability. It is not a mistake by the person scooping; it is built into taking samples.
The wobble has a predictable size. If a computer draws 1,000 random samples of 600 from a population where 60% agree, about 95% of the sample results land between roughly 56% and 64%. Bigger samples wobble less, but not in proportion to their size: the spread shrinks like n1. Flipping a fair coin 100 times usually gives between 40% and 60% heads; flipping it 900 times, nine times as often, narrows that only to about 47% to 53%, three times narrower.
Sampling variability is not bias. Variability scatters results on both sides of the truth and shrinks with larger samples. Bias pushes results in one direction every time and does not shrink: an online poll of volunteers stays off no matter how many people answer. Nor is it measurement error, such as a miscounted marble or a misread survey answer, which comes from recording data badly rather than from which individuals happened to be chosen.
Think of it like this.A basketball player who makes 60% of her free throws shows sampling variability every practice: in sets of 20 shots she might make 10, 13 or 14, though her skill is the same. Longer sets give steadier percentages. A real player has good and bad days; a population stays fixed, so all of its wobble comes from which individuals a sample catches.
Core idea.Every random sample catches a different set of individuals, so its statistic wobbles around the population value by a predictable amount that shrinks like n1; judge any difference between samples against that wobble.
When to Use
Think about sampling variability whenever results from different samples disagree, a survey result is compared over time, or someone treats one sample's number as exact. Ask how much honest random samples of this size usually differ; if the gap is within that range, it may be chance. If one method's results are off in the same direction every time, suspect bias instead; to describe the whole pattern of possible sample results, use a sampling distribution.
Tip.Ask: Could chance alone explain why fair random samples from the same population gave different results?
How to Recognize It
Before calling two sample results different, check whether they could have come from the same population by the luck of the draw.
1.
Were both samples chosen by a fair random method from one population?
Yes means any gap between them is at least partly chance. If one method favored certain people, the gap may be Sampling Bias instead.
2.
Is the gap small compared with how much results usually bounce at this sample size?
Yes means it is ordinary variability. Putting a number on the usual bounce is the job of Standard Error.
3.
Would bigger samples be expected to make the results agree more closely?
Yes is a mark of sampling variability, which shrinks as samples grow. A built-in bias stays the same size however many people you ask.
4.
Is the question about the whole pattern of results over many samples, not just a few?
Then you need Sampling Distribution, which describes the center, spread and shape of a statistic across all possible samples.
Sampling Variability vs Sampling Bias vs Sampling Distribution vs Standard Error
All four describe how sample results differ from the truth or from each other. Ask whether you are looking at chance wobble, a built-in tilt, the full pattern of the wobble, or one number for its size.
Sampling Variability vs Sampling Bias vs Sampling Distribution vs Standard Error
Type
Meaning
Key test
Formula
Example
Sampling Variability
Sampling variability is the chance difference among results from fair random samples of the same population.
The plain fact that fair samples disagree, before any number or graph.
—
Four random samples of 25 students give 44%, 52%, 48% and 60%.
Sampling Bias
Sampling bias is a selection method that keeps favoring some members, pushing results the same way.
Error with a direction, which bigger samples do not reduce.
—
A gaming-site poll keeps over-counting gamers no matter how many people answer.
Sampling Distribution
A sampling distribution is the distribution of a statistic over every possible sample of a fixed size.
The full picture of the variability: center, spread and shape.
μp^=p
A histogram of the sample percentages from 1,000 simulated polls of 25 students.
Standard Error
The standard error is the standard deviation of a sampling distribution, the typical size of the chance error.
One number for how big the variability is.
SE=np(1−p)
Polls of 25 when the true share is 50% have SE=250.25=0.1, or 10 points.
Sampling Variability
Meaning
Sampling variability is the chance difference among results from fair random samples of the same population.
Key test
The plain fact that fair samples disagree, before any number or graph.
Example
Four random samples of 25 students give 44%, 52%, 48% and 60%.
Sampling Bias
Meaning
Sampling bias is a selection method that keeps favoring some members, pushing results the same way.
Key test
Error with a direction, which bigger samples do not reduce.
Example
A gaming-site poll keeps over-counting gamers no matter how many people answer.
Sampling Distribution
Meaning
A sampling distribution is the distribution of a statistic over every possible sample of a fixed size.
Key test
The full picture of the variability: center, spread and shape.
Formula
μp^=p
Example
A histogram of the sample percentages from 1,000 simulated polls of 25 students.
Standard Error
Meaning
The standard error is the standard deviation of a sampling distribution, the typical size of the chance error.
Key test
One number for how big the variability is.
Formula
SE=np(1−p)
Example
Polls of 25 when the true share is 50% have SE=250.25=0.1, or 10 points.
Worked Examples
Example 1 — Two polls, one truth?
Easy
Two newspapers each take a random sample of 600 students. One reports 58% support for online homework, the other 62%. A computer simulation shows that when 60% of a population agrees, about 95% of random samples of 600 give results between 56% and 64%. Must one of the polls be wrong?
Show the steps and the answer
Measure how far apart the polls are: 62%−58%=4 percentage points.
The question is whether a 4-point gap is unusual for samples this size.
Compare each result with the simulated range: 58% and 62% both lie inside 56% to 64%.
If the true support were about 60%, both results are the kind random samples of 600 usually give.
Average the polls as a rough combined estimate: 258+62=60% from 1,200 students.
Both polls are fair estimates, and together they carry more information than either alone.
Answer:No. A 4-point gap is ordinary sampling variability for samples of 600; both polls are consistent with true support near 60%.
Takeaway:Before calling a difference real, compare it with the spread that honest random samples of that size produce.
Example 2 — Nine times the data, three times the precision
Standard
Students each flip a fair coin 100 times and record the percentage of heads; about 95% of them get between 40% and 60%. If each student flips 900 times instead, about what range will 95% of them get?
Show the steps and the answer
Find the spread for 100 flips: the range runs 10 points either side of 50%.
The range is centred on the true value, 50%, and its half-width measures the wobble.
Scale by the square root of the size ratio: 100900=9=3, so the half-width becomes 310≈3.3 points.
Sampling variability shrinks like n1, not like n1.
New range: about 50−3.3=46.7% to 50+3.3=53.3%.
Nine times the flips gives only three times the precision, and the wobble never reaches zero.
Answer:About 47% to 53% heads: nine times as many flips makes the range only three times narrower.
Takeaway:To halve sampling variability you need four times the sample; to cut it to a third, nine times.
Example 3 — Find the mistake: a drop that may be noise
Application
A school paper asks 50 randomly chosen students each month whether they are satisfied with the cafeteria: 64% say yes in March and 56% in April. The editor writes: "Satisfaction fell 8 points, so the new April menu is a failure." For random samples of 50 when about 60% agree, 95% of results fall roughly between 46% and 74%. Find the mistake and correct it.
Show the steps and the answer
Measure the month-to-month change: 64%−56%=8 points.
The editor treats the whole 8 points as a real change in student opinion.
Compare with the chance range: both 64% and 56% sit well inside 46% to 74%.
With only 50 students, a single month's result can wobble about 14 points either way, so an 8-point move is ordinary.
Correct the conclusion and suggest a fix: say the data show no clear change, and survey more students, about 200 a month to halve the wobble.
Four times the sample halves the spread, so a real change of 8 points would then stand out more clearly.
Answer:The 8-point drop is within normal sampling variability for samples of 50, so it does not show that satisfaction fell or that the menu failed.
Takeaway:Small samples wobble a lot; judge a change against the range chance alone produces before explaining it.
Where You'll See It
Two polls, two answers
Two honest random polls of 600 students about online homework might report 58% and 62% support; that 4-point gap is normal sampling variability, not a mistake by either pollster.
Flipping 10 coins again and again
Each set of 10 flips rarely gives exactly 5 heads — you might get 4, then 7, then 5. These differences across samples are sampling variability in action.
Estimating a bag of marbles
Scooping 20 marbles from a big bag and counting reds gives a slightly different proportion each scoop, even though the bag never changes — showing why we report a margin of error, not a single exact number.
Common Mistakes
The slip:If two random samples disagree, at least one of them must be biased or wrong.
Why it's tempting:Different answers from honest samples feel like someone made an error.
Instead:Differences between honest random samples are expected sampling variability, not bias. Bias is a systematic, repeatable tilt; variability is random sample-to-sample fluctuation.
The slip:One good random sample gives the exact population value.
Why it's tempting:A well-done sample feels trustworthy enough to pin down the truth.
Instead:Any single sample statistic is an estimate that varies from sample to sample. We quantify that uncertainty with a margin of error or interval rather than claiming an exact value.
The slip:Bigger samples make sampling variability disappear.
Why it's tempting:More data clearly steadies estimates, so it seems variability goes to zero.
Instead:Larger samples REDUCE variability (the sampling distribution narrows) but never eliminate it. Estimates from different large samples still differ a little.
Mini Practice
Try these on your own. Tap Reveal when you want to check.
1.
Five random samples of 50 students from the same school give these percents who own a bike: 42, 36, 48, 40, 44. Find the mean and the range of the five sample percents.
Hint:Add and divide by 5; then subtract the smallest from the largest.
2.
Samples of size 25 and samples of size 400 are taken from the same population. How many times smaller is the typical sample-to-sample spread of the sample mean when n=400?
Hint:Spread shrinks with the square root of the sample size.
3.
In a population, 60% like a new logo. For random samples of 100, the sample percent typically varies by about 4.9 points. Would a sample result of 56% be surprising? What about 30%?
Hint:Compare each distance from 60 with the typical variation of 4.9 points.
4.
Ana and Ben each weigh a random sample of 30 apples from one orchard. Ana's mean is 152 g and Ben's is 147 g. Ben then finds his scale reads 3 g too low. How much of the 5 g gap is sampling variability?
Hint:Correct Ben's mean first, then compare again.
5.
Poll A randomly samples 400 voters and finds 52% support; Poll B, also random with 400 voters, finds 47%. Each poll's percent typically varies about 2.5 points. A commentator says one poll must be biased. Is the gap more likely sampling variability or bias?
Hint:Think about how far apart two fair polls could land if each can miss by about 2.5 points.
6.
One random sample of 80 of a school's 1,200 students finds 35% skip breakfast. A student writes: "Exactly 35% of all 1,200 students skip breakfast." Correct the statement.
Hint:Would a second random sample of 80 give exactly 35% again?
Want the full set?
50 practice questions for this concept — free to try, every one with a complete worked
solution showing the why, not just the answer.
How big a gap between two samples is too big to be chance?
It depends on the sample size. For percentages near 50%, random samples of 100 usually land within about 10 points of the truth, and samples of 1,000 within about 3. So a 6-point gap between two polls of 100 is unremarkable, while the same gap between two polls of 1,000 deserves a closer look. Hypothesis tests make this judgment precise.
Why do sports statistics swing so much early in a season?
Because early numbers come from small samples. A batter with 3 hits in 10 at-bats is hitting .300; if one of those outs had been a hit, the average would be .400, and one fewer hit would make it .200. Over 500 at-bats, a single hit moves the average by only .002. Early leaders often drift back toward ordinary numbers, not because they got worse, but because the early wobble was large.
Can I see sampling variability without collecting new data?
Yes, by simulation. A computer can draw thousands of random samples from a population you specify, such as one where 60% of students like a new schedule, and record each sample's percentage. Plotting those percentages shows how far fair samples typically wander. Statisticians also resample from their own data, a method called the bootstrap, to estimate the wobble when the population is unknown.
You need Random Sampling (chance chooses the sample) and Population vs Sample (statistics estimate parameters). Sampling variability adds the fact that each random sample gives a slightly different statistic, by an amount that shrinks with sample size. Next, the Sampling Distribution describes the whole pattern of those results, Standard Error measures their typical spread, and Margin of Error turns that spread into a plus-or-minus range for a poll.