Yesterday, eight independently prepared samples had mean response 9.6. Today, eight new samples targeting the same conditions had mean 10.7. The protocol and measurement were the same, but the number changed. Which experiment was wrong?
We cannot conclude that either was wrong. Yesterday and today used two samples made of different independent units. Even when conditions target the same population, different included units can produce different observations and means.
Why do sample results change when they come from the same population?
The previous unit read one sample through mean, SD, distribution, raw points, and replication structure. Now step back: that well-read sample itself can change when the experiment is repeated.
Keep four pairs of terms as a compass and compare their roles in the figures.
First define the boundary of the question
“What is our experimental mean?” is incomplete until we define which cells, donors, times, and culture conditions it concerns. The full target of the research question is the population.
A population need not be a huge visible list. It may be a finite set, such as all specified frozen samples, or the generating process for outcomes from future independent preparations under the same conditions. The essential step is to state where one question ends.
A sample is the observed values from some independent units used to learn about that population. Eight independent cultures prepared today form one sample; eight newly prepared tomorrow can form another sample targeting the same population.
A sample is part of a population but not automatically its perfect miniature. Selecting only convenient samples or overrepresenting one donor, date, or instrument may fail to reflect the question even with many rows. Sample size matters, but cannot replace where and how the sample was obtained.
The samples you can access may cover less than the population in your research question. Statistical software can calculate the entered numbers accurately, but cannot decide whether the result generalizes to a broader target.
Population numbers and sample numbers have different names
A numerical property of the full population is a parameter. We write population mean as μ and population SD as σ. A value calculated from the sample in hand is a statistic: sample mean x̄ and sample SD s.
average of a defined population
Average calculated from this sample
Spread of population values
Spread of this sample values
These symbols are a safety boundary between what we want to know and what we actually calculated. We usually cannot observe the entire population and therefore do not know μ or σ directly. We learn about them through x̄ and s.
Even if software prints Mean = 10.274, that is x̄ for the entered sample. Extra decimal places do not turn it into a direct observation of μ. A new sample can change both x̄ and s.
Why do means differ within the same population?
Build intuition with a tiny finite population of synthetic teaching values:
6, 8, 10, 12, 14
The population mean is 10. Now draw two values at a time.
| Sample | Values drawn | Sample mean x̄ | Interpretation |
|---|---|---|---|
| A | 6, 8 | 7 | A possible sample below μ |
| B | 8, 12 | 10 | A sample that happens to equal μ |
| C | 12, 14 | 13 | A possible sample above μ |
All three come from the same population, yet their means are 7, 10, and 13. B is not the sole success. The means differ because different units entered each sample.
The thought experiment of repeatedly forming samples from new independent units while keeping the same population and sampling rule is repeated sampling. The change in sample composition and statistics from sample to sample is sampling variation.
Sampling variation does not mean nothing is trustworthy. It makes the expected movement of a result an object of study. Instead of elevating one statistic into fixed truth, we examine how it can move when the same procedure is repeated.
Different sample means alone do not establish process change, instrument error, or biological difference. Sampling variation can create a difference. Investigate design, batch, day, donor, instrument records, and the magnitude of the difference before assigning a cause.
A second distribution appears
In the previous unit, one histogram point represented one experimental unit. That was a distribution of raw observations. Now repeatedly draw equal-sized samples and leave one point for each calculated x̄. Their distribution is the sampling distribution of the sample mean.
The two plots may share a numerical axis, but one point means something different:
- Raw-data distribution: one observation from one independent experimental unit.
- Sampling distribution: one statistic x̄ summarizing an entire sample.
Always finish the phrase “distribution of what?” A sampling distribution of means differs from one of medians. This unit follows sample means only.
What changes when independent sample size grows?
Compare means from samples of four with means from samples of twenty-five. A few unusually low or high values can move a small-sample mean greatly. In a larger sample, many values contribute and their chance effects tend to offset one another.
Under the same generating process and valid independent sampling, means from larger n generally cluster more narrowly around μ. Three boundaries remain:
- One large-sample x̄ need not equal μ exactly.
- Collecting more observations from a biased source does not automatically remove bias.
- Repeated readings of the same biological sample do not increase independent n.
The next unit quantifies movement of sample means with standard error and expresses it as a confidence interval. For now, read how the width of the mean-point cloud changes with n.
Be precise about what was repeated
Reading four technical wells from one culture and preparing four independent cultures can both produce four table rows. The first is technical repetition within one biological unit; the second introduces new independent units from the population.
Technical repeats are useful for examining variation in dispensing and measurement. But more readings of one sample do not provide four new pieces of biological information. To claim repeated sampling, identify which new independent experimental units entered the sample.
Draw the sample again yourself
The mini Lab uses a symmetric continuous generating process with population mean μ=10 and population SD σ=3. It is a synthetic model for observing repeated sampling, not real experimental data.
At n=8, press New sample several times. μ remains 10 while this sample's x̄ and s change. Then compare the width of the accumulated mean points at n=4 and n=25.
If this is your first time: What should I press?
- 1. Read the question firstIn the Lab title, check the one thing you will compare this time.
- 2. Change just one conditionInitially, change only one of the inputs: n, effect, or spread.
- 3. New sample pressureNew synthetic data is created. The same conditions may vary depending on the sample.
- 4. current sample and repeated-mean trail CompareWrite in one sentence what moves and what stays the same before and after the change.
If it gets stuckresetGo back to see the default results and change just one condition. This Lab is not a correct answer tester but a pattern observation tool.
Take a new sample from the same population
Fix the training generation process with a population mean of 10 and a population standard deviation of 3. Only the seed and independent n are changed and the traces of the average of this sample and the repeated averages are compared.
Raw data points for the current sample
Average of 24 samples created with the same settings
Division of this run
It is normal even if x̄ this time is different from 10. The set μ is the center of the generation process, and x̄ is the calculated value of the sample selected this time.
Same 40 repetitions: sample mean traces for n=4 and n=25
Model: Symmetric continuous normal generation process, μ=10, σ=3 · PRNG: mulberry32 · Transformation: Box–Muller · simulator_version: u04-guided-v1. This lab is for concept learning purposes only and cannot be used for actual research, quality, or clinical judgment.
The seed is a starting value that makes pseudorandom results reproducible. The same simulator version, population settings, n, and seed reproduce the same synthetic sample. This is computational reproducibility, not evidence that real cell experiments are biologically reproducible.
Each copied row is one newly drawn independent experimental unit. Different values from the article are normal. Check the relationships instead:
- This x̄ need not equal μ exactly.
- Changing seed changes the sample and its statistics.
- With equal repeat counts, mean points from larger n generally cluster more narrowly.
How does JMP express this phenomenon?
JMP does not design valid sampling for you. It summarizes the entered sample and can graph a column of statistics collected from repeated runs.
It shows the raw data shape and Mean·Std Dev·N of the current sample.
Compare how x̄ and s change when you add a new sample.
Expresses the position and shaking of x̄ values collected from repeated execution.
For the current sample, JMP Distribution reports x̄ as Mean, s as Std Dev, and entered row count as N. The researcher must verify that N equals independent experimental units. JMP will not see four technical-repeat rows from one culture and correct “independent n” to one.
Across new samples, x̄ and s change. If a ledger contains sample_id, seed, n, and sample_mean, Distribution of sample_mean expresses the locations and movement of sample means—the sampling distribution described above.
The relationships to read are:
- Mean for the current sample is statistic x̄, not parameter μ.
- Mean and Std Dev can change in a new sample.
- Each row in a sample-mean column summarizes a different whole sample.
- Software output does not guarantee population definition, unbiased sampling, or independence.
The figures and Lab teach the concept. If either program is available, import the copied current sample and sample-mean ledger. Screen layouts differ, but the distinction between a current-sample summary and a distribution of repeated statistics is the same.
Five questions before stopping at one mean
- What exactly is my target population? Define subjects, conditions, time, and independent units.
- Do I have the full population or a sample? Separate observed scope from target scope.
- Is the displayed number a parameter or a statistic? A sample calculation is x̄ or s.
- What new independent units were drawn? Do not count technical-repeat rows as new biological samples.
- How would the statistic move under another draw? Do not treat one result as fixed truth.
Sampling variation is not an annoying exception. It is a necessary structure whenever we learn about a population through a sample. Accepting it prepares the next question: how can one express a plausible range for μ around the observed x̄?
The next unit places a range of uncertainty beside one sample mean. We connect standard error, intervals that move across samples, and the correct repeated-sampling meaning of 95%.
Official supplementary resources
- JMP Statistics Knowledge Portal · Descriptive Statistics
- JMP Statistics Knowledge Portal · Inferential Statistics
- JMP Academic · Probability and Sampling Distributions
- JMP Help · Distributions of Continuous Variables
The values, figures, and In-Silico Lab in this article are synthetic material for explaining statistics. They cannot be used as evidence for real research, clinical, quality, or regulatory decisions.