Twelve independently prepared samples have mean response 10.4. Writing mean 10.4 is arithmetically correct, but it does not show how close the number is to the population mean. A new set of twelve samples could give a different mean.
This unit does not discard the mean. It places beside it a range that reflects sampling variation.
How confident can we be about a population mean from one estimate alone?
The population mean μ is the value we want to know; sample mean x̄ is a statistic computed from this sample. New samples from the same population give new x̄ values. We summarize that movement with standard error and construct a confidence interval around x̄.
Estimation learns an unknown parameter from a sample
When the whole population cannot be observed, μ cannot be calculated directly. We calculate x̄ from a sample to estimate μ. Giving one value for one parameter is a point estimate.
Point estimates are useful: they compare centers across experiments, start later calculations, and make tables and figures compact. But x̄=10.4 alone cannot distinguish:
- a relatively stable 10.4 from many independent, tightly clustered samples;
- a potentially variable 10.4 from few independent, widely spread samples.
The means match, but their precision differs. A point estimate therefore needs information about how much that statistic can move from sample to sample.
Interval estimation is not choosing a safe range from the observed minimum and maximum. Those extrema describe individual observations; a mean confidence interval is an inferential procedure targeting the population mean μ. The targets differ.
A 95% confidence interval for the mean is not an interval containing 95% of individual sample values. It is also not a prediction interval for a new individual observation. This unit concerns estimation of the population mean μ only.
Standard deviation and standard error describe different movement
Sample SD s summarizes how individual observations in the current sample spread around x̄. Standard error SE describes how x̄ moves if new samples are repeatedly drawn in the same way.
The basic estimated relation for a mean is:
SE = s / √n
Here n is the number of independent experimental units. Reading the same culture four times does not make n=4. Putting technical repeats that are not independent into n enlarges only the denominator, makes SE unjustifiably small, and narrows the interval without basis.
| Question | Sample standard deviation s | Standard error of the mean SE |
|---|---|---|
| Spread of what? | Individual observations | Sample means under repeated sampling |
| As n grows | Does not reduce population spread itself | Usually decreases for the same spread |
| Role in a CI | Material used to estimate SE | Direct material for interval width |
For s=3, SE is 1 at n=9 and 0.5 at n=36. Independent n must quadruple to halve SE: a square-root relationship, not a doubling of precision when n merely doubles.
A confidence interval expresses sampling variation; it does not automatically include biased sampling, wrong independence units, batch mixing, or measurement bias. A systematically wrong sample can produce a very narrow and wrong interval.
A mean confidence interval has four calculation layers
In the usual small-sample situation where population SD σ is unknown, use sample s and a Student t critical value.
The two-sided mean CI is:
x̄ ± t* × s / √n
t* depends on chosen confidence level and n−1 degrees of freedom. With the same sample, a 99% interval has a larger t* than a 95% interval and is wider. Demanding a higher long-run coverage means accepting a wider single interval.
For n=20, x̄=10.4, s=3.0, the two-sided 95% t critical value at df=19 is about 2.093.
SE = 3.0 / √20 ≈ 0.671margin of error = 2.093 × 0.671 ≈ 1.40595% CI = 10.4 ± 1.405 ≈ 8.995–11.805
The point is not memorizing decimals. Read the structure: x̄ supplies the center; SE and t supply the width*.
To narrow an interval, state what changes:
- More independent n tends to lower SE and narrow the interval at the same s.
- Larger s raises SE and widens it.
- Raising confidence from 90% to 95% or 99% raises t* and widens it.
- Treating more technical repeats from the same sample as independent n is not a justified way to narrow it.
95% is not a probability inside a completed interval
Before sampling, we do not know which x̄ or interval will occur. Repeatedly sample the same population at the same size and apply the same 95% procedure: intervals occupy different locations. The procedure is designed so that about 95% of those intervals include the fixed μ in the long run.
Once this sample has produced lower and upper limits, μ is either in that interval or it is not. In frequentist confidence intervals, do not say “the interval calculated this time has a 95% probability of containing μ.” The 95% is the long-run coverage of a procedure that creates moving intervals.
A safer statement is:
Under the same sampling conditions and model, repeating this calculation procedure would produce intervals that include population mean μ about 95% of the time.
In ordinary reporting, “the 95% confidence interval for μ is 9.00–11.81” is concise and acceptable. Keep the long-run procedure in mind.
Coverage of a 95% confidence procedure is a property under its sampling, independence, and model conditions. The number 95% does not repair a biased sample, missing batch structure, selective reporting, or an inappropriate model.
Move intervals yourself
The mini Lab below draws independent samples from a normal generating process with mean μ=10 and SD σ=3. It is an educational synthetic model for seeing how confidence intervals work, not real study data.
Start with n=20 and 95%, then press New sample batch. μ remains 10, but x̄, s, SE, and the interval all change. Some batches miss μ; that does not mean the button or formula failed.
Then compare:
- n=5 and n=40 at the same seed and confidence level;
- 90%, 95%, and 99% at the same seed and n;
- why one set of 20 observations need not show exactly the setting's proportion.
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 batch pressureNew synthetic data is created. The same conditions may vary depending on the sample.
- 4. confidence-interval width and whether it contains μ 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.
Directly compare section width and repetition inclusion
We use a training generation process with population mean μ=10 and standard deviation σ=3. Change the independence n and confidence level, and observe how the intervals behave in the new sample.
Confidence interval for the mean of this sample
Four Layers of Computation
20 observations is a small demonstration of long-term coverage, but is not a sufficient number of repetitions to accurately verify the confidence level you set.
Mean confidence interval: two-tailed Student t procedure · Population model: normal, μ=10, σ=3 · PRNG/transformation same as U04 · simulator_version:u05-ci-guided-v1. This is a composite material for educational purposes only and cannot be used for research, clinical, quality, or regulatory decisions.
The Lab reveals true μ because we defined the synthetic generation process. In real experiments μ is unknown and estimated from the sample. Do not confuse an answer check possible only in simulation with actual analysis.
Read three JMP lines as one relationship
For a continuous variable, JMP Distribution output can show Mean, Std Dev, Std Error Mean, and mean confidence limits. Connect the lines through one calculation rather than memorizing them separately.
It is a point estimate of the population mean given by the current sample.
It is the estimated standard error of the sample mean and has a different role from Std Dev.
This is the 95% confidence limit for the mean calculated using the same sample and model.
Mean is x̄ and Std Dev is s. Std Error Mean corresponds to s/√n; Lower 95% Mean and Upper 95% Mean are the endpoints of the t interval around x̄. An Interval Plot can help compare means and intervals across groups, but overlap alone must not replace every hypothesis-test conclusion.
Even if JMP calculates correctly, the researcher must still ask:
- Which population does Mean target?
- Is every N row truly an independent experimental unit?
- Are technical repeats, donors, days, and batches represented in the analysis?
- Is the chosen t procedure appropriate for the data and question?
- Is the interval narrow enough to distinguish an experimentally important difference?
The text and Lab are sufficient for learning point estimation, standard error, and coverage. In any statistics program, identify how Mean, SE, and confidence limits relate even if the column names differ.
What to retain in a result statement
Do not write only mean 10.4. Preserve what mean, how many independent units, and which interval.
The mean response of 20 independent cultures was 10.4, and the two-sided 95% Student t confidence interval for the population mean was 9.00–11.81.
Add units, sample-selection criteria, biological/technical replicate structure, missing-data handling, and analysis model as appropriate. A confidence interval is not two numbers; it is a result jointly created by a question, sample, model, and calculation.
Six statements to check before finishing
- x̄ is a point estimate of μ, not direct observation of μ.
- s describes spread of individual observations; SE describes sampling variation of x̄.
- A mean CI has structure
x̄ ± t* × SE. - Larger independent n tends to narrow the interval at the same spread.
- Higher confidence requires a wider interval.
- 95% is long-run coverage of a repeated procedure, not probability of one fixed interval.
The next unit distinguishes a mean confidence interval from the range for a new observation. Confidence, prediction, and tolerance intervals answer different questions and have different widths.
Official supplementary resources
- NIST/SEMATECH · Confidence intervals
- NIST/SEMATECH · Confidence limits for the mean
- JMP Statistics Knowledge Portal · Confidence Intervals
- JMP Help · Confidence Interval for One Sample Mean
- JMP Help · Distributions of Continuous Variables
The numbers, 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.