Head-to-head
Standard Deviation vs Confidence Interval
Both are intervals built from the same data and the same 1.96, and they answer different questions. A standard deviation describes how far individual observations sit from the mean. A confidence interval describes how far the estimated mean might sit from the true mean. One is about the population, the other is about your knowledge of it.
Last updated
That is why the confidence interval is narrower, and why it narrows further as the sample grows. Its width is driven by the standard error — the standard deviation divided by the square root of n — so quadrupling the sample halves the interval. The standard deviation does not move at all when you collect more data; it converges to the population value instead.
The commonest error is to read a standard deviation as a range for the mean. With a mean of 100 and a standard deviation of 15, roughly 95% of individuals fall between 70.60 and 129.40. The mean of 100 observations, by contrast, is almost certainly between 97.06 and 102.94. Quoting the first as though it were the second makes an estimate look ten times less precise than it is.
There is a subtler trap underneath. The interval for individuals assumes the data are roughly normal, while the interval for the mean leans on the central limit theorem and is far more forgiving. The same 1.96 appears in both, but it is doing different work.
Standard deviation
How To Calculate Standard Deviation
Standard deviation is the typical distance of observations from their mean, measured in the original units. The squaring is what makes it algebraically useful — and the n-1 is what makes it honest for a sample.
Confidence interval
How To Calculate A Confidence Interval
A confidence interval turns a single sample estimate into a range, and the width of that range is the honest measure of how much the sample actually tells you.
How They Differ
| Attribute | Standard deviation | Confidence interval |
|---|---|---|
| What it describes | Spread of individual values | Uncertainty of the estimated mean |
| Formula | √(Σ(x − mean)² ÷ (n − 1)) | Mean ± z × (SD ÷ √n) |
| Changes with sample size | No — it converges instead | Yes, shrinking as 1 ÷ √n |
| Width at n = 100, SD 15 | 58.80 for 95% of individuals | 5.88 |
| Width at n = 400 | 58.80 | 2.94 |
| What 95% means | 95% of observations lie inside | 95% of such intervals contain the true mean |
| Assumption | Roughly normal data | Central limit theorem, so mild |
| Use it for | Describing the data | Reporting an estimate |
The Same Numbers, Both Ways
A sample mean of 100, a standard deviation of 15 and 100 observations, with both intervals built at 95%.
- Sample mean
- 100
- Standard deviation
- 15
- Sample size
- 100
- Confidence level
- 95% (z = 1.96)
| Line | Standard deviation | Confidence interval |
|---|---|---|
| Centre | 100 | 100 |
| Half-width | 29.40 | 2.94 |
| Lower bound | 70.60 | 97.06 |
| Upper bound | 129.40 | 102.94 |
| Width | 58.80 | 5.88 |
| Width with 400 observations | 58.80 | 2.94 |
| What drives the width | The spread of the data | SD ÷ √n |
With a mean of 100 and a standard deviation of 15, 95% of individuals fall between 70.60 and 129.40 — a width of 58.80. The 95% confidence interval for the mean from 100 observations is 97.06 to 102.94, a width of 5.88, because the standard error is 1.50 rather than 15.
When To Use Each
Standard deviation
- You are describing the data itself — how variable it is, or what range a typical case falls in.
- You need a benchmark for what counts as unusual in a single observation.
- You are building other statistics, since the standard deviation is an input to most of them.
Confidence interval
- You are reporting an estimate and need to say how precise it is.
- You are comparing two groups and asking whether the difference could be noise.
- You want an interval that shrinks as evidence accumulates.
Common Questions
Why is the confidence interval narrower?
Because it is about the mean, not about individuals. The mean of n observations varies less than the observations themselves, by a factor of √n — so with 100 observations the interval for the mean is one tenth the width of the interval for individuals.
Does a 95% interval mean a 95% chance the mean is inside it?
Not in the strict frequentist reading. It means that if the whole procedure were repeated, 95% of the intervals produced would contain the true mean. Any single interval either contains it or does not; the 95% is a property of the method.
When should I use 1.96 and when a t value?
1.96 is the normal approximation, which is fine once n is above roughly 30 and the population SD is known. For small samples use the t distribution, which is wider — at n = 10 the multiplier is about 2.26 rather than 1.96.
Can I compare two confidence intervals to test a difference?
Only roughly. Overlapping intervals do not prove there is no difference, because the interval for the difference depends on the correlation between the two estimates. A proper test compares the difference directly, which is narrower than either interval.
What if my data are badly skewed?
The standard deviation is then a poor summary and the individual interval is misleading. The interval for the mean still behaves reasonably in large samples thanks to the central limit theorem, or you can bootstrap, which makes no shape assumption at all.
More Comparisons
- Average vs Median
The average uses every value; the median uses the middle one. On five incomes where one is an outlier, the average moves 8.65 and the median moves 1.5 — here is why that matters.
- Percentile Rank vs Z-Score
A score of 78 against a mean of 70 and a standard deviation of 8 is a z-score of 1.00 and the 84.13th percentile — 421 of 500 test takers below it. One number is measured in standard deviations, the other in people.
- Exponential vs Logistic Growth
Starting at 100 with a 20% growth rate, exponential growth reaches 619 by period ten and 146,977 by period forty. Logistic growth towards a capacity of 1,000 reaches 451 and then 997 — same start, different world.
Sources
- NIST/SEMATECH e-Handbook — confidence intervals and standard error — National Institute of Standards and Technology
- Standard error — the relationship to sample size — Wikipedia
- Confidence interval — interpretation and common misreadings — Wikipedia
Both methods sit in Math.