A confidence interval shrinks as sample size grows, so the interval alone means little without the count

What it is

A confidence interval's width is not an intrinsic property of the data but a function of sample size: with the same mean and variance, growing the sample from 30 to 3,000 narrows the 95% confidence interval roughly tenfold (width scales with 1/√n). So a "narrow interval" on a chart may reflect genuinely low variance—or simply a large sample; a "wide interval" may reflect high data variability—or a small sample. Without sample-size information, the reader cannot attribute interval width to the data's properties versus the collection effort—and the two attributions lead to entirely different actions (improve the product vs collect more data).

Why it happens

The width-sample-size relationship comes from the mathematics of estimation precision: SE = SD/√n, and the confidence interval ≈ SE × constant. Sample size enters through a square root, which means diminishing returns: doubling the sample narrows the interval to only 1/√2 (about 71%) of its width, and growing from 100 to 10,000 (a hundredfold) shrinks it to one-tenth. This square-root structure produces two phenomena on charts: first, at small samples, interval width is extremely sensitive to n (30 → 100 nearly halves the interval), so the wide interval the reader sees might narrow significantly with just a bit more data collection; second, at large samples, the marginal narrowing approaches zero, so two intervals of nearly identical visual width may correspond to a tenfold sample-size difference. Readers need the sample size to judge "is this interval a property of the data or of the collection," and thus whether the next step is improving the product or accumulating data.

Where it stops holding

Sample size is not the only determinant of interval width: variance matters too (high-variance small samples produce wide intervals), so a narrow interval does not directly imply a large sample. Width sensitivity to sample size also depends on the confidence level: a 99% interval is naturally about 30% wider than a 95% one, so across charts with different confidence levels, width differences may come from the level rather than n. For non-independent samples (time series, clustered data), the effective sample size is smaller than the nominal n, and intervals computed from nominal n are systematically too narrow—labeling both nominal and effective n is more honest. For proportions, interval width also depends on the base rate (rates near 0% or 100% naturally have narrower intervals), and sample-size interpretation needs to be read alongside the base rate.

Applying it

• Show the sample size n alongside every confidence interval, in labels, tooltips, or captions.
• When group sample sizes in a comparison chart differ by more than 2x, explicitly display each group's n with a bar or number, preventing readers from attributing width differences entirely to data variability.
• On charts with extremely unbalanced group sizes (e.g., one group n=5, others n=1,000), add an "insufficient sample" tag to the tiny group.
• Verification: ask a reader whether a given group's narrow interval means "stable data" or "large sample"; if they cannot tell because n is absent from the chart, sample-size labeling is missing.

Related

• Same group: U8.02.1 Whether an error bar shows standard deviation, standard error, or confidence interval must be stated · U8.02.2 Overlapping intervals between two groups do not imply a non-significant difference · U8.02.4 Gradient bands create less certainty illusion than hard-edged intervals · U8.02.5 A band's visual width depends on the y-axis scale and cannot be compared across charts
• Nearby: U8.01.1 Point estimates hide interval information · U10.04.4 Check whether uncertainty, missing proportion, and sample size are labeled
• Search terms: sample size annotation · interval width · standard error scaling