Free scoping read. A written assessment of your manuscript in two business days, at no cost.

Start one
Contents

Academy · Research Data, Analytics & AI · Confidence Level vs Width

The Effect of Confidence Level on the Width of the Confidence Interval

Raising the confidence level from 90% to 95% to 99% widens the interval in every case, because a higher level requires a larger critical value from the same distribution. This trade-off between confidence and precision is determined by the statistics rather than by preference.

Pubrica AcademyWritten by a biostatistics analyst Reviewed bya research data methodologist Published10 October 2026 Last checked10 October 2026 Reading time7 minutes Sources10 peer-reviewed

The confidence level attached to an interval, most often 95%, but sometimes 90% or 99%, sets the critical value used to multiply the standard error when computing the interval's margin of error. A higher stated confidence level requires a larger critical value from the relevant distribution, which mechanically produces a wider interval from the identical underlying data; a lower confidence level requires a smaller critical value and produces a narrower one1. This relationship is a fixed mathematical trade-off rather than a matter of judgment: for any given dataset, a researcher can choose greater confidence or greater precision, but cannot increase both simultaneously from the same sample.

Learning objectives

  • How the critical value changes as the stated confidence level changes
  • Why a 99% confidence interval is always wider than a 95% interval from the same data
  • Why 95% became the conventional default level rather than a universally required one

Common misconceptions

  • “A higher confidence level is simply a stronger, more impressive result with no cost.” A higher confidence level is mathematically guaranteed to widen the interval from the same data, because it requires a larger critical value from the sampling distribution; a 99% interval sacrifices precision specifically to gain confidence, and a 90% interval does the reverse, so neither direction constitutes an improvement without cost.

Variation of the critical value with confidence level

A confidence interval's margin of error is calculated as a critical value, taken from the standard normal (or t) distribution, multiplied by the standard error of the estimate1. The critical value itself is determined entirely by the chosen confidence level: a 95% confidence level corresponds to a critical value of approximately 1.96 standard errors on each side of the point estimate under a normal approximation, while a 99% confidence level corresponds to a larger critical value of approximately 2.576, and a 90% confidence level to a smaller value of approximately 1.6451. Because the margin of error scales directly with this critical value while the standard error stays fixed for a given dataset, raising the confidence level always raises the critical value and therefore always widens the resulting interval.

Comparison of 90%, 95% and 99% intervals from the same data

Holding the point estimate and standard error constant, a 99% confidence interval will be noticeably wider than a 95% interval from the identical data, roughly 31% wider given the ratio of their respective critical values (2.576 versus 1.96), and a 90% interval will be narrower than the 95% interval by a comparable proportion in the other direction (1.645 versus 1.96)1. This is the mechanical explanation for why a result that appears clearly significant under a 95% interval, excluding the null value, can become inconclusive under a 99% interval calculated from the same data, simply because the higher confidence level requires a wider interval that now extends across the null value. ICH’s E9 guideline on statistical principles for clinical trials relies on precisely this trade-off when specifying the confidence level a trial’s primary analysis should use, since the chosen level directly determines how wide the reported precision estimate for the primary endpoint will be2.

Origins of the 95% convention

The 95% confidence level, and its complementary 0.05 significance threshold for hypothesis tests, traces back to conventions popularized by Ronald Fisher in the early twentieth century, and it has persisted largely as an institutionalized default rather than because 95% occupies any uniquely correct position on the confidence-versus-precision trade-off3. The American Statistical Association's 2016 statement on statistical significance explicitly cautions against treating any single threshold, including the 0.05/95% convention, as a bright line separating "real" findings from "non-findings," noting that the appropriate level of confidence to report should be justified by the context and stakes of the specific question being asked rather than applied automatically by habit4. Fields and journals that require a 99% or higher confidence level for certain claims, common in some areas of physics and genomics, are making an explicit choice to trade precision for a stronger guard against false positives, not following a universal statistical rule.

Interaction of confidence level and sample size

Confidence level and sample size affect interval width through two independent routes that are readily conflated: the confidence level sets the critical value multiplying the standard error, while the sample size sets the standard error itself1. A researcher facing an uncomfortably wide interval therefore has two genuinely different options: lowering the confidence level, which narrows the interval without changing the underlying data at all, or increasing the sample size, which narrows the interval by collecting more information. Methodological guidance consistently holds that only the second option represents a real gain in evidentiary strength; lowering the confidence level to produce a narrower interval is a presentational choice that leaves the underlying uncertainty in the data unchanged, which is why Cochrane and ICH guidance both expect the confidence level to be fixed by protocol before results are seen, rather than adjusted retrospectively12.

Selecting a confidence level proportionate to the stakes of the question

Because the confidence level is a deliberate trade-off rather than a fixed property of the data, methodological guidance increasingly frames the choice as one that should reflect the consequences of being wrong in each direction4. A screening context in which a missed true effect carries a high cost may justify a lower confidence level (and narrower interval) to avoid excessive conservatism, while a context in which a false positive could lead to the adoption of an ineffective or harmful intervention may justify a higher confidence level despite the wider, less decisive interval it produces. Reporting guidance for clinical trials under ICH E9 requires this choice to be specified and justified in the statistical analysis plan before unblinding, precisely so that the confidence level cannot be selected, consciously or otherwise, to make a particular result appear more or less conclusive once the results are known2.

Confidence level, multiple comparisons and family-wise error

When a single study or meta-analysis reports confidence intervals for several outcomes or several subgroups at once, each individual interval still carries its own stated confidence level, but the probability that at least one of them, by chance, fails to contain its true value rises with the number of comparisons made, a problem usually discussed in terms of P values as the multiple-comparisons or family-wise error rate problem3. Some statistical approaches address this by raising the effective confidence level used for each individual comparison, which widens each interval accordingly, trading some precision on every comparison for a lower chance of any one of many comparisons being misleading. Cochrane guidance recommends treating multiple subgroup or outcome analyses in a single review with particular caution for this reason, since reporting many separate 95% intervals without adjustment, and then highlighting whichever one excludes the null value, can produce a highly misleading overall impression of the evidence even though each individual interval was calculated correctly1.

A checklist for comparing confidence intervals across studies

When comparing confidence intervals reported across different papers, a reader should check four points in sequence: the stated confidence level for each interval, whether that level is the same across the papers being compared, whether any correction for multiple comparisons was applied and, if so, how, and finally, only after confirming all three of the above, whether the reported widths are comparable at face value4. Proceeding directly to a comparison of numeric widths without checking these three preceding points is one of the more common, avoidable errors in informal comparisons of evidence across studies, and it can lead to the mistaken conclusion that one study’s result is meaningfully more or less precise than another’s when the apparent difference merely reflects different reporting conventions.

A worked example illustrates how omitting these checks can mislead. If Study A reports a mean difference of 4 units with a 90% confidence interval of 1 to 7, and Study B, testing a similar intervention, reports a mean difference of 4 units with a 95% confidence interval of 0.5 to 7.5, a reader comparing only the numeric widths might conclude that Study B’s result is less precise because its interval spans a wider range, when in fact Study B used a higher confidence level, which mechanically widens the interval through a larger critical value, and may have an equal or even smaller standard error than Study A once both are recalculated to the same confidence level1. Recalculating Study A's interval to a 95% level using its implied standard error, or Study B's to 90%, before comparing the two directly is the only way to isolate whether the apparent difference in width reflects genuinely different precision in the underlying data, as opposed to simply a different, equally defensible choice of confidence level made at the protocol stage2. The same recalculation should be performed before two studies are cited together in a narrative review or discussion section as showing “similar precision,” since an unstated difference in confidence level is a common, entirely avoidable reason why two results can appear less consistent with each other than they are.

Key points

Key points

Raising the confidence level always widens a confidence interval from the same data, because it requires a larger critical value, while sample size narrows an interval through a separate route entirely; the 95% convention is a widely used default, not a mathematically mandated one, so the appropriate level, and any adjustment for multiple comparisons, should be chosen deliberately, and pre-specified, for the question at hand.

Practical application

Practical application

  1. The confidence level should be stated explicitly whenever an interval is reported. The 95% level should not be left unstated as an assumed default; readers comparing intervals across studies need to know the level used in order to interpret differences in width correctly.
  2. Intervals constructed at different confidence levels should be recalculated, not assumed comparable, before comparison. A 90% interval and a 99% interval from different papers cannot be compared directly for precision without adjustment for the different critical values underlying each.
  3. A non-default confidence level should be chosen deliberately and with a stated reason. If a field or specific claim requires a stricter threshold against false positives, that choice and its cost in precision should be documented rather than changing levels without explanation.

Frequently asked questions

1.96 is the approximate critical value from the standard normal distribution that leaves 2.5% of the distribution in each of the two tails, totaling 5%, which corresponds to a 95% confidence level; different confidence levels require correspondingly different critical values from the same distribution.

No; it reflects only the acceptance of a different trade-off. The underlying standard error, and therefore the actual precision of the data, has not changed; a lower confidence level simply accepts a lower probability of the interval containing the true value in exchange for reporting a narrower range.

References

  1. Deeks JJ, Higgins JPT, Altman DG (editors). Chapter 10: Analysing data and undertaking meta-analyses. In: Cochrane Handbook for Systematic Reviews of Interventions. Cochrane, 2024. read 10 October 2026. View on Cochrane Training
  2. International Council for Harmonisation. E9: Statistical Principles for Clinical Trials. read 10 October 2026. View on ICH
  3. Greenland S, Senn SJ, Rothman KJ, Carlin JB, Poole C, Goodman SN, Altman DG. Statistical tests, P values, confidence intervals, and power: a guide to misinterpretations. European Journal of Epidemiology. 2016;31(4):337-350. read 10 October 2026. View on European Journal of Epidemiology
  4. Wasserstein RL, Lazar NA. The ASA Statement on p-Values: Context, Process, and Purpose. The American Statistician. 2016;70(2):129-133. read 10 October 2026. View on Taylor & Francis Online
  5. Gardner MJ, Altman DG. Confidence intervals rather than P values: estimation rather than hypothesis testing. BMJ. 1986;292(6522):746-750. read 10 October 2026. View on The BMJ
  6. Altman DG, Bland JM. How to obtain the confidence interval from a P value. BMJ. 2011;343:d2090. read 10 October 2026. View on The BMJ
  7. Schulz KF, Grimes DA. Sample size calculations in randomised trials: mandatory and mystical. The Lancet. 2005;365(9467):1348-1353. read 10 October 2026. View on The Lancet
  8. Higgins JPT, Thompson SG. Quantifying heterogeneity in a meta-analysis. Statistics in Medicine. 2002;21(11):1539-1558. read 10 October 2026. View on PubMed
  9. Riley RD, Higgins JPT, Deeks JJ. Interpretation of random effects meta-analyses. BMJ. 2011;342:d549. read 10 October 2026. View on The BMJ
  10. DerSimonian R, Laird N. Meta-analysis in clinical trials. Controlled Clinical Trials. 1986;7(3):177-188. read 10 October 2026. View on ScienceDirect

Every source above is a peer-reviewed article, a publisher’s record of one, or a standard issued by the body that owns it. Each carries the date it was last read. Where a standard moves, this article is revised rather than left up with a note.

How to cite this article

Pick your referencing style and copy it. Your university or target journal may vary the punctuation — check its guide if it has one.

Pubrica Academy. (2026, October 10). The Effect of Confidence Level on the Width of the Confidence Interval. Pubrica. https://pubrica.com/academy/confidence-level-impact-the-width-of-the-confidence-interval/

This article is written to be learned from. Cite it if you are referring to what it says; if you are using the evidence behind it, cite the sources listed above instead.

Confidence Level Reporting Checklist

This checklist should be used before reporting or comparing confidence intervals constructed at different confidence levels.

[ ] Stated the confidence level explicitly next to every reported interval
[ ] Checked whether a comparison across studies used the same confidence level
[ ] Recalculated, rather than assumed, interval widths when comparing different confidence levels
[ ] Documented a deliberate reason if using a non-default (not 95%) confidence level
[ ] Avoided treating the 95%/0.05 convention as a mathematically mandated threshold

Pubrica AcademyWritten and maintained by Pubrica’s subject matter experts. Articles are revised when the underlying standards change, and each article states the date on which it was last reviewed.

How Pubrica can help

Pubrica’s biostatistics team recalculates confidence intervals to a consistent confidence level across a manuscript or comparison, and documents the rationale where a non-default level is clinically or scientifically justified.

Find something else

Search the Pubrica Academy

Access is free and requires no account. Enter a topic, or open the full library to filter by category, sub-topic or resource type.

Consistency of confidence levels across tables and figures

A scoping assessment is free and takes two business days. It provides a written assessment of what the work requires and which specialist would undertake it.

Get a free quote

Regulations

Check your project against this rule

  1. 1Your work
  2. 2Your details
  3. 3Send
What do you need?

Material you send is held under signed confidentiality, with access limited to the assigned team and encrypted transfer. It is not used to train anything and is not passed to third-party tools. Ask for the agreement before you send anything you would rather cover first.