Skip to content

New to Omni?·Get 20% off your first purchase with code WELCOME20 at checkout.See plans

CFA® confidence intervals: what makes the range wider?

A confidence interval reflects uncertainty in a sample mean. Learn how standard error, sample size, critical values and assumptions change the range you report.

ConceptOmni Curriculum Team6 min read

Blank index cards, a pencil and a ruler on a navy study desk.
On this page

A confidence interval puts a range around a sample mean to show how much the estimate may move from sample to sample. Its width comes from the standard error and a critical value chosen for the confidence level and distribution. On CFA® Level I, the useful sequence is to identify what is known about population variability, calculate the standard error, and only then form and interpret the interval.

What the standard error measures

Imagine repeatedly drawing equally sized samples from the same population and calculating each sample's mean. The means would differ even if the population never changed. The spread of those calculated means is called their standard error. It describes uncertainty in an estimate of a mean; it is not the spread of individual observations. CFA Institute's Estimation and Inference reading identifies that distinction as a Level I learning outcome.

The population standard deviation describes how far individual values tend to sit from the population mean. Dividing it by the square root of the sample size converts that individual variation into variation of the mean across repeated samples. If the population standard deviation is unavailable, the sample standard deviation supplies an estimate. Keep the units straight: if observations are measured in percentage points of return, the standard error is also in percentage points. Treating it as a percentage of the mean would change the question.

σxˉ=σnSE^(xˉ)=sn\sigma_{\bar{x}}=\frac{\sigma}{\sqrt{n}}\qquad \widehat{SE}(\bar{x})=\frac{s}{\sqrt{n}}
Standard error of a sample mean using population standard deviation sigma, or an estimate using sample standard deviation s, for sample size n.

The square root matters. Making a sample four times as large halves the standard error if the underlying variability and sampling assumptions stay the same. It does not quarter it. This gives a quick reasonableness check before using a calculator: a larger valid sample should narrow sampling uncertainty, but a small addition to an already large sample may change the interval only modestly. More observations also cannot cure a systematically unrepresentative sample. A precise calculation around a biased sample mean remains a poor description of the target population.

From standard error to interval

An interval starts with the sample mean and extends a margin on either side. The margin is a distribution-specific critical value multiplied by the standard error. The critical value reflects the chosen confidence level and, when using a t distribution, degrees of freedom. CFA Institute's sampling and estimation guide gives the normal and t forms. The form is more important than memorising one isolated number because a question may change the confidence level or the information available about variability.

interval=xˉ±c×SE(xˉ)\text{interval}=\bar{x}\pm c\times SE(\bar{x})
Confidence interval centred on sample mean x bar; c is the appropriate critical value and SE is the standard error of that mean.

If the population standard deviation is known and the sampling distribution is appropriate, use the normal critical value with sigma divided by the square root of n. When the population standard deviation must be estimated from a normal population, use the t critical value with s divided by the square root of n and n minus one degrees of freedom. The t distribution has heavier tails for a finite sample, which reflects the additional uncertainty in estimating variability. With more degrees of freedom it approaches the normal distribution. CFA Institute's hypothesis-testing reading also identifies t as the theoretically correct choice for a normal population mean with unknown variance.

Do not treat a sample-size rule of thumb as permission to ignore the data-generating process. An approximately normal sampling distribution can arise for a sufficiently large, well-behaved sample through the central limit theorem, but severe dependence, selection bias or extreme tails call for scrutiny. The formulas describe sampling variation under their assumptions. They do not repair a poor definition of the population or observations gathered only when a convenient result appeared.

A worked interval with illustrative figures

Suppose, purely for illustration, that an analyst observes 36 independent monthly measurements with a sample mean of 10 units and sample standard deviation of 6 units. The population standard deviation is unknown, and assume a normal population for this example. The estimated standard error is 6 divided by the square root of 36, or 1 unit. The t distribution has 35 degrees of freedom. For a two-sided 95% interval, a t critical value of about 2.03 gives a margin of about 2.03 units. The resulting interval runs from 7.97 to 12.03 units. These are invented measurements and not market data.

The arithmetic has three checkpoints. The denominator is 6, not 36; the estimated standard error is 1, not 6; and the critical value is multiplied by that standard error, not by the raw standard deviation. If a candidate used the normal value 1.96 here, the interval would be a little narrower. That numerical difference may look small, but it reveals the wrong distribution choice for the stated normal population with unknown variance. If the question instead supplied a known population standard deviation, the normal route would be justified on the given conditions.

The interval is a statement about the procedure over repeated samples. If the sampling process and assumptions hold, intervals built this way would contain the fixed population mean at the stated long-run frequency. Once a particular sample has produced a particular interval, the unknown population mean is not being randomly redrawn inside it. This wording is more careful than saying the interval gives a probability distribution over possible values of an unchanged population parameter. It also avoids the opposite error of treating the interval as a guarantee.

What changes the answer

First, a higher requested confidence level generally requires a more extreme critical value and a wider interval for the same sample. Second, greater observed dispersion increases the standard error and widens the interval. Third, a larger sample reduces the standard error when the sampling design remains valid. Fourth, changing from known to estimated population variability can change the critical distribution. A candidate can often eliminate an implausible answer without fully calculating it by checking these directions of change.

Distinguish an interval for a population mean from a range for a future individual observation. A single observation has its own variation around the mean, while the mean of a sample averages some of that variation away. Reusing the standard error of a mean as if it described individual outcomes would produce an unjustifiably narrow range. Likewise, a confidence interval is not the same as a prediction interval in a regression setting; the latter addresses a different target and source of uncertainty.

The point estimate deserves attention too. A sample mean can be unbiased under a suitable sampling design yet still be imprecise in a small sample. Precision concerns how tightly the sampling distribution clusters; unbiasedness concerns its centre. Neither property follows from a narrow displayed interval if observations were chosen in a way that excludes part of the intended population. On the exam, read the population definition, sampling method and variance condition before reaching for a critical value.

A compact exam-day check

  • Name the parameter and the estimator. A population mean is unknown; the observed sample mean is the centre of the interval.
  • Identify whether population variability is supplied or estimated. Match the normal or t critical value to the stated assumptions.
  • Calculate standard error before margin of error. Keep its units, and use the square root of sample size.
  • Check that endpoints straddle the sample mean symmetrically and that greater confidence or variability would widen them.
  • Interpret repeated-sample coverage and state the sampling assumptions. Avoid a claim that the computed interval is certain to contain the mean.

One further check is whether the question asks about a mean at all. A sample proportion has a different standard-error calculation, while two sample means require a standard error for their difference and an appropriate account of dependence or independence. Applying the single-mean formula just because a question mentions a sample can produce a plausible but irrelevant interval. Write the estimand in words before substituting numbers. For the single-mean case, also check whether observations are independent and whether the population description justifies the selected distribution. These steps make the interval answer the stated question, rather than a nearby one that happens to be easier to calculate.

For a broader view of where this sits, the CFA programme page locates the qualification, while Omni's methodology page explains the teaching approach. Use the official CFA Institute readings for the curriculum wording and the simple checks above to keep the calculation tied to the question's assumptions.

Sources

  1. Estimation and Inference | CFA Institute · cfainstitute.org · Retrieved
  2. Hypothesis Testing | CFA Institute · cfainstitute.org · Retrieved
  3. Members’ Guide to 2022 Refresher Readings | CFA Institute · cfainstitute.org · Retrieved

Explore CFA study resources

Educational content, not investment advice. This article is written to help candidates prepare for professional exams. It is not a recommendation to buy, sell or hold any security, and it does not take your personal circumstances into account.

Omni Finance Academy is an independent exam-preparation provider and is not affiliated with or endorsed by CFA Institute, GARP, CAIA Association or CFP Board. CFA®, FRM®, CAIA® and CFP® are trademarks of their respective owners. Full disclaimer.