Skip to content

Actuarial Statistics · Confidence intervals and prediction intervals

Prediction Intervals: Formula and Method for a Future Observation

Updated 11 October 2026 · Fact-checked

A prediction interval gives a range that a future single observation will fall in with stated probability. It is wider than a confidence interval because it adds the variance of the new observation to the variance of the estimate. For a normal sample: x̄ ± t × s × √(1 + 1/n).

Understand Prediction Intervals

A confidence interval is about a fixed unknown parameter, such as the mean μ. It tells you where μ is likely to lie. A prediction interval is about a random quantity that has not happened yet, such as the next claim or the next observation Y.

The key idea is that two sources of uncertainty act together. First, you do not know μ exactly, so your estimate has error. Second, the new observation will itself vary around μ. A confidence interval allows only for the first. A prediction interval allows for both.

Take a normal sample X₁, …, Xₙ from N(μ, σ²) and a new independent observation Y from the same distribution. Then Y − X̄ has mean 0 and variance σ² + σ²/n = σ²(1 + 1/n). The new observation is independent of the sample, so the variances add. This is the extra variance term: the "1" is the variance of Y, and the "1/n" is the variance of X̄.

If σ is known, you use a standard normal pivot. If σ is unknown, you replace it with the sample standard deviation s and use the t distribution with n − 1 degrees of freedom. Then (Y − X̄) ÷ (s√(1 + 1/n)) follows a t distribution with n − 1 degrees of freedom.

The same idea applies in regression. The prediction at a new x₀ has an estimation error from the fitted line plus the error term of the new observation. So the regression prediction interval is always wider than the confidence interval for the mean response at x₀. As n grows, a confidence interval shrinks towards zero width, but a prediction interval does not. It stays at least about as wide as the natural spread of the observation.

Key rules to remember

Variance of prediction error (normal sample)
Var(Y − X̄) = σ² (1 + 1/n)
Y is a new observation independent of the sample. The 1 is for Y and the 1/n is for X̄.
Prediction interval, σ known
x̄ ± z × σ × √(1 + 1/n)
z is the standard normal point, e.g. 1.96 for 95% two-sided.
Prediction interval, σ unknown
x̄ ± t(n−1) × s × √(1 + 1/n)
Use the t point with n − 1 degrees of freedom and s² = Σ(xᵢ − x̄)² ÷ (n − 1). Assumes a normal population.
Regression: prediction interval at x₀
ŷ₀ ± t(n−2) × σ̂ × √(1 + 1/n + (x₀ − x̄)² ÷ Sxx)
Simple linear regression, σ̂² = residual sum of squares ÷ (n − 2), Sxx = Σ(xᵢ − x̄)².
Regression: confidence interval for mean response at x₀
ŷ₀ ± t(n−2) × σ̂ × √(1/n + (x₀ − x̄)² ÷ Sxx)
Same as above without the 1. Use it to compare with the prediction interval.

How to solve Prediction Intervals questions

Use this method for any prediction interval question, whether a simple sample or a regression.

  1. 1Identify what is being predicted: a single future observation, not a parameter. If it is a parameter or a mean response, you need a confidence interval.
  2. 2State the assumptions: observations are independent and normally distributed with common variance, and the new observation is independent of the sample.
  3. 3Write the point prediction: x̄ for a sample, or ŷ₀ = α̂ + β̂x₀ for regression.
  4. 4Write the variance of the prediction error: σ²(1 + 1/n), or σ²(1 + 1/n + (x₀ − x̄)²/Sxx) for regression.
  5. 5Choose the distribution: normal if σ is known, t if σ is estimated. Degrees of freedom are n − 1 for a sample and n − 2 for simple regression.
  6. 6Find the critical value for the required level, using the two-sided tail for an interval.
  7. 7Compute point prediction ± critical value × estimated standard deviation of the error.
  8. 8State the result in words: the probability that the future observation lies in the interval is the stated level.

Quickest way: Confidence interval width, then widen

When to use it: Use when you already have the standard error of the mean, or when the question asks you to compare the two intervals.

  1. Compute the standard error for the mean: s/√n.
  2. Multiply it by √(n + 1) to get the prediction standard deviation s√(1 + 1/n), since √(1 + 1/n) = √(n + 1) ÷ √n.
  3. Multiply by the critical value and add and subtract from the point estimate.
  4. Check the answer: the prediction interval must be wider than the confidence interval at the same level.

Common mistakes in Prediction Intervals

  • Using s√(1/n) instead of s√(1 + 1/n).

    Students copy the confidence interval formula for the mean and forget the new observation has its own variance.

    Fix: Ask what is being predicted. If it is a single new value, add the 1 inside the square root.

  • Using the normal point when σ is estimated.

    The 1.96 value is memorised and used by habit.

    Fix: If s replaces σ, use the t distribution with n − 1 degrees of freedom (n − 2 in simple regression).

  • Using the wrong degrees of freedom in regression.

    Students carry n − 1 over from the one-sample case.

    Fix: Simple linear regression estimates two parameters, so use n − 2.

  • Thinking a large sample makes the prediction interval very narrow.

    Confidence intervals do shrink as n grows, so the same is assumed here.

    Fix: The 1 term does not shrink. The interval tends to μ ± zσ as n grows, not to a single point.

  • Forgetting the (x₀ − x̄)² ÷ Sxx term in regression.

    It is dropped when simplifying the formula under time pressure.

    Fix: Write the full variance expression first. Note that the interval is narrowest at x₀ = x̄.

  • Interpreting the interval as a statement about μ.

    Prediction and confidence intervals look alike on paper.

    Fix: Say the interval is for the future observation. The probability statement covers the sample and the new value together.

Worked examples

Example 1

A sample of n = 9 claim amounts (in ₹ thousands) from a normal distribution has x̄ = 50 and s = 6. Find a 95% prediction interval for the next claim amount. The t point with 8 degrees of freedom for 95% two-sided is 2.306.

Show the solution
  1. The target is a single future claim, so use a prediction interval with σ unknown.
  2. Point prediction: x̄ = 50.
  3. Standard deviation of the prediction error: s√(1 + 1/n) = 6 × √(1 + 1/9) = 6 × √(10/9).
  4. √(10/9) = 3.16228 ÷ 3 = 1.05409, so the estimate is 6 × 1.05409 = 6.3246.
  5. Margin: 2.306 × 6.3246 = 14.585.
  6. Interval: 50 ± 14.585 = (35.415, 64.585).

Answer: The 95% prediction interval is about ₹35,415 to ₹64,585.

Example 2

For the data in the previous example, find the 95% confidence interval for the mean claim amount, and state how much wider the prediction interval is.

Show the solution
  1. Standard error of the mean: s/√n = 6 ÷ 3 = 2.
  2. Margin: 2.306 × 2 = 4.612.
  3. Confidence interval: 50 ± 4.612 = (45.388, 54.612), width 9.224.
  4. Prediction interval width: 2 × 14.585 = 29.170.
  5. Ratio of widths: 29.170 ÷ 9.224 ≈ 3.16, which equals √(n + 1) = √10.

Answer: The 95% confidence interval for the mean is (45.39, 54.61) in ₹ thousands. The prediction interval is about 3.16 times wider, because it includes the variance of the new observation.

Exam tips

  • Read the wording carefully. "Future observation", "next claim" or "new policy" means prediction interval. "Mean" or "parameter" means confidence interval.
  • Write the variance of the prediction error before substituting numbers. Examiners give method marks for it.
  • State your normality and independence assumptions in one line. This is often a mark.
  • In a regression question, check whether the question asks for the mean response or a new observation, and use the right formula.
  • Be ready to explain in words why the prediction interval is wider and why it does not shrink to zero as n grows.

Practice questions from Confidence intervals and prediction intervals

Prediction Intervals in other exams

The same ground in other exams, if you are preparing for more than one or want another angle on it.

Prediction Intervals: frequently asked questions

What is the difference between a confidence interval and a prediction interval?

A confidence interval estimates a fixed unknown parameter such as the mean. A prediction interval gives a range for a single future observation. The prediction interval is wider because it also includes the variability of the new observation.

Why is there a 1 inside the square root?

The 1 represents the variance of the new observation Y, which is σ². The 1/n is the variance of x̄ divided by σ². Because Y is independent of the sample, the two variances add.

When do I use the t distribution instead of the normal?

Use the t distribution when σ is unknown and replaced by s, for normal data. The degrees of freedom are n − 1 for a single sample and n − 2 for simple linear regression.

Where is a regression prediction interval narrowest?

It is narrowest at x₀ = x̄, where the term (x₀ − x̄)² ÷ Sxx is zero. It widens as x₀ moves away from the mean of the observed x values, and prediction far outside the data range is unreliable.