MATH 5010 · Sections 16–17

Interval Estimation

An interactive page for confidence intervals, exact intervals for Poisson and Exponential models, Bayesian credible intervals, coverage simulation, and bootstrap intervals.

1. Big picture

From point estimate to interval estimate

A point estimate gives one number, such as $\bar X$. An interval estimate gives a plausible range for the unknown parameter and reports the uncertainty caused by sampling variation.

$$\text{estimate}\ \pm\ \text{critical value}\times \text{standard error}.$$

In a frequentist confidence interval, the confidence level describes the long-run success rate of the procedure. In a Bayesian credible interval, the probability statement is about the posterior distribution of the parameter.

Core formulas to compare

ModelParameterTypical interval idea
Normal, known $\sigma$$\mu$$\bar X\pm z_{1-\alpha/2}\sigma/\sqrt n$
Poisson$\lambda$Invert exact count distribution
Exponential mean$\mu$Use $2\sum X_i/\mu\sim\chi^2_{2n}$
Bayesian Poisson$\lambda$Gamma posterior quantiles
confidence levelstandard errormargin of errorcoveragepivotexact intervalcredible intervalbootstrap
2. Normal mean interval

Known $\sigma$ confidence interval

Use this when $X_1,\dots,X_n\sim N(\mu,\sigma^2)$ with known $\sigma$, or as a large-sample approximation.

$$\bar X\pm z_{1-\alpha/2}\frac{\sigma}{\sqrt n}.$$

Sampling distribution picture

The shaded central region has area equal to the confidence level under the standard normal curve.

3. Interpretation

What does “95% confidence” mean?

Correct frequentist interpretation:
If we repeated the sampling process many times and built an interval by the same rule each time, about 95% of those intervals would contain the true parameter.
Common mistake:
After the data are observed, the fixed frequentist interval either contains the true parameter or it does not. The 95% refers to the procedure, not to a posterior probability for this realized interval.

Bayesian credible intervals are different: a 95% credible interval can be read as containing 95% posterior probability for the parameter, conditional on the model, prior, and observed data.

4. Exact Poisson interval

Poisson mean $\lambda$

Let $X_1,\dots,X_n\sim\text{Poisson}(\lambda)$ and $S=\sum_i X_i$. Then $S\sim\text{Poisson}(n\lambda)$. A classical exact interval for $\lambda$ is obtained by inverting the count distribution.

$$\left[\frac{\chi^2_{2S,\alpha/2}}{2n},\ \frac{\chi^2_{2(S+1),1-\alpha/2}}{2n}\right].$$

The displayed chi-square quantiles use a fast Wilson-Hilferty approximation, which is excellent for teaching and visualization.

Poisson likelihood and interval

The curve is the likelihood shape as a function of $\lambda$; vertical lines mark the interval endpoints and MLE $\hat\lambda=S/n$.

5. Exponential mean interval

Exact CI for mean $\mu$

Let $X_i\sim\mathrm{Exp}(\mu)$ where $\mu$ is the mean. Since $2\sum X_i/\mu\sim\chi^2_{2n}$, an exact confidence interval is

$$\left[\frac{2S}{\chi^2_{2n,1-\alpha/2}},\ \frac{2S}{\chi^2_{2n,\alpha/2}}\right],\qquad S=\sum_i X_i.$$

Pivot distribution

The central chi-square area determines the interval for $\mu$ by solving inequalities for $\mu$.

6. Bayesian credible interval

Poisson-Gamma posterior

For $X_i\sim\text{Poisson}(\lambda)$ with prior $\lambda\sim\text{Gamma}(a,b)$ using rate $b$, the posterior is

$$\lambda\mid x\sim \text{Gamma}\left(a+S,\ b+n\right).$$

Posterior density and credible interval

The shaded posterior area is the equal-tail credible interval. Because $\text{Gamma}(k,r)=\chi^2_{2k}/(2r)$, the calculator uses chi-square quantiles.

7. Coverage simulation

Do confidence intervals cover?

Simulate many samples from a normal population and build the known-$\sigma$ interval each time.

Last 60 simulated intervals

Blue intervals cover the true mean. Red intervals miss it.

8. Bootstrap interval

Percentile bootstrap for a mean

When a closed-form pivot is unavailable, the bootstrap approximates the sampling distribution by resampling from the observed data.

  1. Resample $n$ observations with replacement from the data.
  2. Compute the statistic for each bootstrap sample.
  3. Use empirical quantiles of bootstrap statistics as interval endpoints.

Bootstrap distribution of $\bar X^*$

The histogram shows bootstrap means; vertical lines show the percentile interval.

9. Quick checks

Self-check questions

1. A 95% confidence interval means...

2. Increasing $n$ while keeping $\sigma$ fixed usually makes the interval...

3. In the Poisson-Gamma model, what distribution is the posterior?