Engineering Fluency OS

Why This Matters

A survey says the average American sleeps 7.2 hours per night. But how precise is that number? Is it 7.2 plus or minus 0.1, or plus or minus 2? Without a confidence interval, a point estimate like 7.2 is almost meaningless because it hides how uncertain we are. Confidence intervals put bounds on our uncertainty: "We are 95% confident the true average is between 7.0 and 7.4 hours."

The margin of error determines the width of the interval, and the confidence level (typically 90%, 95%, or 99%) controls how sure we are that the true value falls inside. A higher confidence level means a wider interval — certainty comes at the cost of precision. This tradeoff is central to A/B testing, clinical trials, polling, and any data-driven decision. Understanding confidence intervals is what separates guessing from science.

Define Terms

Visual Model

Sample Datan observations

Sample Mean (x-bar)Point estimate

Standard Errorsigma / sqrt(n)

Z-score (z*)Based on confidence level

Margin of Errorz* x SE

Confidence Intervalx-bar +/- margin of error

Common Levels90%: 1.645, 95%: 1.96, 99%: 2.576

The full process at a glance. Click Start tour to walk through each step.

A confidence interval wraps a point estimate in a range of uncertainty. The width depends on the standard error and the chosen confidence level.

Code Example

Code

// Confidence interval for the mean
// CI = x-bar +/- z* * (sigma / sqrt(n))

function confidenceInterval(data, confidenceLevel) {
  const n = data.length;
  const mean = data.reduce((a, b) => a + b, 0) / n;
  
  // Sample standard deviation
  const variance = data.reduce((a, b) => a + (b - mean) ** 2, 0) / (n - 1);
  const sd = Math.sqrt(variance);
  const se = sd / Math.sqrt(n);
  
  // Z-scores for common confidence levels
  const zScores = { 0.90: 1.645, 0.95: 1.96, 0.99: 2.576 };
  const z = zScores[confidenceLevel];
  
  const marginOfError = z * se;
  return {
    mean: mean,
    marginOfError: marginOfError,
    lower: mean - marginOfError,
    upper: mean + marginOfError
  };
}

// Example: 50 measurements with known properties
const data = [];
for (let i = 0; i < 50; i++) {
  // Simulate Normal(100, 15) using Box-Muller
  const u1 = Math.random(), u2 = Math.random();
  const z = Math.sqrt(-2 * Math.log(u1)) * Math.cos(2 * Math.PI * u2);
  data.push(100 + 15 * z);
}

const ci95 = confidenceInterval(data, 0.95);
console.log(`Mean: ${ci95.mean.toFixed(2)}`);
console.log(`95% CI: (${ci95.lower.toFixed(2)}, ${ci95.upper.toFixed(2)})`);
console.log(`Margin of error: ${ci95.marginOfError.toFixed(2)}`);

const ci99 = confidenceInterval(data, 0.99);
console.log(`99% CI: (${ci99.lower.toFixed(2)}, ${ci99.upper.toFixed(2)})`);
console.log("99% CI is wider than 95% CI");

Interactive Experiment

Try these exercises:

Compute a 95% CI for 10 data points vs. 100 data points from the same distribution. How much narrower is the larger sample?
Change the confidence level from 90% to 95% to 99%. Watch the interval widen. When would you choose each level?
Simulate 100 different 95% CIs from the same population. How many contain the true mean? It should be close to 95.
What happens to the margin of error if you quadruple the sample size? Verify that it halves.
Try computing a CI for a highly skewed distribution (like income). Does it still work? Why or why not?

Quick Quiz

Coding Challenge

Compute Confidence Interval

Write a function `computeCI(data, zStar)` that takes an array of numbers and a z-score, and returns a string in the format 'lower,upper' where lower and upper are the bounds of the confidence interval, each rounded to 2 decimal places. Use the formula: CI = mean +/- zStar * (sampleSD / sqrt(n)), where sampleSD uses n-1 in the denominator (Bessel correction).

Loading editor...

Real-World Usage

Confidence intervals are essential in every data-driven field:

A/B testing: When testing a new feature, companies report the difference in metrics with a confidence interval. If the CI for revenue lift is ($0.50, $2.00), the feature is profitable. If it is (-$0.10, $2.00), we are not sure.
Polling: Election polls report results like "52% +/- 3%." The +/- 3% is the margin of error, and the full CI is (49%, 55%). If 50% is inside the CI, the race is a toss-up.
Clinical trials: Drug effectiveness is reported with confidence intervals. A drug that reduces blood pressure by 10 mmHg with 95% CI (8, 12) is meaningfully effective. A CI of (-2, 22) is too uncertain.
Software performance: Load test results might report "p99 latency: 200ms, 95% CI (185ms, 215ms)." This tells engineers how reliable the estimate is.
Machine learning: Cross-validation accuracy is a mean of k folds. Reporting the CI tells you whether 92% accuracy is meaningfully different from 90%.

Confidence Intervals