Confidence Interval vs Margin of Error: What's the Difference?

Confidence Interval vs Margin of Error: The Short Answer

Consider a poll that finds 52% support with a 95% CI of (49%, 55%). The margin of error is ±3 percentage points—the distance from the sample estimate (52%) to either bound. If you know the MOE, you can reconstruct the CI. If you know the CI, you can calculate the MOE. The two values contain identical information.

Definitions: Confidence Interval and Margin of Error

What Is a Confidence Interval?

The term was introduced by the statistician Jerzy Neyman in a landmark 1937 paper and formalized alongside the decision-theoretic framework developed with Egon Pearson. Unlike a single point estimate—which gives one number for a population parameter—a confidence interval acknowledges that any single sample will differ from the true population value by some amount of sampling error. The interval captures that uncertainty explicitly.

A 95% confidence interval does not mean there is a 95% probability that the true parameter lies within the computed range. The true parameter either is or is not inside any specific interval you calculate—probability no longer applies to it once the data are observed. The correct interpretation: if you repeated the study 100 times under identical conditions, roughly 95 of the resulting confidence intervals would contain the true population value. This interpretation, though subtle, matters for how you communicate uncertainty in research and reporting. The American Statistical Association provides formal guidance on this distinction in its statements on statistical inference.

What Is the Margin of Error?

The margin of error appears most often in news reports of opinion polls. When a media outlet writes "±3 percentage points," they typically mean the margin of error at the 95% confidence level. Organizations like Gallup and the Pew Research Center publish their methodology alongside reported margins of error, including the confidence level used and the effective sample size after weighting.

The margin of error accounts only for random sampling error. It does not capture non-sampling errors like question wording bias, non-response bias, or coverage errors—errors that arise from how the survey was designed or conducted rather than from the randomness of who was selected. This is a common misconception worth addressing directly: a small margin of error does not guarantee an accurate poll.

The Mathematical Relationship Between CI and Margin of Error

The connection is direct and algebraic. Any symmetric confidence interval can be written as:

This means: if you see a confidence interval reported as (L, U), you can always recover the margin of error as MOE = (U − L) / 2. If you see a margin of error, the confidence interval is simply (point estimate − MOE, point estimate + MOE).

The margin of error itself is built from two pieces: the critical value (z* for proportions and large samples, t* for means with small samples) and the standard error (SE), which measures how much a sample statistic varies from sample to sample.

The critical value z* comes from the standard normal distribution and depends on your chosen confidence level. The three most common values are well established:

Confidence Interval and Margin of Error Formulas

Formula for a Population Mean

When σ is unknown (the usual case in practice), replace σ with the sample standard deviation s and replace the z* critical value with the t* critical value at n−1 degrees of freedom. See the confidence interval for a mean guide for the full derivation using the t-distribution.

Formula for a Population Proportion

This is the formula used by virtually all election pollsters. A common approximation when p̂ is unknown or close to 0.5 is MOE ≈ 1/√n (at 95% confidence), which gives a quick upper bound. With n = 1,000, this yields MOE ≈ 3.16%—close to the ±3% figure you see in most major national polls. You can compute the exact value for your dataset with the confidence interval calculator.

For proportions near 0 or 1 where the normal approximation breaks down, statisticians prefer the Wilson score interval, which is documented on the Wilson score interval page. See also the confidence interval for proportion full guide.

Confidence Interval vs Margin of Error: Full Comparison

Confidence Interval vs Confidence Level

The confidence level (e.g., 95%) is the input to the calculation. The confidence interval is the output. Raising the confidence level from 95% to 99% makes the resulting interval wider, because you need to capture a higher proportion of all possible sample outcomes. The two are often confused in casual conversation but refer to distinct things: one is a probability, the other is a range.

Margin of Error vs Standard Error

Standard error and margin of error are related but not the same. The standard error is the standard deviation of the sampling distribution of the statistic—a measure of how much the statistic would vary across repeated samples. The margin of error scales the standard error by the critical value: MOE = z* × SE. With a 95% CI (z* = 1.96), the margin of error is roughly twice the standard error.

Free CI & Margin of Error Calculator

5 Fully Worked Examples

Example 1 — Election Poll (Proportion)

Example 2 — Clinical Trial (Mean)

Example 3 — Student Test Scores

Example 4 — Customer Satisfaction Survey

Example 5 — Product Quality Control

How Sample Size Affects the Margin of Error

The relationship between sample size and margin of error follows a square-root law. Because SE = σ/√n, the margin of error is proportional to 1/√n. Doubling the sample size multiplies the MOE by 1/√2 ≈ 0.707, reducing it by about 29%. To cut the MOE in half, you must quadruple the sample size.

You can plan your study's sample size in advance using the sample size calculator. For a desired MOE of E at confidence level z*, you need at minimum n = (z* × σ / E)² observations.

The CLEAR Framework: Building Any Confidence Interval

Constructing and reporting a confidence interval correctly involves five decisions, each one feeding into the next. The CLEAR framework captures this sequence:

Real-World Applications

Common Misconceptions

When to Report a CI vs a Margin of Error

Entity and Formula Glossary

Related Concepts: Prediction and Tolerance Intervals

Confidence intervals are sometimes confused with prediction intervals and tolerance intervals. They answer different questions:

For simple linear regression, a similar distinction arises between the CI for the mean response and the prediction interval for a new observation. See simple linear regression for the full treatment of regression intervals.

Calculating CI and MOE in R, Python, and Excel

Python (SciPy)

import scipy.stats as stats
import numpy as np

data = np.array([12.3, 14.1, 13.7, 11.9, 15.2, 13.4])
n = len(data)
xbar = np.mean(data)
se = stats.sem(data)

# 95% CI using t-distribution
ci = stats.t.interval(0.95, df=n-1, loc=xbar, scale=se)
moe = (ci[1] - ci[0]) / 2

print(f"Mean: {xbar:.3f}")
print(f"MOE: ±{moe:.3f}")
print(f"95% CI: ({ci[0]:.3f}, {ci[1]:.3f})")

R

# Proportion CI (exact binomial)

prop.test(x = 540, n = 1000, conf.level = 0.95)

# Or manually:

p_hat <- 0.54
n <- 1000
z_star <- qnorm(0.975)  # 1.96

se <- sqrt(p_hat * (1 - p_hat) / n)
moe <- z_star * se

ci_lower <- p_hat - moe
ci_upper <- p_hat + moe

cat("MOE:", round(moe, 4), "\n")
cat("95% CI: (", round(ci_lower, 4), ", ", round(ci_upper, 4), ")\n")

Microsoft Excel

In Excel, use =CONFIDENCE.NORM(alpha, std_dev, n) for large samples (returns the MOE directly) or =CONFIDENCE.T(alpha, std_dev, n) when σ is unknown. The confidence interval is then =AVERAGE(range) ± CONFIDENCE.NORM(0.05, STDEV(range), COUNT(range)).

Frequently Asked Questions

Further Reading and Resources