Sample Size Determination — How Many Observations Do You Need?

Foundations of Statistics

Planning for Statistical Success

Sample size determination ensures studies have adequate power to detect meaningful effects while avoiding unnecessary data collection. It balances statistical requirements against practical constraints like time, cost, and ethics.

Clinical Trials — Ensuring sufficient power to detect clinically meaningful treatment effects
Market Research — Optimizing survey costs while maintaining estimate precision
Quality Assurance — Determining inspection sample sizes for reliable defect detection

The right sample size is the foundation of trustworthy statistical conclusions.

What Is Sample Size Determination?

DfSample Size Determination

Sample size determination is the process of calculating the number of observations needed to achieve a desired level of precision and power in a statistical study. Too few observations leads to inconclusive results; too many wastes resources.

Core Formulas

Sample Size for Estimating a Mean

n = \left(\frac{z_{\alpha/2} \cdot \sigma}{E}\right)^2

Here,

$n$ =Required sample size
$z_{\alpha/2}$ =Critical z-value for the desired confidence level
$\sigma$ =Population standard deviation (estimated)
$E$ =Desired margin of error

Sample Size for Estimating a Proportion

n = \frac{z_{\alpha/2}^2 \cdot p(1-p)}{E^2}

Here,

$p$ =Estimated population proportion
$E$ =Desired margin of error

Conservative Estimate for p

When $p$ is unknown, use $p = 0.5$ for the most conservative (largest) sample size, since $p(1-p)$ is maximized at $p = 0.5$ with value $0.25$ .

Derivation: Inverting the Margin of Error

ThSample Size from Margin of Error

Starting from the margin of error formula $E = z_{\alpha/2}\sigma/\\sqrt{n}$ , solve for $n$ :

\sqrt{n} = \frac{z_{\alpha/2} \cdot \sigma}{E} \implies n = \left(\frac{z_{\alpha/2} \cdot \sigma}{E}\right)^2

Since $n$ must be an integer, always round up to the next whole number: $n = \\lceil (z_{\alpha/2}\sigma/E)^2 \\rceil$ .

Proof sketch: The margin of error is the half-width of the CI. Setting $E$ to the desired precision and solving for $n$ gives the minimum sample size that achieves that precision. Rounding up ensures the actual margin is at most $E$ .

Sample Size for Hypothesis Testing (Power Analysis)

Sample Size for Two-Sided Test

n = \frac{(z_{\alpha/2} + z_{\beta})^2 \cdot 2\sigma^2}{\delta^2}

Here,

$\alpha$ =Significance level (Type I error rate)
$\beta$ =Type II error rate; power $= 1 - \beta$
$\sigma$ =Population standard deviation
$\delta$ =Minimum detectable effect size

ThPower and Sample Size Trade-off

For a fixed effect size $\delta$ and significance level $\alpha$ , the required sample size scales as:

n \propto \frac{\sigma^2}{\delta^2}

This reveals two critical insights:

Detecting smaller effects requires more data: halving $\delta$ requires $4\times$ the sample.
More variable populations require more data: doubling $\sigma$ requires $4\times$ the sample.

Worked Example: Clinical Trial Design

A pharmaceutical company wants to detect a 3 mmHg reduction in blood pressure with 80% power at $\alpha = 0.05$ . Prior studies suggest $\sigma = 8$ mmHg.

Step 1: Identify parameters: $\delta = 3$ , $\sigma = 8$ , $\alpha = 0.05$ ( $z_{0.025} = 1.96$ ), power $= 0.80$ ( $\beta = 0.20$ , $z_{0.20} = 0.842$ ).

Step 2: Compute:

n = \frac{(1.96 + 0.842)^2 \times 2 \times 64}{9} = \frac{(2.802)^2 \times 128}{9} = \frac{7.851 \times 128}{9} = \frac{1004.9}{9} \approx 112

Step 3: Round up: $n = 112$ per group, total $N = 224$ .

Accounting for Attrition

In practice, inflate the required $n$ by the expected dropout rate $d$ : $n_{\text{adjusted}} = n/(1-d)$ . For a 15% dropout rate: $112/0.85 \\approx 132$ per group.

The Effect Size Pyramid

Sample Size Dependencies

The required sample size depends on four quantities:

Factor	Effect on $n$	Example
Effect size $\delta$	$n \\propto 1/\delta^2$	Halving effect $\\to$ $4\times$ sample
Standard deviation $\sigma$	$n \\propto \sigma^2$	Doubling variance $\\to$ $4\times$ sample
Power $1-\beta$	$n \\propto (z_{\alpha/2}+z_\beta)^2$	80% to 90% power $\\to$ $\\sim 1.7\times$ sample
Significance $\alpha$	$n \\propto z_{\alpha/2}^2$	$0.05$ to $0.01$ $\\to$ $\\sim 1.4\times$ sample

Python Implementation

import numpy as np
from scipy import stats

def sample_size_mean(sigma, E, alpha=0.05):
    """Sample size for estimating a mean with margin of error E."""
    z = stats.norm.ppf(1 - alpha / 2)
    return int(np.ceil((z * sigma / E) ** 2))

def sample_size_proportion(p, E, alpha=0.05):
    """Sample size for estimating a proportion with margin of error E."""
    z = stats.norm.ppf(1 - alpha / 2)
    return int(np.ceil(z**2 * p * (1 - p) / E**2))

def sample_size_two_sample(delta, sigma, alpha=0.05, power=0.80):
    """Sample size per group for two-sample t-test."""
    z_alpha = stats.norm.ppf(1 - alpha / 2)
    z_beta = stats.norm.ppf(power)
    return int(np.ceil(2 * sigma**2 * (z_alpha + z_beta)**2 / delta**2))

# Example 1: Mean estimation
print(f"n for σ=10, E=2, 95% CI: {sample_size_mean(10, 2, 0.05)}")
print(f"n for σ=10, E=1, 95% CI: {sample_size_mean(10, 1, 0.05)}")

# Example 2: Proportion estimation
print(f"n for p=0.5, E=0.03, 95% CI: {sample_size_proportion(0.5, 0.03)}")
print(f"n for p=0.2, E=0.03, 95% CI: {sample_size_proportion(0.2, 0.03)}")

# Example 3: Two-sample test
print(f"n per group for δ=3, σ=8, 80% power: {sample_size_two_sample(3, 8, 0.05, 0.80)}")
print(f"n per group for δ=2, σ=8, 80% power: {sample_size_two_sample(2, 8, 0.05, 0.80)}")

Key Takeaways

Summary: Sample Size Determination

For precision: $n = (z_{\alpha/2}\sigma/E)^2$ — round up
For power: $n = (z_{\alpha/2}+z_\beta)^2 \cdot 2\sigma^2/\delta^2$ — per group
The $1/\delta^2$ relationship means detecting small effects is expensive
Always estimate $\sigma$ from prior studies or pilot data before computing $n$
Account for attrition, clustering, and multiple comparisons in your final $n$

Sample Size Determination — How Many Observations Do You Need?

Sample Size Determination — How Many Observations Do You Need?

Planning for Statistical Success

What Is Sample Size Determination?

DfSample Size Determination

Core Formulas

Sample Size for Estimating a Mean

Sample Size for Estimating a Proportion

Derivation: Inverting the Margin of Error

ThSample Size from Margin of Error

Sample Size for Hypothesis Testing (Power Analysis)

Sample Size for Two-Sided Test

ThPower and Sample Size Trade-off

Worked Example: Clinical Trial Design

The Effect Size Pyramid

Sample Size Dependencies

Python Implementation

Key Takeaways

Summary: Sample Size Determination

Premium Content

Need Expert Statistics Help?