Statistics
& Error Analysis

Scroll to explore
← Physics Labs Error Analysis & Statistics

PHY 113 · 114 · 121 · 123 · 181

Error Analysis
& Statistics

Every measurement you make has an uncertainty. Understanding that uncertainty is what separates a result from a guess.

Scroll to explore
Daily Polling Report — October 2024
"Harris surges ahead of Trump in new national poll!"
Harris
49%
Trump
46%

Each bar carries a ±3% margin of error

⚠ The difference (3%) equals the margin of error — this "surge" is statistically meaningless.

Is 2024 really the hottest year on record?

Global average temperature in 2024 was reported as $+1.60^\text{o}\text{C}$ above pre-industrial levels — about $0.12^\text{o}\text{C}$ warmer than the previous record set in 2023.

But the instrumental uncertainty in global temperature averages is roughly $\pm0.10^\text{o}\text{C}$. That $0.12^\text{o}\text{C}$ difference barely exceeds the measurement error.

The record is likely real — but quantifying uncertainty is precisely how science makes that claim responsibly.

Global temperature anomaly (°C above pre-industrial)
2021
+1.12 °C
2022
+1.15 °C
2023
+1.48 °C
2024
+1.60 ± 0.10 °C

Source: Copernicus Climate Change Service, 2025

Measuring Gravitational Acceleration
9.5 9.6 9.7 9.8 9.9 10.0 10.1 g (m/s²) Student A Student B

Why don't
measurements repeat?

When you measure the same pencil $10$ times with a ruler, you don't get $12.7$ cm every time. You get $12.5$, $12.8$, $12.6$, $12.9$, etc.

This scatter isn't "bad" — it's data. It tells you about the limits of your measurement tool and technique.

Random error: ruler markings are discrete, your eye position varies, you press differently each time.

Systematic error: a bent ruler always reads high; a cold ruler contracts; parallax bias from one direction.

Both shape your result. Understanding which is which helps you improve.

Measuring a pencil: 10 trials Pencil length (cm) 12.2 12.4 12.6 12.8 13.0 13.2 12.7 cm±0.5 cm
Accuracy vs. Precision High Accuracy High Precision Low Accuracy High Precision High Accuracy Low Precision Low Accuracy Low Precision ● = target center ○ = darts

Where is your data clustered?

Measure a pendulum's period $N=50$ times. Each measurement has random error. Plot all values as a histogram — they cluster around $2.018$ seconds in a bell-shaped (normal) distribution.

The population mean $μ$ is the true period (unknown). The sample mean $\bar{x}$ is what you calculate from your N measurements:

$\displaystyle \bar{x} = \frac{1}{N} \sum_{i=1}^{N} x_i$

The more measurements you take, the closer $\bar{x}$ gets to $\mu$. The bell curve shape is universal: most measurements cluster near the true value, with fewer far away. This fact is known as the central limit theorem

50 measurements of pendulum period 1.90 1.95 2.00 2.05 2.10 0 5 10 15 μ = 2.00 s x̄ ≈ 2.018 s Period (seconds) Count Data (50 measurements) Population distribution
Two datasets — different spread small σ large σ
Wait time for a train (100 samples, λ=3 min) 0 1 2 3 4 5 6 7 0 10 20 Minutes waited Count samples Poisson PMF

Combining Samples

Each data sample will have its own mean, $\bar{x}$ and standard deviation, $\bar{s}_x$. When you combine multiple samples, you will get a distribution of sample means.

The standard error of the mean (SEM) is:

$\bar{s}_{\bar{x}} = \frac{\sigma}{\sqrt{N}}$

Each time you double N, the error shrinks by √2. Taking 10 measurements reduces error by √10 ≈ 3.16×. This is why experiments repeat measurements: averaging is powerful.

From measurement to result Input radius r r ± σ_r A = πr² Output area A A ± σ_A Nonlinear functions stretch errors unevenly
Equation Reference Sample Mean: x̄ = (1/N) Σ xᵢ Standard Deviation: σ = √[ (1/N) Σ(xᵢ - x̄)² ] Standard Error (SEM): σ_x̄ = σ / √N Poisson Statistics: σ = √N (for counting) Error Propagation: σ_f ≈ |df/dx| σ_x Keep these formulas close when designing experiments