Q.How can systematic errors be minimised or eliminated?

Measurement Error Estimation: Why the Key Formulas Hold

Measurement error estimation is about quantifying how much a measured value might differ from the true value. The core idea is that no measurement is perfect — every reading contains some uncertainty.

1. The Fundamental Idea: True Value vs. Measured Value

Let’s start with the basic relationship:

The error ($\varepsilon$) is the difference:

Why this matters: We never know the true value exactly — if we did, there would be no error to estimate. So we must infer the error from repeated measurements.

2. Mean Error (Bias) — Why We Average

If you take $n$ measurements $x_1, x_2, \dots, x_n$, the mean is:

Why does the mean estimate the true value?

Assume each measurement has a random error $\varepsilon_i$ with zero mean (no systematic bias). Then:

As $n$ increases, the average of random errors $\frac{1}{n} \sum \varepsilon_i$ tends to zero (by the law of large numbers). So:

Key insight: Averaging cancels out random errors, but not systematic errors (bias).

3. Standard Deviation of the Mean — Why $\sigma / \sqrt{n}$

The standard error of the mean (SEM) is:

Derivation (why this formula):

1. Each measurement $x_i$ has variance $\sigma^2$ (spread around the true value).
2. The variance of the mean $\bar{x}$ is:

3. Since all $\text{Var}(x_i) = \sigma^2$ and they are independent:

4. Standard deviation is the square root of variance:

Why this makes sense: More measurements ($n$ larger) reduce uncertainty — but only as $\sqrt{n}$, not linearly. Doubling $n$ reduces error by only $\approx 30\%$.

4. Propagation of Errors — Why We Add Variances

When a result $z$ depends on measured quantities $x$ and $y$ (e.g., $z = x + y$ or $z = x \cdot y$), errors propagate.

Case 1: Addition/Subtraction

If $z = x + y$, and errors $\Delta x$, $\Delta y$ are independent:

Why?

Variance of sum = sum of variances (for independent variables):

So the uncertainty adds in quadrature (not linearly). This is because errors can partially cancel.

Case 2: Multiplication/Division

If $z = x \cdot y$, then:

Derivation (why relative errors add):

1. Take natural log: $\ln z = \ln x + \ln y$
2. Differentiate: $\frac{dz}{z} = \frac{dx}{x} + \frac{dy}{y}$
3. For small independent errors, variances add:

Key insight: Relative uncertainties propagate the same way absolute uncertainties do for sums.

5. The General Formula (Why It's a Taylor Expansion) …