Every analytical measurement carries uncertainty arising from random (indeterminate) and systematic (determinate) errors. Statistical tools — mean, standard deviation, relative standard deviation, confidence intervals, and significance tests such as the t-test and F-test — allow chemists to characterize measurement uncertainty and compare results rigorously. Propagation of uncertainty describes how errors in individual measurements combine in calculated quantities. Outlier identification using the Q-test or Grubbs' test maintains data integrity.
Practice computing confidence intervals and propagating uncertainty through multi-step calculations by hand before relying on spreadsheet functions. Simulating datasets with known parameters builds intuition for how sample size and variability affect conclusions.
Every measurement in analytical chemistry is imperfect. The goal of error analysis is not to eliminate uncertainty — that is impossible — but to characterize it honestly so that results can be interpreted correctly. Errors fall into two fundamentally different categories. Random (indeterminate) errors scatter results unpredictably around the true value: sometimes too high, sometimes too low, with no consistent direction. They arise from uncontrollable fluctuations in the instrument, the environment, or the analyst's technique. Systematic (determinate) errors push all measurements in the same direction — a miscalibrated balance always reads heavy, a pipette that delivers less than labeled always underestimates concentration. The two types require different remedies: averaging more replicates reduces the impact of random error, but systematic errors must be hunted down and eliminated at the source.
Standard deviation (s) is the primary descriptor of random variability. It tells you how spread out your replicate measurements are — a large s means noisy data, a small s means tight precision. But standard deviation does not tell you how well you know the mean. That is the job of the standard error of the mean (SEM = s/√n): it quantifies how much the sample mean would vary if you repeated the entire experiment. Taking more replicates shrinks the SEM but does not necessarily change s. Confusing these two quantities is one of the most common errors in reporting analytical results.
Confidence intervals connect statistics to real-world decisions. A 95% confidence interval for a mean says: if you repeated this measurement procedure many times, 95% of the intervals constructed this way would contain the true value. Wider intervals reflect either greater variability or smaller sample size. In practice, a confidence interval tells a chemist whether a result is meaningfully different from a target value — for instance, whether a drug formulation meets its specification. The t-test formalizes this comparison by asking whether an observed difference is larger than would be expected by chance alone, given the measured variability.
Propagation of uncertainty addresses a practical reality: most analytical results are calculated from multiple raw measurements, each with its own uncertainty. If you pipette two volumes and subtract them, the uncertainty in the difference depends on the uncertainties of both individual pipettings. Propagation rules (based on partial derivatives) tell you how uncertainties combine. The key insight is that adding or subtracting quantities causes absolute uncertainties to add in quadrature, while multiplying or dividing causes relative (percent) uncertainties to add in quadrature. Tracking uncertainty through a calculation ensures that the final reported result is not falsely precise.