92 terms

Statistics for Everyday Life Glossary

The key vocabulary of Statistics for Everyday Life, each term defined in plain language. Start learning in the Statistics for Everyday Life subject, or drill these terms as flashcards.

A/B test: A controlled experiment comparing two versions with real users to see which performs better.
Absolute risk: The plain chance of an outcome, such as 2 in 1,000 per year.
ANOVA: Analysis of variance: a method to test whether several groups’ averages differ, avoiding many separate t-tests.
Base rate: How common something is beforehand, which determines what a positive result means.
Base-rate fallacy: Ignoring the underlying prevalence and over-weighting specific evidence, leading to badly wrong probability estimates.
Bayes’ theorem: A rule for updating the probability of a hypothesis as new evidence arrives.
Bayesian statistics: The school treating probability as a degree of belief, updated with evidence using Bayes’ theorem.
Bin width: The interval size in a histogram, which determines what structure is visible.
Box plot: A summary showing median, middle 50%, range, and outliers — and hiding shape.
Calibration: How well stated probabilities match observed frequencies — if events called 70% likely happen about 70% of the time.
Categorical data: Labels with no inherent order, such as blood type or department.
Cherry-picking: Selecting the window, subgroup, or comparison that supports a case.
Chi-square test: A test of whether two categorical variables are associated more than chance would predict.
Conditional probability: The probability of one event given that another has occurred.
Confidence interval: A range of values compatible with the data, showing both plausibility and magnitude.
Confounder: A hidden variable that influences both cause and effect, muddying a causal claim.
Conjunction fallacy: Judging a specific combination as more likely than one of its parts.
Consumer price index: A measure of inflation based on the changing cost of a fixed basket of goods and services.
Contingency table: A table of counts crossing two categorical variables, used in the chi-square test.
Continuous data: Numerical data measurable on a scale, such as height or time.
Correlation coefficient: A measure from −1 to +1 of how closely points cluster around a straight line.
Data analysis workflow: The end-to-end process of an analysis: question, collect, clean, explore, analyse, and communicate.
Data cleaning: Fixing errors, missing values, duplicates, and inconsistencies so data can be analysed reliably.
Data literacy: The ability to read, question, and reason with data and data-based claims.
Descriptive statistics: Summaries of the data you have, claiming nothing beyond the group measured.
Discrete data: Countable numerical data, such as number of children.
Distribution: The pattern of how values are spread across their possible range.
Effect size: How large an effect is, as distinct from how confident we are it exists.
Expected value: Each outcome multiplied by its probability and summed; suited to repeated decisions.
Exploratory data analysis: Probing and visualising a dataset to find patterns, oddities, and questions before formal analysis.
Extrapolation: Predicting outside the range of the observed data.
Forecast: A probabilistic statement about a future outcome, best judged by its calibration over many predictions.
Frequentist statistics: The school treating probability as the long-run frequency of an event over many repetitions.
Gambler's fallacy: Expecting independent events to self-correct after a streak.
Gambler’s fallacy: The mistaken belief that past independent outcomes change the odds of the next one — that a run of heads makes tails "due".
Histogram: A chart of numerical distribution, with touching bars and a chosen bin width.
Independence: When one event tells you nothing about another.
Independent events: Events where the outcome of one does not change the probability of the other.
Inferential statistics: Using a sample to make claims about a population you did not measure.
Interquartile range: The span of the middle 50% of the data.
Margin of error: The range attributable to sampling variation alone.
Mean: The arithmetic average; uses every value and is pulled by extremes.
Median: The middle value when sorted; resistant to outliers.
Missing data: Absent values in a dataset, which must be handled thoughtfully because they can bias results.
Mode: The most common value; the only average available for categorical data.
Multiple comparisons: Testing many outcomes, making a chance significant result likely.
Multiple regression: A model relating an outcome to several predictors at once, estimating each while holding the others constant.
Multiple-comparisons problem: The inflated risk of false positives that arises from running many statistical tests.
Natural experiment: A real-world situation where chance-like variation lets researchers infer causes without a designed experiment.
Natural variation: The ordinary fluctuation a process produces with nothing having changed.
Non-response bias: Distortion arising when who responds depends on the topic.
Nonparametric methods: Statistical methods that make few assumptions about data’s distribution, often using ranks.
Number needed to treat: How many must receive an intervention for one to benefit.
Numerical data: Actual quantities on which arithmetic is meaningful.
Ordinal data: Ordered categories with unknown spacing between them, such as satisfaction ratings.
Outlier: A value far from the rest, which may be an error or the most important observation.
Overfitting: When a model memorises noise in its training data and fails to generalise to new data.
p-hacking: Trying many analyses until a result crosses the significance threshold, producing false positives.
p-value: The probability of data at least this extreme if there were no real effect.
Percentage point: The arithmetic difference between two percentages, distinct from a percentage change.
Population: Everyone or everything you want to know about.
Probability: A statement of long-run frequency or of degree of belief.
Publication bias: The tendency to publish positive results and file away null results, distorting the evidence base.
R-squared: The share of variation in the outcome explained by the model.
Random sampling: Selecting so that every member of the population has a known chance of inclusion.
Randomized controlled trial: An experiment that randomly assigns subjects to treatment or control to establish cause.
Randomness: Variation without pattern, which nonetheless produces apparent clusters and runs.
Range: Highest minus lowest value.
Regression: Fitting a line or curve to predict one variable from another.
Regression to the mean: The tendency of extreme values to be followed by less extreme ones, because extremes are partly luck.
Relative risk: The ratio between two absolute risks; uninterpretable without a baseline.
Replication: Repeating a study to see whether its result holds up; a cornerstone of reliable science.
Risk difference: The subtraction of one absolute risk from another.
Risk of ruin: The chance of a loss large enough to end the game, which can make a positive-expected-value bet unwise.
Robust statistics: Methods resistant to outliers and violated assumptions, such as using the median instead of the mean.
Sabermetrics: The statistical analysis of baseball, and by extension the data-driven analysis of sports.
Sample: The subset actually measured, used to reason about the population.
Scatter plot: A chart of two numerical variables, one point per observation.
Selection bias: Distortion arising when some of the population could not be sampled.
Sensitivity: The share of true cases a test correctly identifies as positive.
Simpson’s paradox: When a trend that appears in groups reverses or disappears once the groups are combined.
Skew: Asymmetry in a distribution, which pulls the mean away from the median.
Specificity: The share of true non-cases a test correctly identifies as negative.
Spurious correlation: A correlation that appears by chance or through a hidden cause, common when searching huge datasets.
Standard deviation: Typical distance from the mean, in the same units as the data.
Statistical power: A study's ability to detect an effect that genuinely exists.
Statistical significance: A result unlikely to arise from noise alone, by a chosen threshold.
Statistics: The discipline of drawing conclusions from data in the presence of uncertainty.
Survivorship bias: Analysing only what remained, missing what disappeared.
t-test: A test comparing two groups’ averages to judge whether their difference is likely more than chance.
Training and test data: Splitting data so a model is built on one part and honestly evaluated on unseen data.
Variance: A measure of how far values sit from the mean, in squared units.

← All subject glossaries