Scientists use statistics because scientific observations contain variability and uncertainty, and researchers need mathematical methods for describing data, estimating quantities, comparing groups, identifying relationships and evaluating hypotheses. Natural systems vary from one observation to another, and measurements are affected by sampling limitations and measurement error. Statistics provides a framework for distinguishing patterns that are likely to represent underlying processes from patterns that could reasonably arise through random variation. It also allows scientists to make inferences from samples to populations when appropriate assumptions are satisfied. Statistics is therefore not simply a collection of calculations used after an experiment. It influences study design, sampling, measurement, uncertainty estimation, model construction and interpretation.
Descriptive statistics summarise observed data. Measures such as the mean, median, mode, range, variance and standard deviation describe different aspects of a dataset. Quantiles and percentiles can describe the distribution without assuming symmetry, while graphical methods can reveal skewness, clusters, outliers and temporal patterns. The mean is sensitive to extreme observations, whereas the median is more robust in many skewed distributions. Standard deviation describes the dispersion of observations around the mean under its conventional definition. Descriptive statistics do not by themselves establish causation or generalise beyond the data. Their purpose is to represent the structure of the observed dataset accurately so that researchers can understand what was measured before making broader inferences.
Inferential statistics allow scientists to estimate properties of a larger population using a sample. A sample statistic, such as a sample mean, provides information about a population parameter, but sampling variability means that different samples can produce different estimates. Confidence intervals provide ranges constructed by specified statistical procedures that describe uncertainty associated with an estimate under the relevant assumptions. Hypothesis tests can evaluate the compatibility of data with a null hypothesis. Bayesian methods can combine prior information with observed data to produce posterior distributions. Both frequentist and Bayesian approaches have different interpretations and assumptions. Statistics therefore does not remove uncertainty. Instead, it quantifies and manages uncertainty using explicit mathematical models.
Statistics is also essential for experimental design. Before collecting data, researchers can use statistical power calculations to estimate how many observations may be required to detect an effect of a specified magnitude with a chosen probability under particular assumptions. Randomisation procedures can be designed statistically, and blocking can reduce variation caused by known nuisance factors. Factorial designs can investigate multiple variables and interactions efficiently. Repeated measures designs can account for measurements taken from the same subjects or experimental units. Appropriate design can substantially improve the information obtained from a given number of observations. Statistical analysis cannot fully repair a poorly designed experiment because systematic bias, inadequate sampling or uncontrolled confounding may remain regardless of the sophistication of later calculations.
Scientists use statistics to quantify relationships among variables. Correlation measures certain forms of association, while regression models estimate relationships between a response variable and one or more explanatory variables. More complex models can incorporate nonlinear relationships, interactions, repeated observations, hierarchical structures and spatial or temporal dependence. Statistical modelling can be used for prediction, estimation and hypothesis testing. However, correlation does not automatically establish causation. A relationship may arise because one variable influences another, because both are influenced by a third factor or because of selection or measurement effects. Causal inference requires appropriate study design and assumptions, such as randomisation in experiments or carefully justified identification strategies in observational studies.
Statistics is particularly important when scientists compare observed results with a null hypothesis. A p value measures the probability of obtaining data at least as incompatible with the specified null model as the observed data, assuming that the null hypothesis and statistical assumptions are correct. It is not the probability that the null hypothesis is true. Statistical significance therefore should not be treated as a direct measure of scientific importance.
Effect size and uncertainty are often more informative for understanding practical or biological significance. Multiple testing can increase the probability of apparently significant results occurring by chance, so researchers may use corrections or hierarchical approaches when appropriate. Statistical interpretation requires understanding the design and assumptions rather than relying on a single threshold.
Statistics also helps scientists combine evidence from multiple studies. Meta analysis can estimate an overall effect from several studies when the studies are sufficiently comparable and appropriate statistical models are used. Researchers can examine heterogeneity to determine whether effects differ substantially across studies or contexts. Time series analysis can identify trends and recurring patterns while accounting for temporal dependence. Spatial statistics can analyse geographic clustering and spatial relationships. Survival analysis can investigate time until events occur. Multivariate methods can examine many variables simultaneously. The diversity of statistical methods reflects the diversity of scientific questions. Choosing a method requires understanding the structure of the data, measurement scale, sampling process and scientific objective.
Statistics has limitations because statistical conclusions depend on assumptions, data quality and study design. A mathematically correct calculation applied to biased or unrepresentative data can produce a precise but misleading result. Small samples can produce unstable estimates, while very large samples can detect trivial differences. Missing data, measurement error, confounding, nonindependence and model misspecification can distort conclusions. Statistical models are representations rather than direct descriptions of reality.
Scientists therefore inspect assumptions, perform diagnostic tests, conduct sensitivity analyses and compare alternative models when appropriate. Statistics is most powerful when integrated with subject matter knowledge, careful experimental design, measurement science and transparent reporting. It provides a disciplined framework for reasoning under uncertainty rather than a mechanism for converting uncertain observations into certainty.
