EXIN Data Analytics Foundation key terms

The systematic examination of data to uncover patterns, trends, and relationships that support conclusions and decisions. It spans descriptive, diagnostic, predictive, and prescriptive levels of increasing analytical ambition and complexity.

The simplest analytical level, answering what happened by summarizing historical data through counts, averages, trends, and dashboards. It creates a factual baseline but does not explain causes or forecast future outcomes on its own.

Analysis that asks why something happened, using techniques such as drill-down, correlation, and root-cause investigation to explain observed patterns. It goes beyond summarizing results to identify the drivers and conditions behind them.

The use of statistical models and machine learning to estimate future or unknown outcomes from historical data, such as churn likelihood or demand forecasts. Results are probabilities and estimates, not certainties, and depend on data quality.

The most advanced analytical level, recommending actions or decisions by combining predictions with rules, optimization, and constraints. It answers what should be done, often producing ranked options rather than a single forecast value.

Information organized in a predefined format, typically rows and columns in relational tables, that is easy to query with standard tools. Examples include transaction records and customer tables stored in a database.

Information without a predefined tabular model, such as free text, images, audio, and video, which requires specialized processing before analysis. It represents a large share of organizational data and is often stored in data lakes.

Numeric information that can be measured and counted, such as revenue, temperature, or response time, and supports arithmetic operations. It divides into discrete values, like counts, and continuous values, like weights.

Non-numeric information describing qualities, categories, or opinions, such as customer comments or product colors. It is often coded into themes or categories before statistical analysis can be applied.

A categorical measurement scale whose values are labels with no inherent order, such as country, department, or payment method. Only equality comparisons and frequency counts are meaningful on this scale.

A categorical scale whose categories have a meaningful rank but uneven or unknown distances between them, such as satisfaction ratings or education levels. Ordering is valid, but arithmetic on the codes is not.

Numeric data with equal distances between values but no true zero point, such as temperature in Celsius or calendar years. Differences are meaningful, but ratios such as twice as warm are not.

Numeric data with equal intervals and a meaningful zero, such as revenue, weight, or age, allowing all arithmetic including ratios. Statements like one value being twice another are valid on this scale.

The arithmetic average, calculated by summing all values and dividing by the count. It uses every observation but is strongly influenced by extreme values, so it can mislead in skewed distributions.

The middle value of a dataset when observations are sorted, or the average of the two middle values for an even count. It resists distortion from extreme values and suits skewed distributions such as income.

A measure of how far values typically spread around the mean, expressed in the same units as the data. A large value signals high variability; a small value signals that observations cluster near the average.

An observation that lies unusually far from the rest of the dataset, possibly from measurement error, data entry mistakes, or a genuine rare event. Outliers should be investigated, not automatically deleted.

A statistical measure of the strength and direction of a linear relationship between two variables, ranging from minus one to plus one. Correlation does not establish that one variable causes the other.

A relationship in which a change in one variable actually produces a change in another, established through controlled experiments or strong causal designs. It is stronger than correlation and requires more than observational patterns.

The structured sequence of activities used to turn a question into evidence-based insight, typically covering problem definition, data collection, cleaning, analysis, interpretation, and communication of results. Iteration between stages is normal.

A precise, decision-relevant statement of what the analysis must answer, including the scope, stakeholders, and success criteria. A vague question produces unfocused analysis; a well-framed one guides data selection and method choice.

The planned gathering of data from sources such as databases, surveys, sensors, or external providers to answer the defined question. Method choice affects bias, coverage, and the reliability of later conclusions.

Data gathered directly by the analyst or organization for the specific question at hand, for example through a purpose-built survey or sensor. It offers control over definitions but usually costs more to obtain.

Data originally collected by someone else for another purpose and reused in the current analysis, such as public statistics or internal transaction logs. It is cheaper and faster but may not fit the question exactly.

Selecting a subset of a population to analyze and drawing conclusions about the whole. Random sampling supports generalization; biased or convenience samples can produce conclusions that do not hold beyond the collected cases.

A distortion that occurs when the selected sample systematically differs from the population it is meant to represent, for example surveying only online customers. It undermines the validity of conclusions drawn from the sample.

Detecting and correcting errors, duplicates, inconsistencies, missing values, and formatting problems so the dataset is accurate and usable. It is often the most time-consuming stage and directly affects result reliability.

The degree to which data is fit for its intended use, commonly assessed through dimensions such as accuracy, completeness, consistency, timeliness, validity, and uniqueness. Poor quality propagates into every downstream result.

Absent values in a dataset, caused by non-response, system failures, or fields not applicable to a record. How it is handled, by deletion, imputation, or flagging, can materially change analytical conclusions.

Replacing missing values with estimated ones, such as the mean, a regression prediction, or the last known value. It preserves record counts but introduces assumptions that must be documented and tested.

An open-ended investigation of a dataset using summary statistics and visualizations to reveal distributions, patterns, anomalies, and relationships before formal modeling. It guides which hypotheses and methods are worth pursuing.

A controlled experiment comparing two variants, such as a webpage or campaign, assigned randomly to comparable groups to measure which performs better. It supports causal claims when randomization and sample size are adequate.

A key performance indicator: a quantifiable measure tied to a strategic objective, such as monthly churn rate or average order value. Good KPIs are specific, measurable, and reviewed against targets over time.

The graphical representation of data through charts, maps, and diagrams to make patterns and comparisons easier to perceive. Effective visuals match the chart type to the data and the audience's task.

A chart using rectangular bars whose lengths encode values across categories, suitable for comparing discrete groups. Starting the value axis at zero keeps the visual differences proportional to the actual magnitudes.

A chart that groups continuous numeric data into bins and shows the frequency of values in each bin. It reveals the shape of a distribution, including skew and multiple peaks, unlike a bar chart of categories.

A circular chart dividing a whole into slices proportional to each category's share. It works only for a small number of parts of one total and is poor for precise comparison or for values that do not sum to a whole.

A consolidated visual display of current KPIs and metrics, usually interactive and refreshed regularly, designed for monitoring rather than deep exploration. Good dashboards prioritize a few decision-relevant measures over exhaustive detail.

Presenting analytical findings as a coherent narrative that connects evidence to a recommended decision, using structure, context, and visuals. It translates results into language and framing that non-specialist stakeholders can act on.

The framework of roles, policies, standards, and processes that determines who can access, change, and use data, and how its quality is assured. It balances value creation from data against risk, compliance, and accountability.

The operational responsibility for managing data assets day to day, including definitions, quality checks, access approvals, and issue resolution. Stewards act within the policies set by the broader governance framework.

The protection of personal information so that individuals' data is collected, used, and shared fairly and with appropriate consent and safeguards. It addresses legal obligations and individual expectations, distinct from security controls alone.

The European Union's General Data Protection Regulation, governing the processing of personal data of people in the EU. It sets principles such as lawfulness, purpose limitation, and data minimization, plus rights for data subjects.

Irreversibly removing or altering identifying details so that individuals can no longer be recognized, directly or through combination with other data. Done properly, it takes data outside personal-data rules; weak pseudonymization does not.

Systematic error in data or models that produces unfair or inaccurate outcomes for certain groups, arising from sampling, historical patterns, or measurement choices. It must be detected and mitigated throughout the analytics lifecycle.

The application of moral principles to how data is collected, analyzed, and used, covering fairness, transparency, consent, and avoidance of harm. It goes beyond legal compliance to what is right for people affected by the analysis.