Data glossary
Clear, jargon-free definitions of the terms you'll meet in analytics, statistics, data engineering and visualization.
- A/B test
- A controlled experiment that randomly splits users between two versions (A and B) to measure which performs better on a chosen metric. See the A/B Test Calculator.
- Aggregation
- Combining many values into one summary, such as a sum, count, average, minimum or maximum, usually per group.
- API
- Application Programming Interface: a defined way for programs to request data or actions from another system, commonly over HTTP returning JSON.
- Anomaly detection
- Finding data points that deviate strongly from expected patterns, such as fraud, sensor faults or traffic spikes.
- Baseline
- The reference value you compare against, such as last year's figure or the control group's conversion rate.
- Bias (sampling)
- Systematic error that makes a sample unrepresentative of the population, for example surveying only customers who already love your product.
- Business intelligence (BI)
- Tools and practices for turning business data into reports and dashboards that support decisions.
- Cardinality
- The number of distinct values in a column. Country has low cardinality, while email address has high cardinality.
- Categorical variable
- A variable whose values are labels or groups (plan type, region) rather than measured quantities.
- Churn rate
- The share of customers who stop using a product in a given period.
- Cohort analysis
- Grouping users by a shared start event (for example their sign-up month) and tracking their behaviour over time.
- Confidence interval
- A range, computed from sample data, that would contain the true value in a stated share of repeated samples (for example 95%).
- Conversion rate
- The share of visitors or users who complete a target action, such as a purchase or a sign-up.
- Correlation
- A measure (from −1 to 1) of how strongly two variables move together. Correlation alone does not prove causation.
- CSV
- Comma-separated values: a plain-text table format where each line is a row and commas separate fields. See CSV vs JSON vs Parquet.
- Dashboard
- A visual display of the key metrics needed to monitor a goal at a glance. See KPI dashboard design.
- Data cleaning
- Detecting and fixing errors, inconsistencies and gaps in data before analysis. See the Data Cleaner.
- Data lake
- Storage that holds large volumes of raw data in many formats, structured later when it is read.
- Data warehouse
- A database optimised for analytics, holding cleaned, structured data from many sources.
- Data model
- The structure of tables, fields and relationships that defines how data is organised and joined.
- Dimension
- An attribute you group or filter by, such as date, product or region, as opposed to a measure.
- Distribution
- How values of a variable are spread out: their centre, range and shape (symmetric, skewed, multi-peaked).
- ETL / ELT
- Extract, Transform, Load: moving data from sources into a destination. In ELT, raw data is loaded first and transformed inside the warehouse.
- Feature
- In machine learning, an input variable used to make predictions.
- Funnel
- A sequence of steps users go through toward a goal, analysed to see where they drop off.
- Granularity
- The level of detail of data, for example daily vs monthly, or per order vs per customer.
- Histogram
- A chart that counts how many values fall into each range (bin), revealing a distribution's shape.
- Hypothesis test
- A statistical procedure for judging whether observed data is consistent with a default assumption (the null hypothesis).
- Imputation
- Filling in missing values with estimates such as the median, a model prediction or the previous value.
- Interquartile range (IQR)
- The spread of the middle 50% of values (Q3 − Q1). Values beyond 1.5×IQR from the quartiles are often flagged as outliers.
- Join
- Combining rows from two tables based on a matching key, such as customer ID.
- JSON
- JavaScript Object Notation: a text format for nested data made of objects and arrays, widely used by APIs. See the CSV ⇄ JSON Converter.
- KPI
- Key performance indicator: a metric tied directly to a goal and tracked over time.
- Mean
- The arithmetic average: sum of values divided by their count. Sensitive to outliers.
- Measure
- A numeric value you aggregate, such as revenue, units or duration, as opposed to a dimension.
- Median
- The middle value when data is sorted. Robust to outliers, so often better than the mean for incomes or prices.
- Metadata
- Data about data: source, definitions, units, update date, licence and owner.
- Minimum detectable effect (MDE)
- The smallest true change an experiment is designed to reliably detect. See the Sample Size Calculator.
- Normalization
- Either rescaling numbers to a common range, or (in databases) organising tables to reduce redundancy.
- Null / missing value
- The absence of a value. Different from zero or an empty string, and it needs deliberate handling.
- Outlier
- A value far from the rest of the data. It may be an error or a real, important extreme.
- p-value
- The probability of seeing a result at least this extreme if there were truly no effect. Small p-values suggest the null hypothesis is unlikely.
- Parquet
- A compressed, columnar file format designed for fast analytics on large datasets.
- Percentile
- The value below which a given percentage of observations fall. The 90th percentile exceeds 90% of values.
- Pivot table
- A summary table that groups data by one or more dimensions and aggregates a measure. See the Pivot Table.
- Statistical power
- The probability that a test detects an effect of a given size when it truly exists. 80% is a common target.
- Primary key
- A column (or set of columns) that uniquely identifies each row in a table.
- Regression
- A family of methods that model how an outcome variable depends on one or more input variables.
- Sample
- A subset of a population used to draw conclusions about the whole.
- Schema
- The defined structure of a dataset: column names, data types and constraints.
- Seasonality
- Regular, repeating patterns tied to the calendar, such as weekends, holidays or quarter-ends.
- Segmentation
- Dividing customers or records into groups with shared characteristics to analyse or target them separately.
- Statistical significance
- A result is called significant when its p-value is below a pre-chosen threshold (α, often 0.05).
- SQL
- Structured Query Language: the standard language for querying and managing relational databases.
- Sample ratio mismatch (SRM)
- When an experiment's traffic split differs from its design more than chance allows. It signals a bug that invalidates results.
- Standard deviation
- A measure of how spread out values are around the mean, in the same units as the data.
- Standard error
- The standard deviation of a statistic (like a sample mean) across repeated samples. It shrinks as sample size grows.
- Time series
- Data points recorded in time order, such as daily sales or hourly temperature.
- Tidy data
- A layout where each variable is a column, each observation is a row and each value is a cell. It makes analysis much easier.
- Type I / Type II error
- A false positive (detecting an effect that isn't there) and a false negative (missing a real effect).
- Variance
- The average squared distance of values from the mean. Its square root is the standard deviation.
- Data visualization
- The graphical representation of data to reveal patterns and communicate findings. See the Chart Guide.
- z-score
- How many standard deviations a value lies from the mean. Also the test statistic in z-tests.
Missing a term? Suggest it and we'll add a definition.