EXIN Data Analytics Foundation cheat sheet

Analytics types, ascending in value: descriptive (what happened), diagnostic (why), predictive (what will happen), prescriptive (what action to take).

Data types: qualitative/categorical (nominal = no order, ordinal = ordered) versus quantitative/numerical (discrete = countable values, continuous = any value in a range).

Measurement scales: nominal, ordinal, interval (equal distances, no true zero, e.g. Celsius), ratio (true zero, ratios meaningful, e.g. revenue).

Population versus sample: a parameter describes a population; a statistic describes a sample and estimates the parameter.

Mean is sensitive to outliers; median is robust; mode is the only measure valid for nominal data.

Standard deviation is the square root of variance and is expressed in the same unit as the data; a z-score of (x − mean)/SD states distance in SD units.

Correlation coefficient r ranges from −1 to +1; r = 0 means no linear relationship, not necessarily independence.

Structured data fits rows and columns (relational tables); semi-structured uses tags/hierarchy (JSON, XML); unstructured lacks a predefined model (text, images, audio).

Common lifecycle order: define the problem and questions, collect data, clean/prepare, analyse, interpret, communicate, then act and monitor.

SMART criteria for a problem statement: Specific, Measurable, Achievable, Relevant, Time-bound.

Data cleaning tasks: handle missing values (drop, impute with mean/median/mode, flag), remove duplicates, fix formats and types, treat outliers, standardise units.

Missing-data mechanisms: MCAR (missing completely at random), MAR (explained by other observed variables), MNAR (depends on the missing value itself, e.g. high earners not reporting income).

Outlier detection rules: values beyond Q1 − 1.5×IQR or Q3 + 1.5×IQR (Tukey fence); z-score magnitude above 3 is a common cutoff.

Normalisation rescales to a fixed range such as 0–1; standardisation (z-score) centres on mean 0 with SD 1; both are needed for distance-based algorithms.

Sampling methods: simple random, systematic (every k-th item), stratified (random within homogeneous groups), cluster (whole groups selected).

Bias sources in the process: selection/sampling bias, survivorship bias, confirmation bias, and measurement/instrument bias.

An A/B test compares a treatment and control group; a statistically significant result requires checking effect size and practical relevance, not p-value alone.

Encoding hierarchy: position on a common scale is the most accurate visual channel, then length, angle/slope, area, then colour hue or volume.

Chart selection: bar for comparing categories, line for trends over time, scatter for relationships between two numeric variables, histogram for distribution of one numeric variable, box plot for spread and outliers.

Pie charts work only for part-to-whole with few slices (roughly 5 or fewer); stacked bars beat multiple pies for comparison.

Truncated y-axis exaggerates differences in bar charts; bars must start at zero, whereas line charts may use a non-zero baseline.

Dual-axis charts and 3-D effects invite misreading; prefer small multiples or indexed values instead.

Colour rules: sequential palette for ordered data, diverging palette for data with a meaningful midpoint, qualitative palette for unordered categories; avoid red/green-only encoding for accessibility.

Data-ink ratio (Tufte): maximise ink devoted to data; remove chartjunk, redundant gridlines and decoration.

Report structure: lead with the key message or recommendation, then supporting evidence, then method and caveats in an appendix.

Dashboard design: one question per dashboard, most important KPI top-left, consistent units and time windows, and a visible data-refresh timestamp.

Data governance defines decision rights and accountability for data: who may create, read, update, delete, and who owns quality.

Core roles: data owner (accountable for a domain), data steward (day-to-day quality and definitions), data custodian (technical storage and access), data user.

DAMA-DMBOK knowledge areas include data quality, metadata management, master and reference data, data security, and data architecture.

Data quality dimensions: completeness, accuracy, consistency, timeliness, uniqueness, validity - measured and reported as metrics, not opinions.

GDPR principles: lawfulness/fairness/transparency, purpose limitation, data minimisation, accuracy, storage limitation, integrity and confidentiality, accountability.

GDPR data-subject rights include access, rectification, erasure, restriction, portability, and objection to automated decision-making.

Personal data is any information relating to an identified or identifiable natural person; pseudonymisation reduces risk but does not make data anonymous.

Privacy by design and by default requires privacy protection built into systems and the strictest setting applied without user action.

Ethical safeguards: informed consent, purpose limitation, minimisation, anonymisation where possible, bias auditing of models, and documented provenance and lineage.