IIBA-CBDA cheat sheet

A research question defines the specific, answerable inquiry; a hypothesis states a testable prediction about a relationship or difference.

SMART objectives are Specific, Measurable, Achievable, Relevant, and Time-bound; each objective should include a metric, target, and deadline.

Scope boundaries should state the population, time period, geography, systems, and exclusions to prevent analysis creep.

Descriptive questions ask what happened; exploratory questions identify patterns; explanatory questions test causes; predictive questions forecast outcomes.

Root cause techniques include the Five Whys, cause-and-effect (fishbone) diagrams, and Pareto analysis of contributed causes.

Assumptions are believed but unverified conditions; constraints are fixed limitations such as budget, legal rules, or data access.

A measurable success criterion combines a baseline value, a target value, and a timeframe so results can be judged objectively.

Stakeholder influence-interest analysis identifies who must approve, who must be consulted, and who must be informed.

Primary data is collected for the current question; secondary data was collected earlier or for another purpose and requires fitness checks.

Data quality dimensions commonly include completeness, accuracy, consistency, timeliness, validity, and uniqueness.

Probability sampling methods include simple random, stratified, cluster, and systematic sampling; non-probability methods include convenience, purposive, and quota sampling.

Larger sample sizes reduce sampling error and increase precision, but do not correct bias from a non-representative selection process.

Metadata describes definitions, owners, formats, and refresh dates; data lineage documents the path from source through transformations to reporting.

Data minimization collects only the personal data needed for the stated purpose; consent, retention limits, and access controls govern use.

Instrument validity means the measure captures the intended concept; reliability means repeated measurements produce consistent results.

Data profiling checks null rates, duplicate rates, value ranges, distributions, and referential integrity before analysis.

Batch extraction refreshes data at scheduled intervals; streaming extraction updates continuously and requires different latency and quality controls.

Mean is the arithmetic average and is sensitive to outliers; median is the middle ordered value and is robust to extreme values.

Standard deviation measures average dispersion around the mean; interquartile range measures spread of the middle 50% of values.

A correlation coefficient ranges from -1 to +1; its sign gives direction and its magnitude gives strength of linear association, not causation.

Hypothesis testing compares a null hypothesis with an alternative using a significance level alpha; a p-value below alpha leads to rejecting the null.

A confidence interval is estimate plus or minus critical value times standard error; wider intervals indicate greater uncertainty.

Type I error is rejecting a true null hypothesis; Type II error is failing to reject a false null hypothesis.

Regression coefficients estimate change in the outcome per one-unit change in a predictor, holding other included predictors constant.

Missing data mechanisms are MCAR, MAR, and MNAR; the mechanism determines whether deletion or imputation may introduce bias.

Outliers should be investigated before removal because they may be data errors, rare valid events, or important business signals.

Statistical significance indicates an unlikely result under the null; practical significance compares effect size with business thresholds and costs.

Report confidence intervals and sample limitations so decision-makers understand the uncertainty around estimates.

Use line charts for trends over time, bar charts for category comparison, and scatter plots for relationships between two numeric variables.

Avoid causal language when the study design is observational; confounding and reverse causation may explain associations.

A report structure that supports decisions states the question, method, key findings, limitations, and recommendation in that order.

Tailor detail and vocabulary to the audience; executives need conclusions and implications, while technical reviewers need methods and caveats.

Charts should use proportional scales, clear labels, and non-truncated axes unless truncation is explicitly disclosed and justified.

Triangulation compares findings across multiple data sources or methods to assess whether conclusions are consistent and credible.

A weighted decision matrix assigns criteria weights, scores each alternative, and computes weighted totals to compare options transparently.

Cost-benefit analysis compares total expected costs with total expected benefits; net benefit equals benefits minus costs.

Return on investment equals net benefit divided by cost, often expressed as a percentage for comparing alternatives.

Expected monetary value equals the sum of probability times impact across possible outcomes; it supports comparison under uncertainty.

Sensitivity analysis tests whether a recommendation holds when key assumptions or inputs vary across plausible ranges.

Risk tolerance affects choice: a risk-averse decision-maker may prefer a lower expected value with less variability.

An implementation plan specifies the action, owner, resources, timeline, dependencies, and expected impact.

A KPI monitoring plan defines metric definitions, baselines, targets, data owners, review frequency, and corrective-action triggers.

Post-decision feedback loops compare actual outcomes with predicted outcomes and update assumptions for future decisions.