EXIN Data Analytics Foundation common mistakes
EXIN Data Analytics Foundation · Understand the traps and how to avoid them.
Treating descriptive, diagnostic, predictive and prescriptive analytics as a ladder where each type replaces the previous one, so a question describing a dashboard that reports last month's sales is answered as 'predictive' because it sounds more advanced.
Candidates memorise the four types as a ranked list and assume exam questions reward the highest-sounding answer, rather than matching the type to the question the analysis actually answers (what happened, why, what will happen, what should be done).
What to do instead
For every scenario, name the question being answered before naming the type. Build a four-row table (type, question answered, typical technique, example output) and drill on short scenarios where the verb in the scenario ('reported', 'explained', 'forecast', 'recommended') is the deciding clue.
Confusing data, information, knowledge and wisdom in classification items, for example labelling an aggregated monthly average as 'raw data' or calling a single sensor reading 'information'.
The DIKW hierarchy is taught early and skimmed as common sense, so candidates rely on intuition about which word sounds right instead of applying the processing criterion that separates each level.
What to do instead
Apply a one-question test: has context or processing been added, and does it support a decision? Raw data = unprocessed observations; information = processed and contextualised; knowledge = patterns and experience applied; wisdom = judgement about action. Practise with ten examples from your own work and classify each.
Jumping from a business question straight to a modelling or visualisation technique, skipping problem definition, data requirements and data-quality checks, and then choosing the 'explain' option in process-ordering questions.
Practitioners are used to receiving a dataset and starting work immediately, so the early CRISP-DM-style phases feel like paperwork and are under-weighted when reading answer options.
What to do instead
Memorise the phase sequence as a chain (business understanding, data understanding, data preparation, modelling, evaluation, deployment) and note that evaluation and business understanding are revisited, not strictly linear. For ordering questions, anchor on the phase that produces the input the next phase needs.
Treating data cleaning as a single step and failing to distinguish missing values, outliers, duplicates and inconsistent formats, so a question about a value of 999 in an age column is answered as 'missing data' or 'remove the row' without considering whether it is a sentinel code or a genuine outlier.
Real cleaning is iterative and messy, while exam items present one symptom at a time; candidates pattern-match on the word 'outlier' or 'missing' rather than inspecting what the value represents.
What to do instead
Learn the standard treatments per problem: sentinel codes recoded to null, outliers investigated before removal (winsorising, capping, or exclusion with justification), duplicates deduplicated on a defined key, formats standardised before aggregation. For each scenario, state what the value probably means before choosing a treatment.
Confusing correlation with causation in interpretation items, or accepting a strong correlation as proof that one variable drives another, especially when a third variable such as seasonality or population size could explain both.
Correlation coefficients are visually compelling and the exam's distractors are written to sound decisive; candidates also conflate statistical significance with practical importance.
What to do instead
Apply three checks before accepting a causal claim: was the study controlled or observational, is there a plausible mechanism, and could a confounder explain both variables? Also separate significance (p-value, sample-size dependent) from effect size (magnitude, business relevance) and practise stating both for a given result.
Selecting a chart type by appearance rather than by data type and purpose, for example using a pie chart for ten categories, a line chart for unordered categories, or a 3D chart to compare values.
Default chart options in spreadsheet tools encourage decorative choices, and candidates do not rehearse the mapping from variable type (categorical, ordinal, interval, ratio) and analytical intent (comparison, distribution, trend, relationship, composition) to chart.
What to do instead
Build a decision table: comparison across categories = bar; trend over time = line; distribution of one variable = histogram or box plot; relationship between two numeric variables = scatter; composition of a whole with few parts = stacked bar or pie only for two to four slices. Reject 3D and decorative effects on the grounds that they distort perceived values.
Ignoring the perceptual pitfalls of axis manipulation, truncated y-axes, dual axes and colour scales, and judging a visualisation as acceptable because the numbers are correct even when the scale misleads.
Candidates focus on data accuracy and treat presentation choices as cosmetic, so they miss items that ask what is misleading about a described chart rather than whether it is factually wrong.
What to do instead
Learn the standard misleading patterns: truncated bar axes exaggerating differences, dual y-axes implying a relationship, inconsistent bin widths, rainbow scales for ordered data, and area or volume encoding that overstates differences. For each described chart, ask what a viewer would conclude versus what the data supports.
Treating data governance as purely an IT or legal concern, so questions about data ownership, stewardship, definitions and master data are answered with a technical tool or a security control instead of an accountability structure.
Governance is abstract and rarely visible in day-to-day analysis work, so candidates substitute the more familiar concepts of access control and backup for the organisational roles and policies that governance actually covers.
What to do instead
Learn governance as four concrete artefacts: roles (owner accountable, steward responsible for quality and definitions), policies and standards, a business glossary and data dictionary, and quality metrics with escalation. For each scenario, ask who is accountable for the data and where its definition is documented before considering tools.
Confusing the legal bases and principles of data protection, for example treating consent as always required, or assuming anonymised and pseudonymised data carry the same obligations.
Privacy rules are learned from headlines rather than from principles, and pseudonymisation is easily mistaken for anonymisation because both look like removing names.
What to do instead
Learn the distinction precisely: pseudonymised data can be re-linked with additional information and remains personal data; anonymised data cannot be re-identified and falls outside personal-data rules. Learn that consent is one of several legal bases, and that purpose limitation, data minimisation, storage limitation and accuracy are principles that apply regardless of basis.
Answering ethics items with the legal minimum, choosing 'it is allowed because it is not prohibited' over options that address bias, transparency, fairness and accountability.
Candidates collapse ethics into compliance and assume the exam rewards the permissive answer, whereas ethics items typically test whether the analyst considers impact on affected people and the ability to explain a decision.
What to do instead
For each ethics scenario, run four checks: who is affected, could the analysis disadvantage a group, can the result be explained to those affected, and who is accountable for the outcome. Prefer the option that adds transparency and review over the one that merely stays within the law.