Free IIBA-CBDA practice test: 30 questions with answers

B is a specific, measurable question that names the dimensions to examine and the time frame, so it can guide data gathering. A is too broad to act on, C assumes a solution before any analysis, and D compares against competitors whose data the operations team is unlikely to have.

A well-formed research question is measurable and tied to the decision it will inform. Sponsor agreement is approval rather than evidence, tool choice is a method decision rather than part of the question, and open-ended exploration lacks focus.

The business objective sets the purpose of the analysis, and agreed success measures define what counts as an answer. Chart templates, vendor licences and slide counts are presentation or procurement details that do not determine what data is needed.

The main concern is selection bias: people who join a program often differ from those who do not, so any gap may not be caused by the program. The other options describe requirements that do not exist, since storage type, older records and equal group sizes are not prerequisites for a comparison.

A hypothesis is a testable statement that evidence can confirm or reject. It does not guarantee a result, it complements rather than replaces the research question, and it is not an inventory of variables.

Correlation shows that two variables move together but not that one causes the other, and a common driver such as weather often explains both. The other options assert causation in the wrong direction or wrongly claim that correlation proves independence.

Statistical significance indicates that a difference is unlikely to be chance, but it does not show that the difference is large enough to matter for the business. The other options overstate what significance proves about revenue, completeness, or sample size.

Consistent scales and labeled axes let readers compare values accurately, and line charts show change over time clearly. 3D pie charts distort proportions, unlabeled color schemes confuse, and undisclosed axis truncation exaggerates differences.

Executives need the answer and the decision it supports first, with detail available in supporting sections. Cleansing steps, full output tables and tool names belong in the appendix or methods section, not the opening summary.

When fraud is rare, a model that rarely flags fraud can still score high accuracy, so the analyst should check measures focused on the rare class. Retraining on volume growth, rerunning with fewer variables, and blanket confidentiality claims do not address the risk of misleading accuracy.

Random assignment to two versions creates comparable groups, so differences in conversion can be attributed to the design. A survey, forecast, or cleansing step does not isolate the design's effect on behaviour.

A pilot tests assumptions and shows real impact at limited cost and risk before a full rollout. It does not remove the need for approval, guarantee profit, or replace a business case.

Framing options and trade-offs around the manager's goals helps the findings influence the decision. Sending raw appendices, adding jargon, or withholding results makes the analysis harder to use.

Stated assumptions and limitations let decision makers weigh the evidence appropriately. Disclosure does not excuse later review, approval, or independent checking of the data.

Analysts must be transparent about known limitations that could change the conclusion, even when the sponsor prefers the result. Writing it up without disclosure, deleting the record, or disguising the gap all compromise the decision.

Predictive analytics uses historical patterns to estimate what is likely to happen next. Descriptive analytics summarizes the past, diagnostic analytics explains causes, and prescriptive analytics recommends actions.

A model that scores well on training data but poorly on new cases is overfitted and has learned noise rather than general patterns. The other options do not explain a drop in accuracy on new data, since the statistic type, storage choice and dashboards do not cause it.

Testing on data the model has not seen estimates how it will perform on new cases. Scoring on training records gives optimistic results, adding variables to maximize training accuracy tends to increase overfitting, and removing outliers from the test set biases the evaluation.

The right response is to find out why the values are extreme before deciding how to handle them. Deleting, replacing or hiding them without investigation can remove real business activity or hide data errors.

With a skewed distribution, a few very large values pull the mean upward, so the median better represents a typical customer. The range describes spread rather than typical values, and the mode is not always the most accurate average.

Aligning timestamps to one time zone and documenting the rule keeps the joined data accurate and traceable. Deleting records loses valid data, avoiding the join prevents the failure analysis from using both sources, and waiting on the vendor is unnecessary.

A data warehouse holds structured, processed data for reporting and analysis, while a data lake stores large volumes of raw data in native formats. The reversed description, the claim that only text fits in a lake, and the hosting claim are all inaccurate.

Metadata and data dictionaries show what each field means, and knowing how and when data was collected reveals its limits and bias. Assuming accuracy from ownership, choosing sources for visual appeal, or deleting blank records without review are poor practices.

The cause of missing data determines whether it is safe to impute, remove, or fix the collection process. Filling values with a common code or deleting records without checks can bias results, and dropping the field is premature without understanding the cause.

Random sampling gives each member a known chance of selection, which supports general conclusions about the population. Convenience, judgmental, and volunteer samples introduce selection bias, so their results may not represent the population.

A centre of excellence provides shared standards, practices, and support across the organization. Isolated analysts, full outsourcing, and temporary project teams do not build a lasting organization-level capability.

Shared, owned definitions for key metrics are a prerequisite for consistent, data-driven decisions across departments. Tools, dashboards, or separate definitions built before governance tend to produce conflicting answers.

Ranking by strategic value, expected benefit, and readiness directs investment where it will matter and where it can succeed. Equal funding, technology-first choices, and avoiding all data risk do not link investment to strategic outcomes.

An analytics strategy connects initiatives to business objectives and builds the skills needed to use the results. Spreadsheet inventories can be useful operationally but are not strategy, while fixed chart rules and replacing analysts are not sound strategic elements.

Centralized teams create consistency and shared standards but can be slower to respond to individual units, whereas decentralized analysts sit closer to the business. Reversing these strengths or claiming that either model removes governance and analysts is incorrect.