Free DA0-002 practice test: 30 questions with answers

A t-test compares the means of two groups or periods to assess whether the difference is likely real. Chi-square tests relate to categorical counts, clustering finds groups, and word clouds do not test differences.

Two events occurring together does not prove one caused the other, and confounding factors must be considered. The other options do not describe the inferential error.

The median is resistant to extreme values, so it better represents typical income when a few values are very large. The maximum and range are driven by extremes, and the count does not describe the central value.

Association analysis, such as market basket analysis, finds items that frequently occur together. Forecasting uses time patterns, weighting adjusts surveys, and image classification works on pictures.

Executives need the conclusion first, backed by concise evidence. Leading with calculations, raw data, or methods hides the answer the audience needs.

Bar charts should start at zero because bar length encodes value, so a truncated axis misleads the reader. The other issues do not affect how the values are perceived.

Line charts show change over continuous time clearly. Pie charts with many slices are hard to read, maps show geography, and tables without trend cues hide the pattern.

Reconciling totals and peer review catch calculation and filter errors before decisions are made. Speed, colours, and removing sources reduce accountability rather than improving accuracy.

Aligned numbers with consistent decimals make comparisons faster and more accurate. Mixed alignment and fonts, and missing headers, make values harder to compare.

A descriptive title, labelled axes with units, and a source or date make a chart understandable and checkable. Decorative 3D effects distort values, and random colours without a legend confuse readers.

Part-to-whole comparisons suit a pie with few slices or a stacked bar. Scatter plots show relationships between two numeric variables, word clouds show text frequency, and maps show geography.

A positive relationship shows points trending upward as the x-value increases. Flat bands indicate no relationship, and a circle or single cluster does not describe a positive trend.

Relational databases store data in tables with defined columns and enforced schemas. Key-value stores, schema-free document stores, and graph databases use different structures.

JSON records carry their own field labels and nesting without a rigid table schema, which makes them semi-structured. Audio and images are unstructured, and fixed-width rows are structured.

Parquet is a columnar format that lets engines read only the needed columns and compresses well. Plain text, PNG images, and email files are not optimized for analytical column reads.

Managed cloud data warehouses scale storage and compute for SQL analytics without server management. Laptops, spreadsheets, paper, and USB drives cannot handle that volume efficiently.

Data lakes hold data in native formats, including structured, semi-structured, and unstructured data, and apply structure when the data is read. Warehouses usually hold cleaned, modeled data.

Assigning text to categories with labelled examples is supervised NLP classification. Robotic process automation, image compression, and routing are different tasks.

Lineage tracks the origin and transformations of data, so the source of each figure can be traced and audited. Encryption protects data, compression reduces size, and deduplication removes repeated records.

Masking and role-based access limit visibility to what each role needs, which protects privacy. Public or unrestricted sharing exposes the personal data.

Scheduled profiling and automated threshold tests reveal quality drift early. Removing alerts, assuming constant quality, and discarding failed-check records each hide problems.

A retention policy specifies how long records are kept and when they must be disposed of. Dashboard themes, chart standards, and palettes do not govern data lifecycle.

Version control and change records show who changed what and when, with the rationale. Deleting definitions, relying on memory, or ephemeral messages do not create an auditable history.

Encryption at rest makes stored data unreadable without the keys, even if the media are stolen. The other options do not protect stored data.

Joining on a shared key integrates the two sources into one analyzable dataset. Merging charts or discarding one source does not combine the records.

Identical records loaded twice are duplicates, which inflate totals and counts. Outliers are unusual values, type mismatches concern field formats, and integrity errors involve broken links between tables.

An extreme value may be a mistake or a real event, so verification comes before any treatment. Automatic removal or averaging without checking can distort the analysis.

Column summaries and checks for missing values and duplicates are standard exploration steps. Publishing raw data, ignoring types, and removing numeric columns are not exploration steps.

Selecting only needed columns and filtering in the query reduces transfer and processing. Exporting everything and then filtering, or using screenshots, wastes resources.

Standardization and matching rules align variant spellings so the same entity joins correctly. Keeping both values leads to split records, deleting data loses information, and unmapped numbers do not reconcile names.