CompTIA Data+ DA0-002 · Free study guide

Domain 1 — Data Concepts and Environments (20%)

Domain 1 is the vocabulary layer of the course. The other four domains assume you can classify data, name the storage architecture in a scenario, and recognize the modern analytics stack. This supplemental review gives context; use the five objective lessons for the complete current V2 checklist.

Classifying data: structure

Every piece of data falls somewhere on a structure spectrum. Structured data lives in fixed rows and columns with a schema enforced before the data arrives — the relational table is the canonical case. Semi-structured data carries its own organizational markers without a rigid tabular schema: JSON and XML are common examples, with their nested objects, optional fields, and self-describing keys. Unstructured data has no inherent organization the machine can rely on — free-text reviews, images, audio, PDFs.

A useful recognition scenario is a feed of JSON files with nested, optional fields is semi-structured, full stop. Do not be baited by the word "consistent" in a stem — consistency of convention is not the same as a fixed tabular schema.

Classifying data: measurement

As supplemental statistical context, measurement scales describe what values mean:

• Nominal — named categories with no order (product category, region).
• Ordinal — ordered categories with no measurable distance between them (satisfaction scales from Very Dissatisfied to Very Satisfied). Coding the labels 1–5 does not make them interval data; the gap between "Neutral" and "Satisfied" is not a quantity.
• Interval — equal spacing, no true zero (temperature in Celsius).
• Ratio — equal spacing plus a true zero (revenue, weight, duration).

Cross-cutting this: discrete variables take countable values (number of support tickets — nobody opens 2.4 tickets) while continuous variables take any value in a range (weight, load time, temperature). And in BI tools, numeric values you aggregate are measures while the descriptive attributes you slice them by are dimensions — revenue is a measure; region and product line are dimensions.

File formats

CSV is a common delimited text interchange format — simple and portable, but sensitive to delimiter, quoting, encoding, and type-inference mistakes. XLSX can preserve spreadsheet structure, formulas, formatting, and multiple sheets. JSON carries nested semi-structured payloads. Plain-text files are appropriate for logs and prose; image files carry visual data; and a .dat extension is a generic container whose actual encoding must be documented or inspected rather than guessed from the suffix.

As supplemental infrastructure context, Parquet and other columnar formats store column values together. For large analytical reads that touch only a few columns, that layout can reduce I/O and improve compression.

Databases and platforms

OLTP versus OLAP is a foundational workload split. Online transaction processing systems handle many small, fast reads and writes — the checkout system recording each sale. Online analytical processing systems serve big aggregate scans over history — the warehouse the analysts query for monthly patterns. Same business data, opposite designs.

On the relational side, know your keys: a primary key uniquely identifies each row; a foreign key is a column referencing another table's primary key, and it is what makes a join meaningful. On the non-relational side, match the store to the access pattern: a document database accepts flexible JSON-style records; a key-value store retrieves values through unique keys and can suit caches or simple state lookups; graph databases specialize in relationship traversal; and wide-column stores can distribute sparse records across clusters. Actual latency and durability depend on the implementation and deployment rather than the database label alone.

Warehouses, lakes, and lakehouses

A data warehouse commonly holds modeled, curated reporting tables with governed schemas. A data lake commonly retains files such as logs, images, and exports in object storage and applies structure for a particular use. A lakehouse layers managed table formats, transactions, schema controls, and analytical query features over lake-style storage. These are architectural patterns, not guarantees about the quality of every implementation.

Inside a warehouse, dimensional modeling: a star schema puts measures in a central fact table with foreign keys out to denormalized dimension tables (date, customer, product). A snowflake schema normalizes those dimensions into hierarchies — product → category → department — trading storage redundancy for extra joins. Snowflaking can reduce redundancy while increasing query and model complexity.

Metadata — data about data — shows up as the data catalog question: column names, types, refresh schedules, owners, and lineage are metadata, distinct from master data (the authoritative records of business entities, a Domain 5 concern).

Infrastructure choices

DA0-002 names services from AWS, Microsoft Azure, and Google Cloud, but the durable skill is matching requirements to infrastructure. A public cloud uses provider-operated shared infrastructure; a private environment is dedicated to one organization; and a hybrid design connects cloud and on-premises systems when requirements span both.

Storage choices differ too. Object storage suits large collections of files and API-based access. File storage exposes shared hierarchical paths. Block storage presents raw volumes to a host or database. Local storage is tied to one machine, while shared storage supports access from multiple systems. Evaluate latency, durability, capacity, access pattern, portability, security, and cost instead of choosing by label alone.

Containers package an application and its dependencies into a repeatable runtime image. They improve portability and reproducibility, but persistent data, secrets, networking, scaling, and access controls still require explicit design.

Analysis tools

An IDE provides project navigation, debugging, and integrated development features; a text editor is lighter; and a notebook interleaves executable code, results, and narrative exploration. Tableau, Power BI, and Looker are business intelligence platforms. Anaconda distributes Python and data-science packages; pandas is a Python data-manipulation library, while tidyverse is an R package ecosystem. SAS is an analytics platform and language, and Python, R, Scala, and SQL serve different programming and query workloads. A database client such as SQL Server Management Studio connects to and administers a database; it is not the database engine itself.

AI-assisted analytics

DA0-002's headline addition. Two things to hold at once. First, the capabilities are real: natural language querying lets a user type "average order value by month" and get a chart without query syntax, and generative assistants can draft SQL, summarize datasets, and explain anomalies. Second, the discipline is non-negotiable: generative models produce plausible, fluent, and sometimes wrong output. Any specific figure an assistant produces gets verified against the underlying data before it enters a report — and asking the model to confirm itself is not independent verification.

What to drill

Use the five objective lessons as the checklist. Practice matching database models, current file formats and data types, repositories, cloud deployment and storage models, containers, analysis tools, and AI concepts to explicit requirements. For every choice, state the trade-off and the evidence that would verify it in a real environment.