Study guide · Core notes
Reference notes for data stores and tools, acquisition and cleaning, statistical methods, visuals, and the controls that keep a report trustworthy.
How to read these notes: Items tagged Extra sit outside the DA0-002 objective bullets, such as the Type 1 and Type 2 names for slowly changing dimensions. Learn the unmarked items first. New to the exam? Start with the overview for the format, the 675 passing score, and the April 14, 2026 retirement of English DA0-001. A vendor pipeline in one cloud belongs on the AWS Data Engineer path.
A relational database stores tables linked by keys. A non-relational store does not require that fixed table model. Structured data is a table: a fact table holds measures, a dimensional table holds the labels you slice by, and a bridge table links a many-to-many pair such as students and courses. A slowly changing dimension records how a label changes over time. Semi-structured data, such as JSON, can nest fields inside fields. Unstructured data has no agreed columns.
The objectives name the idea. They do not print the usual type numbers. Extra Type 1 overwrites the old value and keeps only the current one. Type 2 keeps a new row so history remains. Overwriting a city on a customer dimension drops the old city from every past report that uses that row.
Strings include char, varchar, and nvarchar. Numeric values include integer, decimal, and float. A timestamp is a datetime. Null means the value is missing, not zero. Spatial data is location. Boolean is true or false. A BLOB is a binary large object, such as an image file stored in the database. A CLOB is a character large object, such as a long text. A GUID or UUID is an identifier meant to be unique.
| Extension | What you should expect |
|---|---|
| .csv and .txt | Text. A CSV separates fields, often with commas. A text file may have no agreed delimiter. |
| .xlsx | A spreadsheet workbook. It can hold more than one sheet and hidden formatting. |
| .json | Named fields, often nested. It is semi-structured. A comma-separated file does not use braces. |
| .jpg | An image. It is not a table of sales. Store or describe it as unstructured or as a large object. |
| .dat | A generic data file. Do not assume the delimiter until you open a sample. |
Data arrives from databases, APIs, website data, files, and logs. A repository is where an organization keeps a collection of that data. The objectives separate five repository ideas.
OLTP and OLAP are on the Data+ acronym list. OLTP records current transactions. OLAP answers analytical questions. Do not run a heavy scan on the checkout system.
The objectives name Amazon Web Services, Azure, and Google as cloud providers. A private cloud is for one organization. A public cloud is shared infrastructure offered to many customers. A hybrid design uses both. On-premises means the organization runs the systems itself.
| Type | How you use it |
|---|---|
| Object | Files stored as objects, usually with an identifier. Common for lakes and cloud buckets. |
| Block | Disks presented in blocks. Databases often sit on block storage. |
| File and shared | Folders and paths. Shared file storage is available to more than one system. Local storage stays on one machine. |
| Containerization | Packages an application and its dependencies so it runs the same way on different hosts. A container is not a data type and not a chart. |
Recognize the category. The exam does not ask you to click through a product menu.
| Job | Examples the objectives name |
|---|---|
| Coding environment | An IDE such as RStudio or VS Code, a text editor, or a notebook. |
| Business intelligence | Tableau, Power BI, and Looker. |
| Packages | Anaconda, pandas, and tidyverse. |
| Languages | Python, R, SAS, and Scala. |
| Database clients | SQL Server Management Studio, MySQL Workbench, MongoDB Compass, DBeaver, Toad, and Azure Data Studio. |
pandas and tidyverse shape tables in code. Tableau and Power BI build visuals. MongoDB Compass is a client for a non-relational database. SSMS and MySQL Workbench are clients for relational engines. Do not swap those jobs.
DA0-002 asks you to identify the idea, not to train a model.
Integration brings sources together. Collection methods the objectives name are surveying and sampling. A sample is a subset. A survey asks people. Neither one is a full copy of a warehouse.
| Operation | What it does |
|---|---|
| Join | Combines tables by a key. A bad key double-counts. |
| Union and concatenate | Stack compatible rows. They do not add columns the way a join does. |
| Filter | Keeps rows that match a condition. |
| Grouping and aggregate | Roll rows up, such as a sum by region. |
| Nested query | A query inside another query. Read the inner result before you blame the outer filter. |
ETL transforms, then loads. ELT loads the raw extract, then transforms it. Basic optimization uses an index to speed a matching read, a parameter to reuse one statement, a subset to shrink the working set, and a temporary table to hold an intermediate result. An index is not a backup, and it can slow writes.
Look for defects before you publish a number. Missing values are gaps. A duplicate is the same row twice. Redundant data is the same fact in more than one place. An outlier is an extreme value. Completeness asks whether required fields are present. Validation asks whether values follow the rule, such as a date that is a real date.
Imputation fills a missing value. Deletion removes a row or a field. Confirm an outlier before you delete it. A real wholesale order is valid data even when it is large.
| Technique | Use it when |
|---|---|
| Parsing and regular expressions | You need to split or match text, such as a full name or a product code. |
| Conversion | A field is the wrong type, such as a number stored as text. |
| Binning and clustering | You group values, such as ages into bands. |
| Scaling and standardization | Variables need a comparable scale. That is not the same as normalizing a dimension table into more tables. |
| Merging and appending | A merge brings columns together by a key. Append adds rows. |
| Exploding | A nested value, such as a list inside one cell, should become rows. |
| Augmentation | You add information from another source to a record you already have. |
| Derived or calculated field | The value is computed, such as margin. It is not a missing value you invented. |
Start from the audience. A mock-up shows the layout before you build it. Accessibility includes people who need the result in sound and people who need it in a visual form. A technical reader can take field definitions and filters. A non-technical reader needs the conclusion in plain language. An executive, the objectives say C-suite, usually wants a few KPIs. An individual contributor may need the working detail. Internal and external audiences do not get the same fields. Sensitive data stays off a public page.
Level of detail follows the decision. A KPI is the number that matches the business question. Extra columns that do not change the decision bury it.
| Method | The question it answers |
|---|---|
| Descriptive | What happened in this set? Center, spread, and counts belong here. |
| Inferential | What can you say about a larger group from this sample? |
| Predictive | What value is likely next? |
| Prescriptive | What action should someone take? |
Measures of central tendency include the mean, the median, and the mode. The mean is the average and moves when a few values are huge. The median is the middle of an ordered list. The mode is the value that appears most often. Measures of dispersion describe spread, such as the range from the smallest value to the largest. Logical functions test a condition. Date functions work with timestamps. String functions work with text. A percent change uses the original value as the base: from 80 to 100 is 25 percent, because the change is 20 and the base is 80.
Extra A t-test, chi-squared, a p-value, and correlation are common inferential tools the V2 bullets do not name. A small p-value is evidence against a null hypothesis. Correlation is an association, not a cause. Use those ideas only after you can name the four families above.
When a result looks wrong, name the failure before you redraw the chart. Connectivity problems mean the tool cannot reach the source. A user-reported issue is a complaint you still have to reproduce. Basic SQL problems include a bad join, a bad filter, or a type mismatch. Corrupted data is damaged content, not a slow network.
Useful checks are to enable logging, validate the data source, and consult vendor communities or other published references when the product itself is failing. Do not delete rows until you know which of those causes you have.
Charts compare values, maps place them, pivot tables summarize by categories you can turn, and infographics combine a few designed facts for a broad reader. Design elements are labels, legends, branding, and color schemes. A legend that uses brand colors still has to distinguish the series. Color alone is a weak encoding when the reader cannot see that difference.
| Visual | Use it for |
|---|---|
| Line | Values in time order. |
| Bar | Categories compared with one another. |
| Histogram | One continuous variable split into bins. It is not a bar chart of product names. |
| Map | Values by place. Word frequency is not a map. |
| Pivot table | Totals you can regroup by dragging fields. It is a table, not a geographic map. |
An executive summary is the short conclusion. A self-service portal lets people ask their own questions. A dashboard can be static or dynamic. Recurring delivery repeats on a schedule. Ad hoc delivery is one time. A snapshot freezes one moment. Real-time delivery moves with the source. Compare the refresh time and the run time when two totals disagree.
| Symptom | Look here first |
|---|---|
| Excessive load time or large data size | Filter to the needed subset. A picture of every row is not a dashboard. |
| Slow refresh or stale data | The source and the refresh interval. A new legend does not update old numbers. |
| Filter not working | The field type and the filter logic. Review the calculation and ask a peer to read it. |
| Corrupt data | Validate the source. A structure change upstream can break a field the report still expects. |
Other techniques the objectives name are code review, calculation review, peer review, source validation, and monitoring alerts. An alert tells you something drifted. It does not choose the chart.
Integration is how systems share data. Documentation the objectives name includes a data flow diagram, a data explainability report, a data dictionary, a hierarchy, and data lineage. A dictionary defines names and meanings. Lineage shows where a value came from and what changed it. A hierarchy is a tree, such as region, country, and city. An explainability report says why a result looks the way it does. The source of truth is the record you trust when two systems disagree. Data versioning uses snapshots and refresh intervals. Metadata is data about the data, such as when a file was loaded.
Master data management is on the acronym list. The trusted customer or product record is the practical idea. A merger is a reason to reconcile two masters. The dictionary is not that record.
Retention says how long you keep data. GDPR is a jurisdictional privacy rule. Other jurisdictions can add their own requirements. Replication copies data for resilience or distribution. Storage is where it sits. Data ethics covers uses that may be legal and still wrong to publish. PCI DSS applies to payment card data. An audit checks that the controls happened. Classification labels sensitivity. Incident reporting covers a data breach and other security events. Report the incident through the path the organization defined. Do not paste the exposed file into a chat to "show the problem."
Requirement testing checks that the data meets the stated need. A unit test checks one piece. User acceptance testing checks that the people who will use the result accept it. A stress test pushes volume or load. Source control tracks changes to the code or the definition. A data health check looks for data drift, which is a shift in the shape or meaning of the data over time. Automated data quality monitoring watches those checks continuously. Data profiling measures the set, including quality metrics such as how many values are missing. ISO is the standards body the objectives name beside these practices.
Accuracy means the values are correct. Completeness means required fields are present. A blank required field fails completeness even when the other fields are right. Check quality at the source, after you transform, and on the delivered report. The PDF alone hides where a country was dropped.
No. They are an original summary of the public DA0-002 objectives, document version 2.0. Confirm the current PDF with CompTIA before you book.
Extra marks a useful idea the DA0-002 objective bullets do not name, such as the Type 1 and Type 2 labels for slowly changing dimensions. Learn the unmarked items first.
The median. The mean is pulled by extreme values. Check an outlier before you delete it, because a real large order is still valid data.
Use the flashcards to recall these pairs, then test them on a timed exam.
Exam format, the 675 passing score, five DA0-002 domain weights, and how V2 differs from retired DA0-001.
Reference notes for sources, preparation, statistics, visuals, and governance.
40 recall cards in four decks, plus memory notes for stores, statistics, and reports.
200 original questions with custom exams, explanations, and an unofficial score on the 100-900 scale.
Useful companions while you study.
Calculate trigonometric functions, logarithms, powers, roots, factorials, and advanced expressions.
Convert length, weight, volume, temperature, area, speed, pressure, and other measurements.
Calculate elapsed time, add or subtract hours and minutes, and convert time to decimal hours.
Compare two text blocks and highlight added, removed, and changed content.
Certicab provides general informational, educational, and convenience resources. Calculations, conversions, estimates, and learning materials may contain errors or become outdated. Financial, tax, medical, legal, and travel information is not professional advice. Verify important results and current requirements with qualified professionals or authoritative sources. Protect sensitive files and personal information, review each tool’s privacy limitations, and use only content you are authorized to process. You are responsible for how you use and share results. Study resources are independent and do not guarantee exam success or imply certification-provider endorsement. CompTIA and Data+ are trademarks of CompTIA, Inc. Certicab is not affiliated with, endorsed by, or sponsored by CompTIA.
Certi