All certifications / Data+ / Cheat sheet

Data+ DA0-002 cheat sheet

Every exam tip and key term from the free Data+ lessons, by domain. Use your browser's Print to save it as a PDF.

Domain 1: Data concepts and environments (18%)

Exam tips

JSON and XML are the go-to examples of semi-structured data. If a question mentions nested fields, optional keys or tags, choose semi-structured; if it mentions photos, recordings or free text, choose unstructured.
If a question asks which column links two tables, the answer is the foreign key in the 'many' table. If it asks what removes repeated customer details from every order row, the answer is normalization.
Match the keyword to the type: flexible JSON records means document; lookup by ID only means key-value; massive time-series writes means column-family; relationships and hops means graph.
Numbers you never do math on (ZIP codes, phone numbers, IDs) should be text. For money, choose decimal over float.
Columnar plus compressed plus analytics means Parquet. Nested data from an API means JSON. Plain, universal, typeless tabular exchange means CSV.
Schema on write means warehouse; schema on read means lake. Transactions mean OLTP; analysis means OLAP.
Measures go in facts; descriptions go in dimensions. Keep history means SCD Type 2; overwrite means Type 1. Dimensions split into sub-tables means snowflake.
Reproducible code plus narrative means notebook; interactive dashboards for business users means BI platform; elastic, pay-as-you-go capacity means cloud.
Querying and aggregating data where it lives points to SQL; automation and machine learning point to Python; heavy statistical analysis points to R. Expect to read short snippets and say what they do.
Rule-based UI automation is RPA, not machine learning. Predicting a known label is supervised learning; finding groups with no labels is unsupervised. LLM output must always be validated.

Key terms

Structured data: Data that follows a fixed schema of fields and types, such as rows in a relational table.
Semi-structured data: Self-describing data with keys or tags but a flexible shape, such as JSON or XML.
Unstructured data: Data with no predefined model that queries can use, such as free text, images, audio and video.
Schema: The definition of the fields, types and relationships that data is expected to follow.
Primary key: A column or set of columns whose values uniquely identify each row and are never null.
Foreign key: A column that refers to another table's primary key, linking the tables and enforcing referential integrity.
Normalization: Organizing tables into normal forms (1NF, 2NF, 3NF) so each fact is stored once, reducing redundancy and update anomalies.
Junction table: A table holding pairs of foreign keys that resolves a many-to-many relationship.
Document store: A NoSQL database that stores self-contained, often nested JSON-like documents with flexible fields.
Key-value store: A NoSQL database that stores and retrieves values by a unique key, optimized for fast lookups.
Column-family store: A distributed NoSQL database whose rows hold varying sets of columns grouped into families, suited to heavy write loads.
Graph database: A database that stores nodes and relationships as first-class data, optimized for traversing connections.
String: A text data type that stores characters; used for names and for numeric-looking identifiers such as ZIP codes.
Decimal (numeric): An exact numeric type with fixed precision and scale, suited to currency.
Floating point: An approximate numeric type stored in binary that can introduce small rounding errors.
Casting: Explicitly converting a value from one data type to another, such as text to date.
CSV: Comma-separated values: a plain-text flat file with one record per line and comma-delimited fields.
JSON: A text format of key-value pairs and arrays that supports nesting; common in APIs.
Parquet: A compressed, columnar binary file format with stored types, built for analytic queries.
Delimiter: The character that separates fields in a flat file, such as a comma, tab or pipe.
OLTP: Online transaction processing: systems optimized for many fast, small reads and writes that run the business.
OLAP: Online analytical processing: systems optimized for complex, read-heavy analytic queries over historical data.
Data warehouse: A central repository of integrated, cleaned, structured historical data modeled for reporting (schema on write).
Data lake: A repository of raw data in native formats where structure is applied when read (schema on read).
Data mart: A subject-specific subset of warehouse-style data for one team or function.
Fact table: A table of numeric measures for business events at a defined grain, with foreign keys to dimensions.
Dimension table: A table of descriptive attributes (such as product, customer or date) used to filter and group facts.
Star schema: A design with a central fact table joined directly to denormalized dimension tables.
Snowflake schema: A star schema whose dimensions are normalized into related sub-tables.
Slowly changing dimension (SCD): A method for handling changes to dimension attributes; Type 1 overwrites, Type 2 adds a new row to keep history.
On-premises: Infrastructure owned and operated in the organization's own facilities.
Cloud: Computing and storage rented on demand from a provider, often as managed services billed by use.
Notebook: An interactive document that combines runnable code, output and narrative text, such as Jupyter.
BI platform: Software that connects to data, models it and publishes interactive dashboards and reports.
SQL: A declarative language for querying and managing data in relational databases.
pandas: A Python library for working with tabular data in DataFrames.
R: A programming language and environment designed for statistical computing and graphics.
Declarative language: A language where you state the result you want rather than the steps to produce it.
Supervised learning: Machine learning trained on labeled examples to predict a label or value for new data.
Large language model (LLM): A foundation model trained on large amounts of text that generates and interprets language.
Natural language processing (NLP): Techniques that let computers analyze and generate human language, such as sentiment analysis.
Robotic process automation (RPA): Software bots that repeat rule-based user-interface tasks, such as data entry, without learning.

Domain 2: Data acquisition and preparation (22%)

Exam tips

If an API exists, it is almost always the preferred answer over scraping. For surveys, look for the sampling method that makes sure every subgroup is represented: stratified.
Transform before load is ETL; transform inside the warehouse after load is ELT. Seconds-level needs mean streaming. Only changed rows means incremental.
Condition on a single row means WHERE; condition on a total, count or average means HAVING. NULL comparisons need IS NULL, never = NULL.
'Include records even with no match' means an outer join (usually LEFT). 'Stack files with the same columns' means UNION or append. If a join increases the row count unexpectedly, suspect duplicate keys.
Know the fix that pairs with each problem: deduplicate duplicates, standardize inconsistent formats, validate invalid values, investigate outliers before removing them, and impute or flag missing values.
For skewed numeric data such as income or house prices, median imputation beats mean imputation. Replacing blanks with zero is almost always wrong unless zero is the true value.
Split one field into several means parsing; join several into one means concatenation; create a new calculated column means a derived variable.
0-to-1 range means min-max normalization; mean 0 and standard deviation 1 means standardization. Don't average averages; aggregate from the underlying totals.
Columns-to-rows is unpivot; rows-to-columns is pivot. BI tools and databases generally want long format.
Index the columns you filter and join on, filter as early as possible, and name only the columns you need. Filtering after data reaches the BI tool is the slow answer.

Key terms

API: An interface that lets programs request data from a system in a documented, structured way, often over HTTPS returning JSON.
Web scraping: Extracting data by downloading and parsing web pages, used when no API is available.
Stratified sampling: Dividing a population into groups and randomly sampling within each group.
Sampling bias: Systematic error when a sample does not represent the population it is meant to describe.
ETL: Extract, transform, load: data is transformed in a separate layer before loading into the target.
ELT: Extract, load, transform: raw data is loaded first and transformed inside the target system.
Incremental load: A load that processes only new or changed records since the previous run.
Change data capture (CDC): A technique that identifies and delivers changes made in a source database, often from its transaction log.
WHERE: A clause that filters individual rows before grouping.
GROUP BY: A clause that combines rows with the same values into groups for aggregation.
HAVING: A clause that filters groups after aggregation, allowing conditions on aggregate functions.
Aggregate function: A function such as COUNT, SUM, AVG, MIN or MAX that summarizes many rows into one value.
Inner join: Returns only rows with matching keys in both tables.
Left join: Returns all rows from the left table plus matches from the right, with NULLs where there is no match.
UNION ALL: Stacks rows from two result sets with the same columns, keeping duplicates; UNION removes them.
Fan-out: Row duplication caused by joining to a table with repeated key values, which inflates totals.
Duplicate record: A record that represents the same entity or event as another record in the dataset.
Invalid value: A value that breaks a field's rules for type, range, format or allowed values.
Outlier: A value that lies far from most other values; it may be an error or a real extreme.
Redundancy: Storing the same data in multiple places, which risks inconsistency between copies.
Listwise deletion: Removing every record that has a missing value in any field used by the analysis.
Imputation: Replacing missing values with estimated values such as the mean, median, mode or a model prediction.
Missing at random: When the chance that a value is missing is unrelated to the missing value itself, so remaining data stays representative.
Imputation flag: An indicator column that marks which values were filled in rather than observed.
Parsing: Extracting meaningful parts from a structured string or document, such as splitting a name at a comma.
Concatenation: Joining two or more values into one string.
Recoding: Mapping existing values to new, standardized or grouped values.
Derived variable: A new field calculated from existing fields, such as profit margin or days to ship.
Min-max normalization: Rescaling values to a fixed range, usually 0 to 1, using the minimum and maximum.
Standardization: Converting values to z-scores with mean 0 and standard deviation 1.
Binning: Grouping continuous values into ranges or categories.
Aggregation: Summarizing detailed rows into totals or statistics at a higher level, such as monthly by region.
Wide format: A layout with one row per subject and repeated measures spread across columns.
Long (tidy) format: A layout with one row per observation, using attribute and value columns.
Pivot: Reshaping long data to wide, turning row values into column headers with aggregated cells.
Unpivot (melt): Reshaping wide data to long, turning column headers into values of a new attribute column.
Index: A data structure that speeds up finding rows by a column's values, at the cost of storage and slower writes.
Execution plan: The database's description of how it will run a query, including scans, joins and index use.
Temporary table: A table that holds an intermediate result for the current session so later steps can reuse it.
Common table expression (CTE): A named subquery defined with WITH that makes complex queries easier to read.

Domain 3: Data analysis (24%)

Exam tips

Skewed data or outliers mean choose the median. Categorical data means the mode. Right skew means mean greater than median.
Memorize the IQR outlier fences: Q1 − 1.5 × IQR and Q3 + 1.5 × IQR. Sample standard deviation divides by n − 1; population divides by N.
Know 68-95-99.7 and z = (x − mean) ÷ SD cold. A two-peaked histogram suggests two populations that should be analyzed separately.
Percent change always divides by the old value. Distinguish percentage points from percent. Compare groups of different sizes with ratios or rates, not raw totals.
p < alpha means reject the null hypothesis; otherwise fail to reject it, never 'accept' or 'prove' it. Larger samples mean narrower confidence intervals.
False positive is Type I (alpha); false negative is Type II (beta). Bigger samples raise power (fewer Type II errors) at the same alpha; the Type I error rate is set by alpha, not by sample size.
r is between −1 and 1, and the sign gives direction. Correlation is never proof of causation on the exam. Plug numbers into the regression equation carefully: multiply the slope by x, then add the intercept.
Map the question word: what happened is descriptive, why is diagnostic, what will happen is predictive, what should we do is prescriptive. Open-ended profiling of new data is exploratory.
A KPI needs a measurable definition and a target. Variance percentage divides by the plan. Know leading versus lagging with an example of each.
GROUP BY collapses rows; window functions keep every row. Running totals, rankings and prior-period comparisons point to window functions.
If a total is exactly double or suspiciously high after a join, suspect fan-out from duplicate keys. If records vanished, suspect an inner join or a leftover filter.

Key terms

Mean: The sum of values divided by their count; sensitive to outliers.
Median: The middle value of sorted data; robust to outliers and skew.
Mode: The most frequent value; the only central measure for categorical data.
Right skew: A distribution with a long tail of high values, where the mean is usually greater than the median.
Standard deviation: The square root of the variance; the typical distance of values from the mean, in the data's own units.
Variance: The average of squared deviations from the mean.
Interquartile range (IQR): Q3 minus Q1: the spread of the middle 50% of the data.
Percentile: The value below which a stated percentage of observations fall.
Normal distribution: A symmetric bell-shaped distribution where mean, median and mode are equal.
Empirical rule: In a normal distribution, about 68%, 95% and 99.7% of values fall within 1, 2 and 3 standard deviations of the mean.
Z-score: The number of standard deviations a value lies from the mean: (x − mean) ÷ SD.
Bimodal distribution: A distribution with two peaks, often indicating two mixed groups.
Frequency distribution: A table or chart showing how often each value or category occurs.
Percent change: (new − old) ÷ old × 100: the relative change from the earlier value.
Percentage point: The arithmetic difference between two percentages, such as 4% to 5% being 1 point.
Ratio: A comparison of two quantities, such as agents per 1,000 customers.
Population vs sample: The whole group of interest versus the subset actually measured to estimate it.
Confidence interval: A range, computed from a sample, that is likely to contain the true population value at a stated confidence level.
Null hypothesis: The default claim of no effect or no difference that a test tries to find evidence against.
p-value: The probability of results at least as extreme as observed, assuming the null hypothesis is true.
Type I error: A false positive: rejecting a null hypothesis that is actually true; its probability is alpha.
Type II error: A false negative: failing to reject a null hypothesis that is actually false; its probability is beta.
Statistical power: The probability (1 − beta) that a test detects an effect that really exists.
Practical significance: Whether an effect is large enough to matter in the real world, separate from its p-value.
Correlation coefficient (r): A value from −1 to +1 that measures the strength and direction of a linear relationship.
Confounding variable: A third variable that influences both variables being studied, creating a misleading association.
Slope: In regression, the average change in y for a one-unit increase in x.
R-squared: The proportion of variation in the dependent variable explained by the model.
Exploratory data analysis: Initial, open-ended investigation of data to understand its structure, quality and patterns.
Diagnostic analysis: Analysis that explains why an outcome occurred, often by drilling down and segmenting.
Predictive analysis: Analysis that uses historical data to forecast future outcomes or probabilities.
Prescriptive analysis: Analysis that recommends the best action, often using optimization or simulation.
KPI: A key performance indicator: a metric tied to a strategic objective, with a defined target.
Leading indicator: A measure that changes before an outcome and helps predict it.
Lagging indicator: A measure that reflects outcomes that have already happened.
Variance to plan: The difference between actual results and the target or budget, often also stated as a percentage of plan.
Window function: A SQL function that calculates across a set of related rows using OVER, without collapsing them into groups.
PARTITION BY: The part of a window definition that restarts the calculation for each group, such as each region.
XLOOKUP: A spreadsheet function that finds a key in one range and returns the matching value from another range.
LAG: A window function that returns a value from a previous row, used for period-over-period comparisons.
Sanity check: A quick test of whether a result is plausible, such as comparing its magnitude with expectations.
Reconciliation: Comparing results with an independent trusted source to confirm they agree.
Control total: A known sum or count used to verify that data was processed completely and correctly.
Integer division: Division of whole numbers that discards the remainder, such as 7 / 2 returning 3 in some SQL dialects.

Domain 4: Visualization and reporting (20%)

Exam tips

Trend over time means line; category comparison means bar; relationship means scatter; distribution means histogram or box plot; part of a whole with few categories means pie or 100% stacked bar.
Match the dashboard type to the audience: strategic for executives, tactical for managers, operational for real-time staff. Put the key KPIs at the top left.
Bars start at zero. Don't encode meaning with color alone. A diverging palette fits above and below target; a sequential palette fits low to high.
A formal record that must not change means static. Ongoing monitoring with interaction means dynamic. A one-time question means ad hoc. Business users building their own views means self-service.
Lead with the conclusion for executives, tailor detail to the audience, and always state limitations. Correlational findings should be presented as associations, not proven causes.
Detail that supports but clutters goes in an appendix. How current the data is means the refresh or as-of date. How the numbers were produced means the methodology.
Operational, react-now needs mean real-time; formal month-end history means snapshot. If a scheduled report shows old data, check whether the refresh ran after the upstream load, and whether it failed.
Conflicting copies point to versioning with numbers, dates and a change log, plus one authoritative location. Consistent fonts, colors and formats point to a style guide.
Old data means check refresh history, credentials and upstream loads. A filter that doesn't affect a visual means check relationships and slicer connections. Doubled totals mean suspect duplicate keys or many-to-many joins.

Key terms

Histogram: A chart of a numeric variable's distribution, with adjacent bars showing counts in each value range (bin).
Box plot: A chart summarizing a distribution with its median, quartiles, whiskers and outliers.
Choropleth map: A map that shades geographic areas by the value of a measure.
Scatter plot: A chart plotting two numeric variables against each other, one point per record.
KPI card: A dashboard visual showing a single key measure, usually with a comparison to target or a prior period.
Drill-down: Navigating from summary data to more detailed levels of a hierarchy within a visual.
Slicer (filter): An interactive control that limits the data shown in a dashboard's visuals.
Wireframe: A simple sketch of a dashboard's layout used to agree on content before building.
Sequential palette: Shades of one hue from light to dark, used for ordered values from low to high.
Diverging palette: Two contrasting hues around a neutral midpoint, used for values above and below a reference.
Truncated axis: A value axis that does not start at zero, which exaggerates differences in bar charts.
Chart junk: Visual elements such as 3D effects and heavy gridlines that add no information and distract from the data.
Static report: A fixed report, such as a PDF, that captures data at a point in time and does not change.
Dynamic report: A report connected to data that refreshes and allows interaction such as filtering and drill-down.
Ad hoc report: A one-time report created to answer a specific question.
Self-service reporting: Letting business users build their own reports from governed, trusted data sources.
Data storytelling: Presenting data findings as a structured narrative of context, insight and recommendation.
Limitation: A known weakness or constraint of an analysis, such as missing data or a small sample.
Assumption: A condition taken as true for an analysis, such as stable prices, that readers should know about.
Actionable insight: A finding specific enough that a stakeholder can decide or act on it.
Methodology: The description of how the analysis was carried out, including data preparation and calculations.
Data as-of date: The point in time the report's data reflects, shown so readers know how current it is.
Disclaimer: A note limiting how the report should be interpreted or used, such as 'preliminary, unaudited figures'.
Appendix: A section at the end of a report holding supporting detail such as full tables and definitions.
Scheduled refresh: An automated update of a report's imported data at set times.
Live (direct) connection: A report connection that queries the source whenever users interact, so data is always current.
Snapshot: A stored copy of data values at a point in time, used to report history as it was.
Subscription: A scheduled delivery of a report or link to users, often by email.
Version control: Tracking and managing changes to files or reports over time, with the ability to compare and restore versions.
Change log: A record of what changed in each version, when, why and by whom.
Style guide: A documented set of rules for fonts, colors, formats, terminology and chart conventions.
Template: A predefined layout and theme that new reports start from to ensure consistency.
Stale data: Data in a report that is older than expected because of a failed or mistimed refresh or pipeline.
Data model relationship: A link between tables in a BI model that lets filters and calculations flow between them.
Cardinality: The number of distinct values in a column, or the one-to-one, one-to-many or many-to-many nature of a relationship.
Pre-aggregation: Summarizing detailed data in advance to the level visuals need, to speed up queries.

Domain 5: Data governance (16%)

Exam tips

Decides and is accountable means owner. Defines and maintains quality means steward. Implements technical controls such as backups, permissions and encryption means custodian.
Where data came from and what transformed it means lineage. What a column means means data dictionary. A searchable inventory of datasets means data catalog.
Missing means completeness; wrong format or impossible value means validity; disagreement between systems means consistency; duplicates mean uniqueness; out of date means timeliness; not matching reality means accuracy.
Profiling discovers what the data looks like; validation rules enforce what it should look like; monitoring runs the rules continuously with alerts. Fix at the source when you can.
The same customer or product differing across systems points to master data management. One authoritative version everyone uses is the single source of truth.
Health data linked to a person means PHI (HIPAA). Card numbers mean PCI DSS. Removing names alone may still leave quasi-identifiers that re-identify people.
EU personal data means GDPR; US health data held by providers and insurers means HIPAA; card data means PCI DSS (a standard, not a law). Data location laws mean sovereignty or residency.
Reversible with a key means pseudonymization; irreversible means anonymization. Realistic fake values for testing means masking. Viewing only your region's rows means row-level security.
Data kept past its retention period, especially stray copies, is a disposal-stage failure. Collect only what you need at the start, and delete every copy at the end.

Key terms

Data owner: The accountable business leader who decides a data domain's classification, access and use.
Data steward: The business-side role that manages data definitions, quality and proper use day to day.
Data custodian: The technical role that stores, secures and maintains data according to the owner's decisions.
Data consumer: Anyone who uses data for their work and must follow the policies on its use.
Metadata: Data that describes other data, such as its structure, meaning, owner and processing history.
Data dictionary: Documentation of each field's name, definition, type, format, allowed values and source.
Data catalog: A searchable inventory of data assets and their metadata that helps users find and understand data.
Data lineage: A record of where data came from and every transformation it passed through to reach its destination.
Completeness: The degree to which required data values are present.
Validity: The degree to which values conform to rules for type, format, range and allowed values.
Consistency: The degree to which data agrees across systems and within a dataset.
Timeliness: The degree to which data is current and available when needed.
Data profiling: Analyzing a dataset's structure and content, such as nulls, distinct values, ranges and patterns, to understand its quality.
Validation rule: A check that data meets an expectation, such as a type, range, format or allowed value.
Referential integrity check: A test that every foreign key value matches an existing record in the related table.
Data quality scorecard: A report tracking quality metrics against thresholds over time.
Master data: Core shared data about key business entities such as customers, products and suppliers.
Master data management (MDM): The processes and tools that create and maintain one consistent, authoritative version of master data.
Golden record: The single, trusted, consolidated record for an entity produced by MDM.
Single source of truth: The principle that each data element has one authoritative source everyone uses.
PII: Personally identifiable information: data that can identify a specific individual directly or in combination.
PHI: Protected health information: individually identifiable health information covered by HIPAA.
Quasi-identifier: An attribute such as birth date or postal code that can identify a person when combined with others.
Data classification: Assigning data a sensitivity level (such as public, internal, confidential or restricted) that determines its handling.
GDPR: The EU regulation governing processing of personal data of people in the EU, including data subject rights.
HIPAA: The US law protecting individually identifiable health information held by covered entities and business associates.
Data sovereignty: The principle that data is subject to the laws of the country where it is located.
Retention schedule: A policy listing how long each type of record is kept and how it is disposed of.
Least privilege: Granting users and systems only the minimum access needed to do their work.
Data masking: Hiding sensitive values with realistic substitutes or partial values while keeping data usable.
Pseudonymization: Replacing identifiers with artificial values that can be re-linked using a separately held key.
Anonymization: Irreversibly removing the ability to identify individuals from data.
Encryption at rest: Encrypting stored data so it is unreadable without the key if storage is accessed.
Data life cycle: The stages data passes through from collection to storage, use, sharing, archiving and destruction.
Archiving: Moving inactive data that must be retained to long-term, lower-cost storage with restricted access.
Secure disposal: Permanently destroying data at the end of its retention period so it cannot be recovered.
Cryptographic erasure: Making encrypted data unrecoverable by destroying its encryption keys.

Study Data+ for free
Lessons, quizzes, exam simulations and hands-on labs.

Open the Data+ study plan