CompTIA Data+ (DA0-002) Study Guide

Domain 1: Data Concepts and Environments

Key concepts you must know · 153 practice questions

Structured data fits a defined schema, semi-structured data carries self-describing tags without a rigid schema, and unstructured data has no predefined model at all. Most enterprise data by volume is unstructured, which is why the distinction drives both storage and technique choices.
Know the measurement scales: nominal categories have no order, ordinal categories have order but uneven intervals, interval data has equal gaps without a true zero, and ratio data has a true zero so ratios are meaningful. The scale decides which statistics and which charts are legitimate.
A warehouse holds integrated structured data modelled for analysis; a lake holds raw data in native formats until a use is found; a mart is a subject-specific subset; a lakehouse adds warehouse structure and transactions over lake storage. Transactional systems are optimised for many small writes, analytical systems for large aggregate reads.
In a dimensional model the fact table holds numeric measures and foreign keys, and the dimension tables hold the descriptive attributes used to slice them. Granularity - what one fact row represents - is the first and least reversible modelling decision.
A star schema keeps dimensions flat; a snowflake normalises them further, reducing repetition at the cost of more joins. Slowly changing dimensions force a choice between overwriting a value and adding a row that preserves history.
ETL transforms before loading; ELT loads first and transforms inside the target, which suits elastic cloud warehouses and keeps the raw data available for reprocessing. Both are preceded by a staging area that decouples the source from the processing.
Metadata describes data; a data dictionary defines each field, its type and its permitted values; a data catalogue makes datasets discoverable with their owner and sensitivity attached. Master data management maintains one authoritative version of core entities such as customer and product.
Batch processing handles accumulated data on a schedule and streaming handles records as they arrive. The choice follows the decision being supported, since streaming costs more to build and operate and a monthly report gains nothing from it.

Domain 2: Data Acquisition and Preparation

Key concepts you must know · 169 practice questions

Primary data is collected first-hand for the question at hand; secondary data was collected for something else and is cheaper and faster but rarely matches your definitions. First, second and third party data differ in how much you know about the collection and its consent basis.
Learn the sampling methods by what they do: simple random gives every item an equal chance, stratified samples each subgroup proportionally, systematic takes every nth item after a random start, cluster takes whole natural groups, and convenience takes whoever is easiest to reach and cannot support population estimates.
Bias does not shrink as the sample grows; sampling error does. A larger badly drawn sample is a more confident wrong answer, so bias is fixed by changing the frame or the selection rather than by collecting more.
Profiling comes before anything else: row counts, column types, missing value rates, distinct value counts, ranges and patterns. A category with far more distinct values than it should have is announcing inconsistent spelling, casing or whitespace.
Cleansing covers deduplication, imputation, outlier handling, standardisation, trimming, parsing and type conversion. An extreme value is investigated before it is removed, because it may be the most important observation in the dataset rather than an error.
Know the joins by what they keep: inner keeps only matches, left keeps every left row, full keeps everything, cross pairs everything. Joining on a non-unique column multiplies rows and double counts measures, which is the usual cause of a suspiciously large total.
Union stacks rows and requires matching column count, order and types; a join places columns side by side on a key. Pivoting turns column values into columns and unpivoting collapses them back into name and value pairs.
In a query, WHERE filters rows before grouping and HAVING filters groups after aggregation, so the two produce materially different answers. Window functions keep every row while adding a rank, a running total or a group comparison.

Domain 3: Data Analysis

Key concepts you must know · 184 practice questions

Mean, median and mode describe the centre; range, interquartile range, variance and standard deviation describe the spread. The median is preferred on skewed data because a single extreme value moves the mean substantially and the median hardly at all.
A gap between the mean and the median is the quickest indicator of skew, and its direction tells you which way. Report a measure of spread and a base count alongside any average, since very different distributions can share the same summary statistics.
A z score expresses distance from the mean in standard deviations. Under the empirical rule roughly 68, 95 and 99.7 per cent of a normal distribution lie within one, two and three deviations of the mean.
The null hypothesis states there is no effect; the p-value is the probability of seeing a result this extreme if the null were true. It is not the probability that either hypothesis is true, and it says nothing about how large the effect is.
A type one error rejects a true null - a false positive whose rate is the significance level. A type two error misses a real effect and is far more common than people expect, because studies are so often too small.
A confidence interval expresses precision, and quadrupling the sample halves its width because precision follows the square root. On very large datasets almost any difference reaches significance, so effect size matters more than the test result.
Correlation measures the strength and direction of a linear relationship and never establishes causation, because a confounding third factor, reverse causation or coincidence all produce the same pattern. Always plot the scatter before quoting the coefficient.
Time series analysis separates trend, seasonality, cyclical movement and noise. Compare against the same period last year to control for seasonality, and remember that a compound annual growth rate uses only the first and last values and hides everything between.

Domain 4: Visualization and Reporting

Key concepts you must know · 153 practice questions

Choose the chart from the question: bar for comparing categories, line for change over time, scatter for the relationship between two numeric variables, histogram or box plot for a distribution, stacked bar or pie for composition, map for geography.
Bar charts must start at zero because length is read as proportional to the value; line charts may start elsewhere because they encode change rather than magnitude. A truncated bar axis is the classic way a chart misleads while every number on it remains correct.
Pie charts fail because people compare angles and areas poorly, so keep them to two or three very different shares and combine the rest. Dual axis charts are objectionable because the apparent relationship depends entirely on how the two axes were scaled.
Sort categorical bars by value unless the categories have a natural order such as age bands or a satisfaction scale, where rearranging destroys the sequence that gives them meaning.
A dashboard carries the small set of measures that show how things stand, with the most important element top left, a visible last-refreshed stamp and a clear statement of the period covered. Filters need sensible defaults and a visible indication of what is currently applied.
Accessibility means enough colour contrast, encoding that does not rely on colour alone, keyboard reachable controls and alternative text that states the finding. Around one man in twelve cannot reliably separate red from green.
Anything a reader must see belongs on the chart itself, because charts are copied into slides and messages without their surrounding text and tooltips vanish in print and on touch devices. Units, scaling and period belong in the subtitle.
Lead with the conclusion, tailor depth and vocabulary to the audience but never the finding, and state a caveat prominently if it could reverse the conclusion. A report that answers no decision anybody is taking goes unused however well it is built.

Domain 5: Data Governance

Key concepts you must know · 108 practice questions

Governance is the framework of ownership, policy and standards. A policy states what must be achieved and a standard states how; a data owner is accountable for a dataset including access, and a steward maintains its definitions and drives quality fixes.
The quality dimensions are accuracy, completeness, consistency, timeliness, uniqueness and validity. Fitness for purpose is judged against the intended use rather than in the abstract, so address data adequate for regional analysis may be unfit for delivery routing.
Personal data identifies an individual alone or in combination, which is why removing names is not sufficient - a postcode, a date of birth and a gender together identify very few people. Special category data such as health or biometrics carries additional protection.
Know the protection techniques and how they differ: encryption is reversible with a key and does nothing about a legitimate user; masking obscures the display; tokenisation replaces the value with a meaningless token that maps back only in a vault; pseudonymisation keeps a reversible link so the data is still personal; anonymisation is irreversible.
Role based access control plus least privilege is the standard model, backed by periodic access reviews that remove what people no longer need. Segregation of duties means nobody both makes and approves a change unchecked.
A breach is unauthorised access, disclosure or loss, including accidental loss. Contain and report immediately through the defined process, because notification deadlines run from discovery rather than from confirmation.
Retention states how long each category is kept and what happens then; disposal must be irreversible and recorded, and it must reach backups, extracts and reporting copies rather than only the primary system. A legal hold suspends normal disposal.
Lineage supports impact analysis before a change and locating every copy for deletion. An audit tests whether controls actually operated, so contemporaneous records are what count as evidence rather than assertions or the policy document itself.

CompTIA Data+ (DA0-002) exam tips

For every chart question, name the underlying question first - comparison, composition, distribution, relationship or trend - and the chart type usually follows in one step. Most wrong answers on this topic come from choosing a chart and then justifying it.
Learn the join behaviours cold, and specifically what happens on a non-unique key. A large share of scenario questions describe a total that came out wrong and the answer is either a many-to-many multiplication or an inner join that silently dropped unmatched rows.
Do not confuse the two senses of normalisation. In database design it means structuring tables so each fact is stored once; in statistics it means rescaling values to a common range. The exam uses both.
Memorise the difference between bias and sampling error: bias does not shrink as the sample grows. That single fact answers several questions about sampling design and about what a larger sample does and does not fix.
Be precise about what a p-value is - the probability of data this extreme if the null were true - and about what it is not. Questions frequently offer the plausible misreadings as distractors, and effect size is what the business actually needs.
Learn the protection techniques as a set: encryption, masking, tokenisation, pseudonymisation and anonymisation. The exam tests which one is reversible, which keeps a link back to the individual and which takes the data outside privacy regulation.
For governance questions, prefer the answer that fixes the problem at source and leaves a record. Correcting a figure in a report, deleting evidence and adding a note while leaving the cause in place are recurring wrong answers.
Watch for the honest-presentation questions. Truncated axes, missing periods, flattering comparison bases, tiny denominators and undisclosed filters all mislead without stating anything false, and the exam expects you to name them.

Study guide FAQ

What is the Data+ exam format and passing score?

DA0-002 is a maximum of 90 questions in 90 minutes, combining multiple-choice with performance-based items. The passing score is 675 on a scale of 100 to 900, and the exam is available in English and Japanese. It is proctored, at a test centre or online.

How are the Data+ objectives weighted?

CompTIA publishes five domains: Data concepts and environments 20 percent, Data acquisition and preparation 22 percent, Data analysis 24 percent, Visualization and reporting 20 percent, and Data governance 14 percent. They sum to exactly 100, so analysis and preparation together are nearly half the exam. Our practice sets draw to the same distribution.

Do I need to know a specific tool such as SQL, Excel or a BI product?

No. Data+ is vendor-neutral, so it tests concepts rather than product syntax. You should understand what a SELECT, a GROUP BY and a HAVING clause do and how join types behave, but you will not be asked to write working code in a particular dialect or to navigate a specific tool's menus.

How much statistics does Data+ require?

More than candidates expect, but at an interpretive level. You need descriptive statistics, distributions, hypothesis testing vocabulary, confidence intervals, correlation and regression, and time series concepts - understood well enough to choose the right measure and read a result correctly, rather than to derive anything by hand.

How does Data+ differ from the previous DA0-001 version?

DA0-002 is the current version and reorganises the objectives into five domains weighted 20/22/24/20/14, with data acquisition and preparation treated as its own substantial domain and governance expanded to reflect modern privacy obligations. Study against the DA0-002 objectives rather than older material.

What is the most common reason candidates fail?

Preparing only for the analysis domain and neglecting preparation, visualization and governance, which together are 56 percent of the exam. Candidates from a reporting background also lose marks on sampling and statistical interpretation, while those from a statistics background lose them on joins, data quality and governance obligations.

Related Data resources

Test yourself: 767 CompTIA Data+ (DA0-002) practice questions