Overview
Prepare Data for Exploration is the third required course and covers the prepare phase of the analysis process: collecting data, judging whether it can be trusted, and storing it so it can be found and protected. It teaches data types and structures (structured versus unstructured, wide versus long, first-, second- and third-party), types of bias, the ROCCC test for good data sources, data ethics and privacy, and open data. The longest module covers relational databases, primary and foreign keys, metadata, importing data, sorting and filtering, and first SQL queries in BigQuery.
The course ends with organizing and securing data through file naming conventions and access controls, plus a short module on building your presence in the data community. Its job in the program is to make sure the data entering the process phase is relevant, credible and ethically obtained.
How it's tested
Course 3 module challenges are mostly identification and judgment items: which bias, which ROCCC element is missing, which ethics principle is violated, which column is a primary key, or what a short BigQuery query returns. Scenarios are brief descriptions of a data source or collection method. The best answer names the precise term, not a near-synonym.
-
Judge the source
A dataset is described by who collected it, when and how, and you decide which ROCCC element it fails or whether it is first-, second- or third-party data.
-
Name the bias or ethics issue
A collection or analysis practice is described and you identify the bias type (sampling, observer, interpretation, confirmation) or the ethics aspect involved.
-
Read the table or query
A small table or SQL statement is shown and you identify the primary or foreign key, the metadata type, or the rows a WHERE clause returns.
A scenario describes an analyst using a popular aggregator site's summary of hospital wait times, last updated four years ago, with no link to the original source. The question asks what is wrong. Candidates notice only the age and pick 'not current'. The data also fails original and cited, because it is a secondhand summary with no traceable source. The best answer addresses credibility: find the original agency's current, cited data.
Modules 5
Official modules from the Google outline, in outline order.
3.1 Data types and structuresHow data is collected and the types, formats and structures analysts work with.
- What it means
- Collection decisions cover how data is gathered, which source to use, how much to collect and over what time frame. Data is described by source (first-party, second-party, third-party), format (qualitative or quantitative, discrete or continuous, nominal or ordinal, internal or external) and structure (structured or unstructured).
- What to understand
- First-party data is collected by your own organization, second-party is another group's first-party data shared directly with you, and third-party data comes from an outside provider that did not collect it themselves. In wide data each subject has one row with many attribute columns; in long data each row is one observation at one point in time, so a subject repeats across rows.
- How it shows up
- A description of a dataset or a spreadsheet layout asks you to name its source type, format or whether it is wide or long.
- Common confusion
- Confusing second-party with third-party data; the difference is whether the provider collected the data itself.
3.2 Data responsibilityBias, credibility (ROCCC), data ethics, privacy and open data.
- What it means
- This module teaches how to recognize biased data, test credibility with ROCCC (reliable, original, comprehensive, current, cited), and apply data ethics and privacy, including the value and limits of open data.
- What to understand
- Sampling bias means the sample does not represent the population; observer bias is seeing what you expect while recording; interpretation bias is reading ambiguous results through your own lens; confirmation bias is favoring evidence that supports what you already believe. Google lists six aspects of data ethics: ownership, transaction transparency, consent, currency, privacy and openness.
- How it shows up
- A collection method or source is described and you identify the bias, the failing ROCCC element or the ethics aspect at stake.
- Common confusion
- Treating 'currency' in data ethics as data being up to date; in the ethics list it means people should know about the financial transactions resulting from use of their data.
3.3 Database essentialsRelational databases, keys, metadata, data sources, sorting and filtering, and SQL in BigQuery.
- What it means
- A relational database stores data in tables linked by keys. Metadata is data about data, and this module also covers importing data, sorting and filtering in spreadsheets, and writing basic queries in BigQuery.
- What to understand
- A primary key uniquely identifies each row, cannot be null and is defined once per table; a foreign key is a column in one table that refers to the primary key of another, which is what lets tables be joined. Metadata is descriptive (what the data is), structural (how it is organized) or administrative (its technical source, rights and access).
- How it shows up
- A pair of tables is shown and you pick which column is the primary key or the foreign key, or a query is shown and you identify its output.
- Common confusion
- Thinking a foreign key must be unique in its own table; it can repeat, such as one customer ID appearing on many orders.
3.4 Organize and protect dataFile naming conventions, folder structure and data security.
- What it means
- Organizing data means consistent file naming, logical folders and archiving older files; protecting it means controlling who can view or change it.
- What to understand
- Good file names are consistent and meaningful, include a date in a sortable format such as YYYYMMDD, use a version number such as v02, and avoid spaces and special characters. Security measures include encryption, which makes data unreadable without a key, and tokenization, which replaces sensitive values with randomly generated tokens, along with access permissions and protected sheets.
- How it shows up
- Several file names are listed and you choose the one that follows good conventions, or a security need is described and you choose the control.
- Common confusion
- Confusing encryption with tokenization; encryption transforms the data with an algorithm and key, while tokenization swaps in a stand-in value and stores the original elsewhere.
3.5 Engage in the data communityBuilding an online presence and networking with data professionals.
- What it means
- This short module encourages learners to build a professional presence, for example on LinkedIn and Kaggle, and to connect with analysts through communities and mentors.
- What to understand
- Sharing work publicly, such as Kaggle notebooks or datasets, gives employers evidence of skills, and networking and mentorship help with learning and job searching.
- How it shows up
- This module has a self-reflection activity and no module challenge, so it is lightly assessed.
- Common confusion
- Skipping it entirely; its Kaggle profile work is reused in the capstone portfolio.
High-yield concepts
The terms, rules and formulas this domain turns into questions most often.
-
ROCCC
Good data sources are reliable, original, comprehensive, current and cited.
- Remember
- Original means you can trace data back to the organization that collected it; cited means the source is documented and vetted.
- Tested as
- The core credibility test in Module 2 and in the capstone.
- Trap
- Accepting a large, popular dataset as credible without checking originality and citation.
-
Types of bias
Sampling, observer, interpretation and confirmation bias each enter at a different point in collection or analysis.
- Remember
- Random sampling reduces sampling bias; standardized procedures reduce observer bias.
- Tested as
- Bias-identification scenarios.
- Trap
- Calling any skewed result sampling bias when the issue is how observations were recorded or interpreted.
-
Data ethics aspects
Ownership, transaction transparency, consent, currency, privacy and openness.
- Remember
- Individuals own their raw data; consent must be informed and explicit; transaction transparency means processing activities and algorithms should be explainable to the people whose data is used.
- Tested as
- Ethics scenarios in Module 2.
- Trap
- Reading 'currency' as timeliness.
-
Data anonymization and PII
Personally identifiable information (PII), such as names, phone numbers and IDs, can identify a person; anonymization masks or removes it.
- Remember
- Removing a name is not enough if other fields together still identify someone.
- Tested as
- Privacy questions.
- Trap
- Assuming anonymized data can never be re-identified.
-
First-, second- and third-party data
First-party is collected by you, second-party is collected by another group and sold or shared directly, third-party is aggregated from sources by a provider that did not collect it.
- Remember
- Third-party data needs the most scrutiny for reliability and originality.
- Tested as
- Source-classification items.
- Trap
- Labeling a partner's own customer survey as third-party.
-
Wide vs. long data
Wide data gives each subject one row with many columns; long data stores each time-point or measure as its own row.
- Remember
- Long data repeats the subject identifier across rows.
- Tested as
- Layout identification items.
- Trap
- Judging width by the number of rows rather than how subjects and values are laid out.
-
Primary key vs. foreign key
A primary key uniquely identifies a row; a foreign key in another table references that primary key.
- Remember
- Primary keys are unique and not null; a table can have several foreign keys but one primary key, which may be composite.
- Tested as
- Table-reading items and the basis for JOINs in Course 5.
- Trap
- Choosing a column like last name, which can repeat, as a primary key.
-
Metadata types
Descriptive metadata describes content, structural metadata describes how data is organized, administrative metadata describes technical source, rights and management.
- Remember
- A photo's camera model and file size are administrative; its title and keywords are descriptive.
- Tested as
- Metadata classification items in Module 3.
- Trap
- Calling everything descriptive metadata.
-
Open data
Data that is freely available, reusable and redistributable, with universal participation.
- Remember
- Openness still has to be balanced with privacy.
- Tested as
- Module 2 items on open data.
- Trap
- Assuming open data is automatically credible.
-
BigQuery basics
Google's cloud data warehouse, organized as project, dataset and table, queried with SQL.
- Remember
- Table names are referenced as project.dataset.table, usually inside backticks.
- Tested as
- Hands-on activities and query-reading items.
- Trap
- Mixing up dataset and table in the reference path.
-
File naming conventions
Consistent, descriptive names with sortable dates and version numbers, such as SalesReport_20260115_v02.
- Remember
- Use YYYYMMDD dates, underscores or hyphens instead of spaces, and no special characters.
- Tested as
- Module 4 items.
- Trap
- Choosing a name with a date like 1-15-26, which does not sort correctly.
-
Encryption vs. tokenization
Encryption scrambles data with a key; tokenization replaces sensitive values with randomly generated tokens.
- Remember
- Both protect data, but tokenized originals are stored in a separate secure location.
- Tested as
- Data security items.
- Trap
- Treating them as the same control.
Side by side
Concepts candidates mix up, compared on the point that decides the answer.
Types of bias
| Concept | Meaning | Key difference | Common trap |
|---|---|---|---|
| Sampling bias | Sample is not representative of the population | Problem is who was included | Using it for errors in recording or interpreting |
| Observer bias | Different people observe or record things differently | Problem is how data was recorded | Calling it interpretation bias |
| Interpretation bias | Reading ambiguous results based on prior assumptions | Problem is how results are explained | Calling it confirmation bias |
| Confirmation bias | Seeking or favoring evidence that supports existing beliefs | Problem is selective search or weighting of evidence | Calling any wrong conclusion confirmation bias |
Primary key vs. foreign key
| Concept | Meaning | Key difference | Common trap |
|---|---|---|---|
| Primary key | Column or columns that uniquely identify each row | Unique, not null, one per table | Picking a column with possible duplicates |
| Foreign key | Column that references another table's primary key | May repeat; links tables | Thinking it must be unique |
Metadata types
| Concept | Meaning | Key difference | Common trap |
|---|---|---|---|
| Descriptive | Describes the content, such as title, author or keywords | What the data is about | Applying it to file size or format |
| Structural | How data is organized, such as page order or table relationships | How parts fit together | Calling it descriptive |
| Administrative | Technical source, rights and management, such as file type, creation date and access | How data is managed | Calling it structural |
Exam traps
Each trap is an answer that looks right. Learn the clue that should change your mind.
-
Popular means credible
- What candidates pick
- A widely shared dataset from an aggregator blog.
- Why it looks right
- It is easy to find and many people use it.
- Why it's wrong
- ROCCC requires data that is original and cited; a secondhand summary with no traceable source fails even if it is popular.
Clue to watch for No link to the organization that collected the data, or a vague attribution.
-
Every skew is sampling bias
- What candidates pick
- Sampling bias, when two researchers recorded the same behavior differently.
- Why it looks right
- Sampling bias is the most familiar term.
- Why it's wrong
- Differences in how observers record data are observer bias; sampling bias is about who is in the sample.
Clue to watch for Multiple people collecting or rating data without a shared standard.
-
Second-party called third-party
- What candidates pick
- Third-party, for survey data a partner company collected and sold to you.
- Why it looks right
- The data comes from outside your organization.
- Why it's wrong
- Second-party data is another organization's own first-party data shared directly; third-party data is aggregated by a provider that did not collect it.
Clue to watch for Whether the provider collected the data itself.
-
Names as keys
- What candidates pick
- Customer last name as the primary key.
- Why it looks right
- It identifies people in everyday life.
- Why it's wrong
- Names repeat and change; a primary key must be unique and not null, so a system-generated ID is correct.
Clue to watch for Look for the column whose values are always unique, usually an ID.
-
Anonymized equals safe
- What candidates pick
- Share a dataset publicly because names were removed.
- Why it looks right
- Removing names looks like full anonymization.
- Why it's wrong
- Remaining fields such as zip code, birth date and gender together can re-identify people; PII must be fully masked or removed.
Clue to watch for Detailed demographic or location fields left in a shared dataset.
-
Pretty file names
- What candidates pick
- 'Final Sales Report Jan 15.xlsx'.
- Why it looks right
- It is readable and descriptive.
- Why it's wrong
- Spaces, the word 'final' and a non-sortable date break conventions; a name like SalesReport_20260115_v03 sorts correctly and shows the version.
Clue to watch for Compare date format, version number and spaces across the options.
Answer framework
Use these steps whenever a Course 3 item asks whether data is fit to use or how to handle it.
- Identify where the data came from and classify it as first-, second- or third-party data. Third-party data needs the closest scrutiny.
- Run ROCCC: confirm it is reliable, original, comprehensive, current and cited, and name the element that fails. A single failing element is enough to look for a better source.
- Check collection for bias, asking who was sampled, how data was recorded and how results were interpreted.
- Check ethics and privacy: ownership, consent, transparency and whether any PII needs masking before use or sharing. Remove or mask identifiers before sharing data outside the team.
- For storage questions, choose consistent file names, keys that uniquely identify rows, the right metadata type and suitable access controls.
Practice question
Want more like this? The Google Data Analytics practice test has questions for every domain.
Checklist
0 of 13
Tick each one you can do without notes. Your ticks stay in this browser.
Study strategy
Work through the modules in order, but split your effort: Module 2 (data responsibility) and Module 3 (database essentials) deserve most of your time because they hold the densest lists and the most assignments. Build flashcards for the four bias types, ROCCC, the six ethics aspects and the three metadata types, each with your own one-line example, since module challenges test the precise term.
For Module 3, do every BigQuery hands-on activity rather than watching. Create a free sandbox project, load a public dataset and write SELECT, FROM, WHERE queries until you can predict their output. Sketch two linked tables on paper and label the primary and foreign keys; this pays off in Course 5 when you write JOINs.
The underestimated areas are data types and file naming: they look trivial but generate easy-to-lose classification items. Afterward, practice judging three real public datasets with ROCCC, then move to Course 4, where you will clean the data you have learned to choose.
FAQ
What does ROCCC stand for?
Reliable, original, comprehensive, current and cited. It is Google's checklist for deciding whether a data source is credible enough to use.
How is Course 3 graded?
Modules 1 to 4 each end with a graded module challenge, and the course includes many knowledge checks and hands-on activities in Kaggle, spreadsheets and BigQuery. The short Module 5 has a self-reflection only.
Do I need to pay for BigQuery?
The course uses BigQuery activities that can be completed without paid usage, typically through a free sandbox. Check the current course instructions for setup details, since Google Cloud offers can change.
What is the difference between observer bias and interpretation bias?
Observer bias happens while data is being recorded, when different people see or measure things differently. Interpretation bias happens afterward, when ambiguous results are explained through the analyst's assumptions.
Which module takes the longest?
Database essentials, listed at about nine hours with twelve assignments. It covers databases, metadata, importing data, sorting and filtering, and SQL in BigQuery.
How much time should I plan for Course 3?
Coursera lists about 19 hours. Our editorial planning estimate is 25 to 35 hours, mostly for BigQuery practice and memorizing the classification lists.
How does this course connect to Course 4?
Course 3 teaches you to choose credible, ethical data; Course 4 teaches you to check its integrity and clean it. Recognizing biased or low-quality data here makes the cleaning decisions in Course 4 easier.
Other Data & Analytics certifications
Where the same subject shows up in other exams. Cards marked "Same subject" open the matching domain.
Test your Domain 3 knowledge
You have covered data types, bias, ROCCC, data ethics, keys, metadata and file organization. Check that you can judge a data source and read a pair of tables quickly.
Keep going
- Training Training Built on your Weak areas
- Practice test Questions by domain, with explanations
- Mock exam Full length, timed like the real exam
- Readiness test Find your weakest domain in minutes
- Cheat sheet Key terms and formulas on one page
- Ask the Google Data Analytics AI tutor Get any concept explained your way