Google Data Analytics · Domain 4 of 8

Google Data Analytics Domain 4: Process Data from Dirty to Clean

The process phase: checking integrity, deciding whether you have enough data, finding and fixing dirty data in spreadsheets and SQL, then verifying and documenting every change so others can trust the result.

Exam weight estimate
Not published
Official modules
5
Estimated questions estimate
4 module challenges
Suggested study time estimate
20–30 hrs

About 16 min read Reviewed October 2026

Overview

Process Data from Dirty to Clean is the fourth required course and covers the process phase of the analysis process. It begins with data integrity (accuracy, completeness, consistency and trustworthiness) and what to do when data is insufficient, including sample size, margin of error, confidence level and statistical power. It then teaches the main types of dirty data and how to clean them in spreadsheets with tools such as remove duplicates, TRIM, SPLIT, conditional formatting and COUNTIF.

The third module moves cleaning into SQL with DISTINCT, TRIM, LENGTH, SUBSTR, CAST, COALESCE and CONCAT. The fourth teaches verifying results against the original and recording changes in a changelog. A short module on adding data skills to a resume follows. The course is where the certificate stops trusting data by default and starts proving it is fit for analysis.

How it's tested

Course 4 module challenges combine concept items with practical ones. Expect scenario questions about whether data is sufficient and what to do next, items that ask which dirty-data type a sample shows, and short spreadsheet or SQL snippets where you choose the function or predict the cleaned result. Verification items ask what to check or record before sharing.

  • Is the data good enough?

    A dataset is described as limited by source, time, geography or size, and you choose the best response, such as gathering more data, using proxy data or adjusting the objective with stakeholders.

  • Match the fix to the dirt

    Sample values show duplicates, extra spaces, mixed formats or nulls, and you pick the spreadsheet tool or SQL function that fixes it.

  • Verify and document

    Cleaning is finished and you choose what to compare, check or record in a changelog before analysis.

Reasoning it through

A scenario shows a SQL table where some state values are 'OH ' with a trailing space, so a GROUP BY splits Ohio into two groups. The question asks which function fixes this. Candidates pick DISTINCT because the result looks duplicated. DISTINCT removes identical rows, but 'OH' and 'OH ' are not identical. TRIM removes leading and trailing spaces so the values match, after which grouping works correctly.

Modules 5

Official modules from the Google outline, in outline order.

4.1 The importance of integrityData integrity, insufficient data, sample size and margin of error.
What it means
Data integrity is the accuracy, completeness, consistency and trustworthiness of data across its life cycle. It can be compromised by data replication, transfer, manipulation, human error, malware and system failures.
What to understand
Data can be insufficient when it comes from only one source, keeps updating, is outdated or is geographically limited. The responses include using available data to identify trends, waiting for more data, discussing a changed objective with stakeholders, or finding a new or proxy dataset. Margin of error is the maximum expected difference between sample and population results, and confidence level, typically 90% to 99%, is how sure you are that the true value lies within it.
How it shows up
A scenario describes limited or unrepresentative data and asks what the analyst should do, or gives survey figures and asks what the margin of error or confidence level means.
Common confusion
Treating confidence level and margin of error as the same; one is how sure you are, the other is the width of the range around the estimate.
4.2 Clean data for more accurate insightsDirty versus clean data and cleaning in spreadsheets.
What it means
Dirty data is incomplete, incorrect or irrelevant to the problem. Google names the main types as duplicate, outdated, incomplete, incorrect or inaccurate, and inconsistent data.
What to understand
Spreadsheet cleaning tools include data validation, conditional formatting, remove duplicates, text to columns or SPLIT, CONCATENATE, COUNTIF, LEN, LEFT, RIGHT, MID and TRIM. Data mapping matches fields from one source to another when merging datasets, and checking field lengths and data types catches errors before they spread.
How it shows up
A small table with problems such as mixed date formats or repeated rows is shown and you choose the right tool or name the dirty-data type.
Common confusion
Calling a misspelled value 'inconsistent' when it is incorrect; inconsistent data uses different formats for the same thing, such as 'USA' and 'United States'.
4.3 Data cleaning with SQLUsing SQL queries and functions to clean and transform data.
What it means
For large datasets, SQL cleans data where it is stored, using statements such as CREATE TABLE IF NOT EXISTS, INSERT INTO, UPDATE and DROP TABLE IF EXISTS, together with string and type functions.
What to understand
DISTINCT returns unique rows, TRIM removes leading and trailing spaces, LENGTH checks string length, SUBSTR extracts part of a string, CAST converts a value to another data type, CONCAT joins strings, and COALESCE returns the first non-null value in a list. MIN and MAX quickly expose out-of-range values.
How it shows up
A query snippet and a data problem are shown and you choose the function that fixes it or predict the output.
Common confusion
Using DISTINCT to fix values that differ only by spaces or case; they are not exact duplicates until they are trimmed or standardized.
4.4 Verify and report on cleaning resultsChecking cleaned data and documenting changes.
What it means
Verification confirms that cleaning worked and that the data still answers the business question; reporting documents what changed so teammates can trust and reproduce it.
What to understand
Verification means comparing cleaned data with the original, stepping back to the big picture, and using tools such as find and replace, pivot tables, COUNTA and SQL CASE statements to catch remaining errors. A changelog records each change, why it was made and when; version history in spreadsheets and SQL comments also support documentation.
How it shows up
Cleaning is done and the question asks which check to run or what to document before analysis begins.
Common confusion
Treating version history as a changelog; version history stores snapshots automatically, while a changelog explains the reasons for each change.
4.5 Add data to your resumePresenting data analytics skills on a resume.
What it means
A short career module on building a resume that highlights transferable skills and the analytics experience gained in the program.
What to understand
Describe accomplishments with specific, measurable results and name tools such as spreadsheets, SQL and Tableau, tailoring the resume to each role.
How it shows up
This module lists no graded assignments; it prepares material reused later in the capstone and job search.
Common confusion
Listing course titles instead of what you did and the result it produced.

High-yield concepts

The terms, rules and formulas this domain turns into questions most often.

  • Data integrity

    The accuracy, completeness, consistency and trustworthiness of data throughout its life cycle.

    Remember
    Replication, transfer and manipulation are the three process risks Google highlights, alongside human error, malware and system failures.
    Tested as
    Opening concept of Module 1.
    Trap
    Equating integrity with security; secure data can still be inaccurate.

  • Insufficient data

    Data that is too limited to answer the question, for example from one source, outdated, constantly updating or geographically narrow.

    Remember
    Options are to use what you have to identify trends, wait for more, agree a revised objective with stakeholders, or find new or proxy data.
    Tested as
    'What should the analyst do next' scenarios.
    Trap
    Running the analysis anyway and adding a footnote.

  • Margin of error

    The maximum amount sample results are expected to differ from the actual population.

    Remember
    A larger sample gives a smaller margin of error for the same confidence level.
    Tested as
    Interpretation items, such as whether two survey results genuinely differ.
    Trap
    Concluding a difference is real when the results overlap within the margin of error.

  • Confidence level and confidence interval

    Confidence level is how sure you are the true value lies in the range; the confidence interval is that range.

    Remember
    Common confidence levels are 95% and 90%; higher confidence needs a larger sample.
    Tested as
    Sample-size questions.
    Trap
    Swapping the two definitions.

  • Statistical power and significance

    Statistical power is the probability that a test detects a real effect; statistical significance is how unlikely a result is to be due to chance.

    Remember
    A statistical power of 0.8 (80%) is the usual minimum for a test to be considered reliable.
    Tested as
    Testing-your-data items.
    Trap
    Assuming any observed difference in a small test is significant.

  • Types of dirty data

    Duplicate, outdated, incomplete, incorrect or inaccurate, and inconsistent data.

    Remember
    Inconsistent means the same thing recorded different ways; incorrect means wrong values.
    Tested as
    Classification items in Module 2.
    Trap
    Labeling a typo as inconsistent rather than incorrect.

  • Spreadsheet cleaning functions

    TRIM, LEN, LEFT, RIGHT, MID, SPLIT, CONCATENATE and COUNTIF handle spacing, length checks, extraction, splitting, joining and counting.

    Remember
    COUNTIF counts cells meeting a condition, which quickly reveals misspellings or duplicates.
    Tested as
    Hands-on and function-choice items.
    Trap
    Using LEN to fix lengths; it only reports them.

  • Data validation and conditional formatting

    Data validation restricts what can be entered in a cell; conditional formatting highlights cells that meet a rule.

    Remember
    Validation prevents bad entries; formatting reveals existing ones.
    Tested as
    Prevent-versus-detect items.
    Trap
    Choosing conditional formatting to stop bad entries.

  • SQL cleaning functions

    DISTINCT, TRIM, LENGTH, SUBSTR, CAST, CONCAT and COALESCE clean and transform data in queries.

    Remember
    COALESCE(a, b) returns the first non-null value; CAST(x AS INT64) changes data type in BigQuery.
    Tested as
    Snippet-reading items in Module 3.
    Trap
    Thinking COALESCE deletes null rows.

  • Data mapping and schema

    Data mapping matches fields from one source to another; a schema describes how data is organized.

    Remember
    Mapping is essential when merging systems with different field names or formats.
    Tested as
    Merging-data scenarios.
    Trap
    Merging datasets without checking that fields mean the same thing.

  • Changelog

    A file that records notable changes to a dataset or project in chronological order, with what changed, why and when.

    Remember
    It lets others reproduce and trust the cleaning.
    Tested as
    Module 4 documentation items.
    Trap
    Relying on automatic version history as documentation.

  • Proxy data

    Data from a similar, related situation used as a stand-in when direct data is not available, such as another city's ride data to estimate a new market.

    Remember
    Proxy data must be clearly flagged as an approximation.
    Tested as
    Insufficient-data scenarios.
    Trap
    Presenting proxy results as direct evidence.

Side by side

Concepts candidates mix up, compared on the point that decides the answer.

Margin of error vs. confidence level vs. confidence interval

ConceptMeaningKey differenceCommon trap
Margin of error Maximum expected difference between sample and population Expressed as plus or minus a value, such as plus or minus 3 points Thinking it measures certainty
Confidence level How sure you are that the population value is in the range A percentage such as 95% Treating it as the range itself
Confidence interval The range of likely values for the population Estimate plus and minus the margin of error Confusing it with the confidence level

SQL cleaning functions

ConceptMeaningKey differenceCommon trap
DISTINCT Returns unique rows Removes exact duplicates only Using it on values that differ by spaces or case
TRIM Removes leading and trailing spaces Fixes hidden-space mismatches Expecting it to remove spaces inside a string
CAST Converts a value to another data type Fixes numbers stored as text and similar issues Using it to change formatting only
COALESCE Returns the first non-null value from a list Substitutes a fallback for nulls Thinking it removes null rows

Data validation vs. conditional formatting

ConceptMeaningKey differenceCommon trap
Data validation Rules that restrict what can be entered in a cell Prevents bad data at entry Using it to find errors already present
Conditional formatting Formatting applied when a cell meets a rule Highlights existing problems Expecting it to block bad entries

Exam traps

Each trap is an answer that looks right. Learn the clue that should change your mind.

  1. Run it anyway

    What candidates pick
    Proceed with the analysis using data from a single store to represent the whole chain.
    Why it looks right
    It meets the deadline and some data is better than none.
    Why it's wrong
    Data from one source or location is insufficient for a chain-wide conclusion; the analyst should flag the limit, seek more data or agree a narrower objective.

    Clue to watch for The scenario mentions a single source, region or short time period.

  2. Overlap read as a difference

    What candidates pick
    Conclude version A beat version B because 52% is higher than 49%.
    Why it looks right
    The numbers differ.
    Why it's wrong
    With a margin of error of plus or minus 4 points, the ranges overlap, so the difference may not be real.

    Clue to watch for A stated margin of error that is larger than half the gap between results.

  3. DISTINCT for messy text

    What candidates pick
    SELECT DISTINCT to merge 'Ohio' and 'Ohio '.
    Why it looks right
    The problem looks like duplication.
    Why it's wrong
    DISTINCT only removes identical values; TRIM is needed first so the strings match.

    Clue to watch for Values that look identical but group separately.

  4. Formatting as prevention

    What candidates pick
    Conditional formatting to stop staff entering negative quantities.
    Why it looks right
    It makes bad values visible.
    Why it's wrong
    Conditional formatting only highlights; data validation is what blocks invalid entries.

    Clue to watch for The goal is to prevent entry rather than find existing errors.

  5. Version history is enough

    What candidates pick
    Rely on the spreadsheet's version history to document cleaning.
    Why it looks right
    Every change is technically recorded.
    Why it's wrong
    Version history shows what changed but not why; a changelog explains decisions so teammates can verify and repeat them.

    Clue to watch for A teammate needs to understand or reproduce the cleaning.

  6. Delete every null

    What candidates pick
    Remove all rows with any missing value.
    Why it looks right
    It produces a tidy dataset quickly.
    Why it's wrong
    Deleting rows can bias results or discard useful data; the analyst should understand why values are missing and use a fallback such as COALESCE where appropriate.

    Clue to watch for Many rows have nulls in a non-critical field.

Answer framework

Use these steps when a Course 4 item asks how to make data fit for analysis.

  1. Check integrity first: is the data accurate, complete, consistent and trustworthy, and could replication, transfer or manipulation have damaged it?
  2. Decide whether the data is sufficient for the objective; if not, choose more data, proxy data or a revised objective agreed with stakeholders.
  3. Name the dirty-data type precisely: duplicate, outdated, incomplete, incorrect or inconsistent. Misnaming the problem leads to the wrong fix.
  4. Pick the tool that fixes that exact problem, such as TRIM for spaces, CAST for types, COALESCE for nulls or validation to prevent entry errors.
  5. Verify against the original and the business question, then record each change and its reason in a changelog.

Practice question

In a BigQuery table, the column zip_code was imported as text, and some rows are null. An analyst wants a numeric zip column where null values are replaced with 0. Which expression does this?

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

Start with Module 1 and make the statistics concrete. Use an online sample-size calculator with different confidence levels and margins of error to see how sample size responds, and practice deciding whether two survey results really differ. Learn the four kinds of insufficient data with one remedy for each.

Spend the most time on Modules 2 and 3. Download a messy public dataset and clean it twice: once in Google Sheets with TRIM, SPLIT, COUNTIF and remove duplicates, then again in BigQuery with TRIM, CAST, COALESCE and DISTINCT. Doing the same fix in both tools fixes the function names in memory and shows when SQL is the better choice.

The underestimated area is Module 4. Write a real changelog for your cleaning exercise and compare row counts and key totals before and after. Afterward, take a practice set mixing statistics interpretation, dirty-data classification and SQL function choice, then move to Course 5, which assumes your data is clean.

FAQ

Is there a lot of statistics in Course 4?

Only the basics. You need to understand sample size, margin of error, confidence level, confidence interval, statistical power and significance well enough to interpret results, not to derive formulas.

How is Course 4 assessed?

Modules 1 to 4 each end with a graded module challenge, supported by knowledge checks and hands-on cleaning activities in spreadsheets and SQL. The resume and wrap-up modules list no graded assignments.

What are the five types of dirty data?

Duplicate, outdated, incomplete, incorrect or inaccurate, and inconsistent data. Questions often show a small table and ask which type is present.

When should I clean data in SQL rather than a spreadsheet?

Use SQL when the data is large, stored in a database or shared by many users, because it cleans data where it lives and can be rerun. Spreadsheets suit smaller datasets you can inspect directly.

What is the difference between DISTINCT and TRIM?

DISTINCT removes exact duplicate values or rows. TRIM removes leading and trailing spaces so values that look the same actually match, which often has to happen before DISTINCT or GROUP BY works.

How long should I plan for Course 4?

Coursera lists about 16 hours. Our editorial planning estimate is 20 to 30 hours, mostly for hands-on cleaning practice.

Why does a cleaning course include documentation?

Because cleaned data is only trustworthy if others can see what was changed and why. The capstone also expects you to document your cleaning steps.

Other Data & Analytics certifications

Where the same subject shows up in other exams. Cards marked "Same subject" open the matching domain.

Test your Domain 4 knowledge

You have covered data integrity, insufficient data, sample-size terms, cleaning in spreadsheets and SQL, and documenting changes. Test whether you can match each data problem to the right fix.

Take the Google Data Analytics practice test Check your Google Data Analytics readiness

Keep going