Data Quality Management Flashcards

Question 1

What does the accuracy dimension of data quality measure?

Accepted Answer

Accuracy measures how closely a data value matches the real world entity or event it describes. A customer address is accurate when it reflects where the customer actually lives. Inaccurate values can look valid in format yet still be wrong. Checking accuracy often requires a trusted reference source.

Question 2

What does completeness mean in the context of data quality?

Accepted Answer

Completeness describes whether all required data values are present in a dataset. A record missing a postal code or phone number when those fields are mandatory is incomplete. Completeness is often measured as the percentage of populated required fields. Optional fields should not be counted against completeness.

Question 3

How does consistency differ from accuracy in data quality?

Accepted Answer

Consistency differs from accuracy because it asks whether the same fact is represented the same way across systems or within a dataset. A customer name stored as "Smith, Ann" in one system and "Ann Smith" elsewhere is inconsistent. Accuracy asks whether a value is true, while consistency asks about agreement.

Question 4

What does timeliness mean, and why does it matter for data?

Accepted Answer

Timeliness measures whether data is available and up to date when people need to use it. A stock level report built from data refreshed last week may be unfit for same day ordering decisions. Stale data can be accurate at one moment yet still mislead decision makers.

Question 5

What does validity check in a data value?

Accepted Answer

Validity checks whether a value conforms to the format, type, range, or allowed list defined for its field. A postal code with too many characters fails this check. A valid value is not necessarily accurate, since a well formed phone number can still belong to the wrong person.

Question 6

What is uniqueness, and what problem does it address?

Accepted Answer

Uniqueness means each real world entity appears only once within a dataset. Duplicate records for the same customer cause conflicting balances, repeated mailings, and inflated counts. Teams detect duplicates through matching rules that compare names, addresses, and identifiers, then merge or retire the extra copies.

Question 7

Why should data quality be judged against a specific business purpose?

Accepted Answer

Data quality is relative to how the data will be used, so the same dataset may be fit for one purpose and unfit for another. A mailing list with slightly outdated phone numbers may work for postal campaigns but fail for urgent callbacks. Quality without a stated purpose has no clear target.

Question 8

What is the difference between a data quality dimension and a data quality rule?

Accepted Answer

A dimension is a broad characteristic of quality, such as completeness or validity. A rule is a specific, testable condition that applies a dimension to a particular field, such as requiring an email address to contain an at sign. Dimensions guide the framework, while rules provide measurable checks.

Question 9

What does data profiling involve when a team first examines a dataset?

Accepted Answer

Data profiling is the process of examining a dataset to summarize its structure, content, and quality characteristics. Analysts calculate counts, null rates, value distributions, minimum and maximum values, and pattern frequencies for each column. The results reveal anomalies and hidden problems before any cleansing begins, and they provide a baseline for later comparison.

Question 10

What is a null rate, and why is it useful in profiling?

Accepted Answer

A null rate is the proportion of records in a column that lack a value. Calculated as missing entries divided by total entries, it highlights fields that are sparsely populated. A column with an unexpectedly high null rate may signal a broken data feed, an optional field used as if it were required, or a form that does not capture the information.

Question 11

What does a value frequency distribution reveal about a column?

Accepted Answer

A value frequency distribution shows how often each distinct value appears in a column. It exposes dominant categories, rare outliers, and placeholder entries such as "unknown," "N/A," or repeated default codes. If one generic value appears thousands of times, that often points to a default setting in a source system.

Question 12

How does pattern analysis help assess the format of text fields?

Accepted Answer

Pattern analysis converts each value into a generic shape, such as replacing letters with "A" and digits with "9", then counts how many values share each shape. Phone numbers stored as "999-999-9999" and "(999) 999-9999" reveal formatting drift. Rare patterns usually indicate entry errors or mixed sources.

Question 13

What is the purpose of a cross column check during data assessment?

Accepted Answer

A cross column check tests whether values in one field logically agree with values in another field in the same record. For example, a ship date earlier than its order date is impossible. These rules catch problems that single column checks miss, because each value looks acceptable on its own.

Question 14

Why should a team document the results of a profiling exercise?

Accepted Answer

Documented profiling results create a shared record of what the data looked like at a specific point in time. This record supports root cause analysis and gives future assessments a baseline for measuring improvement. Without documentation, teams often repeat investigations or lose track of problems that were already reported.

Question 15

What does comparing a dataset against reference data reveal during assessment?

Accepted Answer

Comparing a dataset against reference data checks whether its values match an authoritative list, such as approved product codes, postal code tables, or standardized country names. Values that fall outside the reference list reveal misspellings, obsolete codes, and unofficial variants. This comparison identifies accuracy problems that internal checks alone cannot detect.

Question 16

How should a team scope a data quality assessment before profiling begins?

Accepted Answer

A team should begin by identifying the business processes, reports, and decisions that depend on the data, then select the critical data elements that support them. Scope also defines the time period, source systems, and sample size to be examined. A focused scope keeps the assessment manageable.

Question 17

What does data cleansing involve, and what is its main goal?

Accepted Answer

Data cleansing is the process of detecting and correcting errors, inconsistencies, and omissions in a dataset. Its main goal is to make records accurate, complete, and fit for their intended use. Typical actions include fixing typos, removing duplicate rows, filling gaps from trusted sources, and flagging values that cannot be repaired.

Question 18

Why should values be standardized before records are matched for duplicates?

Accepted Answer

Standardization puts values into one consistent form, such as expanding "St." to "Street" and converting all state names to two letter codes. Without it, matching rules treat "Main St" and "Main Street" as different addresses and miss true duplicates. Standardizing first allows the later comparison to work on comparable values.

Question 19

What is parsing, and when is it needed in cleansing?

Accepted Answer

Parsing breaks a combined text value into its component fields, such as splitting a full name into first, middle, and last name, or an address line into street number, street name, and unit. It is needed when source data arrives as free text. Rules should handle common variations in names and addresses.

Question 20

What is a survivorship rule used when merging duplicate records?

Accepted Answer

A survivorship rule decides which value is kept when several duplicate records are merged into one. Common approaches favor the most recently updated value, the value from the most trusted source, or the most complete entry. Clear survivorship rules make merged records predictable.

Question 21

How does a lookup table help standardize values across a dataset?

Accepted Answer

A lookup table maps each known variant of a value to one approved standard value. For example, "NY," "N.Y.," and "New York" can all map to a single code. When new variants appear, staff add them to the table, so the mapping grows over time. Lookup tables are easy to audit.

Question 22

Why should cleansing changes be logged rather than overwritten silently?

Accepted Answer

Logging keeps a record of each original value, the corrected value, the rule or person responsible, and the time of the change. This audit trail lets teams reverse mistakes, explain results to stakeholders, and confirm that corrections follow approved rules. Silent overwrites destroy the evidence needed to investigate problems later.

Question 23

When should a value be flagged for manual review instead of corrected automatically?

Accepted Answer

A value should go to manual review when automated rules cannot determine the correct answer with confidence. Examples include two plausible spellings of a name, a match score that falls near the decision threshold, or a value that conflicts with a trusted source. Human judgment protects against confident but wrong automated changes.

Question 24

What is a common risk of applying cleansing rules without testing them first?

Accepted Answer

A common risk is that a rule changes correct values, damaging data that was fine before cleansing. A rule that strips all punctuation might turn an apostrophe in an organization name into a meaningless string. Testing rules on a sample, comparing before and after results, and reviewing exceptions helps catch such side effects.

Question 25

What does data governance mean for an organization that depends on information?

Accepted Answer

Data governance is the set of policies, roles, standards, and processes that define how data is managed, protected, and used across an organization. It establishes who is accountable for decisions about data and how disagreements are resolved. Good governance gives data quality efforts a durable structure, so improvements outlast any single person.

Question 26

What responsibilities does a data owner typically hold within an organization?

Accepted Answer

A data owner is an accountable business leader who makes final decisions about a dataset, including who may access it, how it is defined, and what quality level is acceptable. Owners approve policies, resolve escalated disputes, and fund improvements. Owners are usually chosen from the business function that uses the data.

Question 27

What does a data steward do that differs from the owner's role?

Accepted Answer

A data steward handles the day to day care of data, applying standards, monitoring quality, and coordinating corrections. Where an owner sets policy and makes final decisions, a steward carries out those decisions and keeps definitions, metadata, and issue logs current. Stewards often notice problems first.

Question 28

Why is a shared business glossary important for data quality?

Accepted Answer

A shared business glossary defines each key business term once, so people across departments interpret data the same way. Without common definitions, two teams may report different revenue totals from the same source and both be technically correct. Glossary entries also make quality rules easier to write.

Question 29

What should a data quality issue log record for each reported problem?

Accepted Answer

An issue log should record a clear description of the problem, the affected data elements and records, the business impact, and the person who reported it. It should also show the assigned owner, the current status, and the resolution taken. It turns scattered complaints into a prioritized work queue.

Question 30

What does a data quality service level agreement typically specify?

Accepted Answer

A data quality service level agreement is a formal commitment between data producers and data consumers that states expected quality targets and response times. It typically specifies which datasets are covered, the dimensions and thresholds that apply, how often quality is reported, and what happens when targets are missed.

Question 31

How does a RACI style responsibility matrix clarify who handles data quality tasks?

Accepted Answer

A responsibility matrix lists each data quality task against the roles involved, marking who does the work, who is accountable for the outcome, who must be consulted, and who must be informed. The format prevents gaps where everyone assumes someone else is handling a problem. It also prevents overlaps where several people duplicate the same effort.

Question 32

What is metadata, and why do governance programs manage it?

Accepted Answer

Metadata is data about data, including field definitions, source systems, lineage, and ownership details. Governance programs manage it because people cannot judge whether a dataset is fit for use without knowing what its fields mean and where they came from. Well maintained metadata supports searching, impact analysis, and faster troubleshooting.

Question 33

How is a data quality metric defined and why is it tracked over time?

Accepted Answer

A data quality metric is a measurable indicator that expresses how well data meets a quality expectation, usually as a percentage, count, or score. Tracking it over time shows whether quality is improving, holding steady, or declining. Good metrics are simple to calculate, tied to a business need, and reported consistently.

Question 34

How is a completeness rate calculated when used as a monitoring metric?

Accepted Answer

A completeness rate divides the number of records with all required fields populated by the total number of records in scope, then multiplies by one hundred to express it as a percentage. For example, if 950 of 1,000 customer records have every required field filled, the rate is 95 percent.

Question 35

Why should monitoring thresholds be set rather than aiming for perfect data?

Accepted Answer

Perfect data is rarely achievable or economical, so monitoring thresholds define the acceptable level of quality for each metric and business use. A threshold turns a number into a decision, showing when a dataset is good enough to use and when an alert should trigger investigation.

Question 36

What are data quality alerts, and what should trigger one?

Accepted Answer

Data quality alerts are automated notifications sent when a metric falls outside its agreed threshold or when a rule fails. An alert should trigger when a change is significant, actionable, and tied to a defined owner who can respond. Alerts that fire constantly for trivial reasons train people to ignore them.

Question 37

What is a root cause analysis in data quality improvement?

Accepted Answer

Root cause analysis traces a data quality problem back to the process, system, or decision that created it, rather than only fixing the symptom. For example, a surge of missing postal codes might trace to a web form that made the field optional after a recent change. Fixing that form prevents future gaps.

Question 38

What does a data quality scorecard communicate to executives and other leaders?

Accepted Answer

A scorecard summarizes key quality metrics for a dataset or domain in a compact view, showing current status, trend direction, and progress against targets. It helps leaders see where quality stands without reading detailed profiling reports. Scorecards work best when they link each metric to a business consequence, such as delayed invoices.

Question 39

How does a feedback loop help sustain data quality over time?

Accepted Answer

A feedback loop routes lessons from monitoring and issue resolution back into source systems, data entry standards, and training. Each cycle of detection, correction, and prevention reduces the number of problems that reach downstream users. Without feedback, teams keep cleaning the same records while the underlying causes remain unchanged.

Question 40

What is continuous improvement in the context of data quality management?

Accepted Answer

Continuous improvement is an ongoing cycle of measuring quality, identifying gaps, making targeted changes, and checking whether the changes worked. Rather than treating quality as a one time cleanup project, it builds routine review into daily operations. Each cycle uses evidence from metrics and feedback to decide the next priority.

Data Quality Management Flashcards

1Core Dimensions of Data Quality

2Data Profiling and Assessment

3Data Cleansing and Standardization

4Governance, Ownership and Stewardship

5Monitoring, Metrics and Improvement

Learning Resources