Your starting point
Prerequisites: Use the ideas and vocabulary from lesson 5; revisit its worked examples if needed.
Definition: Data quality is fitness for a particular use. Completeness concerns whether needed values are present; validity concerns format or allowed values; consistency concerns agreement; uniqueness concerns unintended duplication; accuracy concerns agreement with reality.
A tidy table can still be wrong
A column can contain only valid numbers and still measure the wrong quantity. A support duration recorded in seconds but labeled minutes passes a numerical type check while producing a misleading result. Accuracy often needs a source comparison, instrument check or domain knowledge. Automated checks identify symptoms; they cannot establish that every observation is true.
Prioritize errors by consequence
A missing optional comment may not affect a revenue total. A missing quantity can. Identify the fields necessary for the specific calculation and count the records affected. Report both the number missing and the denominator. Decide whether to exclude, investigate or substitute values based on the meaning of missingness, not on the desire to make a chart look complete. Keep the original table and describe the decision.
Build a small quality report
Start with row and column counts, expected grain, missingness by field, valid ranges, repeated keys and obvious inconsistencies. Inspect unusual records rather than automatically removing them. Repeat the checks after cleaning, because a transformation can introduce new problems. A defensible report states what was checked, what changed and what remains uncertain. Quality is not a single universal score that makes all uses safe.
Two worked examples
Missing hours
A fictional billing table has ten records; two have blank hours. Missingness is 2 ÷ 10 = 20% of records. Replacing those blanks with zero would assert no work occurred, which the source does not establish. A revenue summary should disclose the two excluded calculations and investigate their importance.
Valid but inconsistent
A task has planned_hours = 4 and actual_hours = 40. Both values are valid nonnegative numbers. The large difference is worth checking, but it may represent real rework. Compare the task record or ask its owner before changing 40 to 4. A range check flags the question; it does not answer it.
Where this becomes useful
A small business wants to summarize invoices. Inconsistent payment labels, a repeated invoice and missing hours affect different decisions: grouping, totals and billable value. Use targeted checks rather than deleting every unusual row.
Make an attempt
Load Small business operations in the Cleaning Workbench. Inspect the repeated record, whitespace and blank hours before applying changes. Write a before/after row count and explain which problem each transformation addresses. Do not fill hours unless you can justify the replacement.
Misconception check
Claim: Cleaning means making every row look normal.
Inspect the correction
Cleaning addresses documented problems. Legitimate unusual observations and uncertainty must not disappear merely to simplify a story.
Independent application
Create a quality checklist for a non-sensitive table: expected grain, required fields, units, valid ranges and duplicate rules. Run the checks, retain the source, and write a short limitation statement that distinguishes verified corrections from unresolved questions.
Completion records practice and self-review. It is not certification of statistical competence.
