NR-542 · Week 4 of 8 · The data quality audit

NR-542 Week 4 The Data Quality Audit: How to Write It

The short answer

NR-542 Week 4 is the audit stage. You take a set of records and interrogate it on named quality dimensions before you interpret anything: completeness, accuracy, consistency, timeliness, validity and uniqueness. The graded requirement is that every defect you find is counted and then followed through to its effect on the answer, because a quality section that lists problems without saying what they do to the conclusion has stopped one step short. Your section may print this as NR 542 or NR542; it is the same course. Chamberlain publishes no syllabi outside Canvas. The placement here is our teaching judgment from the course's catalog arc; your section's rubric decides what your week actually asks.

NR-542 Week 4 grading scale at Chamberlain, the criterion levels this assessment is scored on, from Chamberlain Tutors
How Chamberlain grades NR-542 Week 4, visualized by Chamberlain Tutors.

What NR-542 Week 4 asks for

The fastest hour in any data project is the one where you sort each column and look at the top and bottom twenty values. A date of birth in 1900 surfaces, and it is not a very old patient; it is what a registration system stores when a date is unknown. A weight of 700 appears next to weights around 70, which is a unit error or a slipped decimal. A discharge timestamp lands eleven minutes before its admission timestamp, which means two clocks were involved. Three sex values exist where the field allows two, because a legacy code survived a conversion. None of this is exotic. It is what every real extract looks like on first contact, and the professional habit this stage installs is that you find it deliberately rather than discovering it after you have written a recommendation.

Written work at this stage typically asks you to assess the quality of a data set against recognized dimensions and describe the implications for use. Some sections push toward a remediation or governance angle. The failure that recurs is the reassurance paragraph: a student writes that the data appeared complete and accurate, which is a claim that no check was run, and then proceeds to interpret confidently.

Two disciplines separate a strong audit from a weak one. The first is counting. Every defect gets a numerator, a denominator and a scope: 118 of 1,442 records missing a discharge disposition, concentrated in a single month. The second is consequence. Missingness that is spread evenly across records is a nuisance. Missingness that clusters in one unit, one shift or one type of patient is a bias, and it can move your answer in a direction you can name. Saying which of those you are looking at is the analytic core of this stage.

It is worth being explicit that quality is judged against use. A field populated in only six records out of ten is fatal for a question that depends on it and irrelevant for a question that does not. Papers that grade a data set in the abstract end up producing a report card nobody can act on.

The NR-542 Week 4 method, step by step

Six moves that produce an audit rather than a reassurance.

  1. Profile every field before you judge any of them

    Record count, distinct values, minimum and maximum, count of blanks, and the ten most frequent values. That profile is the raw material for everything else and it takes minutes to produce for a modest set.

  2. Run the dimensions as named checks, one at a time

    Completeness by field, validity against allowed values and plausible ranges, consistency across related fields, uniqueness of the identifier, timeliness relative to the events described. Name the dimension in the sentence where you report the result.

  3. Count every defect and locate it

    How many, out of how many, and where they sit. Location matters more than volume: the same number of missing values is a rounding problem if scattered and a bias if concentrated in one unit or one month.

  4. Test whether missingness is random

    Compare records with the missing field against records without it on one or two other characteristics. If they differ, say so plainly and say which direction it would push your result. This single check lifts a paper out of the middle band.

  5. Decide the disposition of each defect and justify it

    Exclude, retain and flag, correct with a documented rule, or leave and note the limitation. State the rule in words precise enough for someone else to apply and land on the same set.

  6. Write the effect on the answer, not just on the file

    Close each finding with the consequence sentence: with these 118 records excluded the denominator falls and the rate rises by an amount that could plausibly account for the difference the committee is reacting to.

A layout and word budget that turns defects into consequences

Our frame for a data quality assessment, sized for roughly 1,200 to 1,600 words. It is our own outline rather than anything the university publishes, and your week's guide outranks it wherever the two disagree.

SectionWhat belongs in itWord target
Set and intended useThe records, the question they are meant to answer, and the fields that question depends on.150 to 180
ProfileCounts, ranges, distinct values and blanks for the fields that matter, reported plainly.200 to 250
Findings by dimensionCompleteness, validity, consistency, uniqueness and timeliness, each with counts and locations.380 to 450
Missingness analysisWhether absent values cluster, what they cluster with, and the direction of the likely bias.200 to 250
DispositionsThe rule applied to each defect class, stated so another person could reproduce your final set.200 to 250
Effect on the answerWhat the cleaned set can now support, what it still cannot, and how much the answer moved.180 to 220

Evidence craft for a quality audit

Use a published set of quality dimensions and name it. Data quality frameworks in health information management are documented and attributed, and adopting one by name makes your checklist defensible instead of personal. Give the issuing body and the year.

Report every defect as a fraction, never as an impression. Ninety-four of 1,442 records carrying a birth date of 1 January 1900 is a finding. Some implausible dates were present is a note to yourself. This is the habit the whole course is built to install.

Distinguish an implausible value from an impossible one. A weight of 700 pounds is implausible and occasionally real. A discharge before an admission is impossible and tells you about clocks or about a data merge. Handling them identically is a reasoning error a grader can see.

Document your cleaning as a reproducible rule. Excluded records with a discharge timestamp earlier than the admission timestamp, 11 of 1,442, is reproducible. Removed obvious errors is not, and in a course about data management the difference is the point.

Five mistakes that cost points in this week's territory

  • The reassurance paragraph. Stating that the data appeared accurate announces that no check was performed.
  • Defects listed without counts. Some missing values and a few duplicates cannot be weighed by anyone, including you.
  • Missingness assumed random. The clustering check is quick, and skipping it leaves the most important finding in the file undiscovered.
  • Silent cleaning. Rows removed without a stated rule make the entire result impossible to reproduce or defend.
  • No consequence sentence. A defect that is never followed through to the answer is trivia, however precisely it was counted.

Before you submit

  • The intended use of the set is stated before any quality judgment
  • A named quality framework governs the dimensions you check
  • Every defect carries a numerator, a denominator and a location
  • Missingness is tested for clustering, with the direction of bias named
  • Each disposition is written as a rule another person could reproduce
  • Each finding closes with its effect on the answer

Auditing a data set for NR-542?

Send the file or scenario and the rubric from Canvas. A premium original draft comes back in 24 to 48 hours with every defect counted, located and followed through to its effect on the answer, and revisions run until the grade lands.

Questions students ask about this stage

The data set my section supplied looks clean. What do I write about?
Teaching sets are usually cleaner than reality and they are almost never spotless. Profile every field anyway and report the profile, because a documented absence of defects is itself a finding when it is supported by counts: all 1,200 records carry a value in every required field, identifiers are unique, and no timestamp pairs are inverted. Then look harder at the places tidy sets still fail. Check whether categories are used consistently rather than merely present, whether a default value appears far more often than plausible, whether every record covers the full date range the file claims, and whether the row count matches what the source should have produced. Finish with fitness for purpose, which is a genuine judgment even about a clean set: clean and complete is not the same as sufficient for the question.
How do I handle protected information if I use data from my workplace?
The short answer is that you should not be moving workplace data into coursework at all. Do not export records, do not attach extracts to a submission, and do not send patient-level information anywhere for a class. If your setting has a formal process for approving the use of de-identified data for education and you have gone through it, follow that process exactly and say in the paper that the data were used under it. In every other case, use a public data set with a published data dictionary, a practice set supplied by your section, or a constructed example you build yourself and label as illustrative. A constructed set has a real advantage for this stage: you can place specific defects in it deliberately and then audit them, which demonstrates the method cleanly.
Should I fix the problems I find or just report them?
Report all of them, act on some of them, and be explicit about which is which. An audit section that quietly repairs values leaves the reader unable to tell what the original set looked like, and a section that repairs nothing produces a paper that cannot proceed to analysis. The disciplined pattern is to report the defect with its count, state the disposition you chose, give the rule, and carry a running record of how the set changed: 1,442 records received, 11 excluded for inverted timestamps, 94 retained with birth date flagged as unknown, 1,431 analyzed. That short lineage is worth including as its own short paragraph or small table, and it is the single clearest signal to a grader that you understand data management as a documented process rather than as tidying.

Keep going

Online now