An independent study reference written by Dr Phuc V. Nguyen. It is not official subject material — for assessment requirements always follow your subject outline and vUWS.
Data quality as fitness for use
Data quality is fitness for use, not an absolute property of a dataset. The same records can be entirely adequate for one decision and unusable for another, so the question is never whether data is good but whether it is good enough for this purpose. This framing, established in the information quality literature, puts the person accountable for the decision in charge of the standard rather than the person who built the pipeline. It also makes quality work finite. You stop when the data supports the decision at an acceptable risk, rather than when the data is perfect, because perfect has no definition.
Try it yourself
Quality is a relation between data and a purpose. Two uses below carry an explicit contract of fields and thresholds. The same working table is measured against both, and each verdict is read off its contract rather than asserted.
| Contract clause | Required | Measured | Result |
|---|---|---|---|
| Validity, postcode (order records) | at least 99.0% | 100.0% | pass |
| Validity, order amount (order records) | at least 99.0% | 100.0% | pass |
| Validity, order date (order records) | at least 99.0% | 100.0% | pass |
| Order value held in quarantine | at most 5.0% | 3.1% | pass |
| Contract clause | Required | Measured | Result |
|---|---|---|---|
| Completeness, mobile number (customer table) | at least 99.5% | 91.7% | fail |
| Uniqueness, customer id (customer table) | at least 99.5% | 91.7% | fail |
Why it matters
A street address accurate to the suburb is fine if you are deciding where to open a store. It is useless if a driver has to deliver a parcel this afternoon. Nothing about the data changed between those two sentences. The purpose changed. So the sensible question is never whether the data is clean, but clean enough for what. That is also the only honest way to finish a cleaning project, which otherwise runs until the budget does.
Two teams disagree about whether a customer table is of acceptable quality. What is the most useful next question?
Worked examples
A team spends three months lifting customer email deliverability from 91 per cent to 97 per cent before launching a campaign. Halfway through, the marketing lead asks whether the effort was needed.
It depends entirely on what the campaign decides. For a broadcast newsletter, six points of extra reach is worth something, and the campaign would have run at 91 per cent. For a regulatory notice that must reach every affected customer, 97 per cent is still a failure, and the correct response is a second channel rather than a better email list. Naming the decision first converts an open-ended cleaning project into a target with a stopping rule.
An analyst refuses to release a demand forecast because 4 per cent of transactions have a missing store code.
The right question is what the forecast drives. If it sets national purchasing volumes, missing store codes barely matter, because the totals are unaffected and the work can proceed with a note. If it sets store-level rosters, 4 per cent of transactions unattributed could shift staffing in a small store by a meaningful amount, so the gap has to be closed or the affected stores flagged. One defect, two conclusions, decided by the use rather than by the defect rate.
Common mistakes
- ✗Data quality is a property of the dataset. It is a relation between the data and a purpose, which is why the same table can pass for one report and fail for another.
- ✗Higher quality is always worth pursuing. Improvement costs money and time, and past the point where the decision would not change, further spending buys nothing.
- ✗The data team decides when data is good enough. The accountable decision owner sets the threshold, because only they know what an error would cost.
- ✗Data that is fit for its current use will stay fit. New uses arrive constantly, and a dataset built for billing is routinely repurposed for analysis with quality assumptions nobody rechecked.
Revision bullets
- •Quality is fitness for use, judged against a stated purpose
- •The same dataset can be fit for one decision and unfit for another
- •The decision owner sets the threshold, not the pipeline builder
- •A stated purpose gives a cleaning project a stopping rule
- •Repurposed data inherits assumptions that were never checked for the new use
Quick check
Two teams disagree about whether a customer table is of acceptable quality. What is the most useful next question?
A dataset built for invoicing is reused to model customer churn. Which assumption most deserves rechecking?
Connected topics
More in Data Foundations
Sources
- Wang & Strong (1996)Wang, R. Y., & Strong, D. M. "Beyond Accuracy: What Data Quality Means to Data Consumers." Journal of Management Information Systems, 12(4), 5-33, 1996.Establishes data quality as fitness for use by data consumers and derives its dimensions from what consumers say they need.
- Juran & Godfrey (1999)Juran, J. M., & Godfrey, A. B. (eds.) Juran's Quality Handbook, 5th ed. McGraw-Hill, 1999.Source of the fitness-for-use conception of quality that the data quality literature adapted.
- Office of the Australian Information Commissioner. Australian Privacy Principles, made under the Privacy Act 1988 (Cth).APP 10 requires personal information to be accurate, up to date and complete having regard to the purpose of its use, which is fitness for use written into law.