In this topic 6 sections
Judge information against its intended use
Information is useful when it helps the intended user decide. Accuracy matters, but an accurate figure for the wrong population or period can still mislead. In the guide's measurement sense, validity asks whether a measure represents the concept being examined. Reliability concerns dependable, consistent measurement. A repeatable process can consistently measure the wrong thing, so the two tests are complementary.
Other characteristics make evidence usable. Relevance connects it to the choice; faithful representation avoids a distorted account. Comparability requires compatible definitions and periods. Verifiability allows checks against records or a reproducible calculation. Timeliness, clarity, understandability and accessibility determine whether people can act on it. Completeness matters, but every possible detail is rarely necessary. For performance evaluation, distinguish outcomes a manager can influence from external effects.
Terminology varies: data-processing frameworks may use validity more narrowly for conformity with expected formats or ranges. A valid date field does not prove that the date is correct.
Concept sources: IFRS Foundation, Conceptual Framework Project Summary, March 2018, page 6, qualitative characteristics of financial information; UK Government Data Quality Framework, December 2020, fitness for purpose and technical validity. The financial-reporting characteristics are useful analytical prompts here, not a mandatory framework for internal reports.
Examine quality from four perspectives
Wang and Strong's 1996 research broadens assessment beyond accuracy. Its four categories can be applied as successive questions rather than a single overall score.
- Intrinsic
- Are the observations accurate and objective, and is there reason to believe the data and its source? Reputation supports investigation; it cannot replace checking.
- Contextual
- Does this information add value for this task, with relevant coverage, adequate completeness, suitable timing and an appropriate amount of detail?
- Representational
- Can users interpret it easily, with consistent definitions and concise presentation? Explain units, labels and the meaning of an unusual result.
- Accessibility
- Can authorised people obtain the information when needed while access security protects it from inappropriate use?
A dataset can satisfy one category and fail another. A dependable national series may have excellent intrinsic quality but insufficient local detail for a branch staffing decision. Conversely, a highly relevant internal estimate may require qualification because supporting observations are incomplete.
Concept sources: Wang and Strong, Beyond Accuracy: What Data Quality Means to Data Consumers, 1996, journal pages 19–22 — four categories and their underlying attributes.
Turn uncertainty into explicit reporting choices
Separate recorded facts, estimates, assumptions and forecasts. State missing coverage and explain whether it could change the recommendation. Use ranges or scenarios where a precise-looking point estimate would conceal uncertainty. Do not infer causation merely because two measures move together.
Define performance indicators clearly. SMART prompts ask whether an objective is specific, measurable, achievable, relevant and time-based. Add an agreed calculation and data owner. An easily measured activity may still be a poor indicator of the desired outcome.
Concept sources: Oracle Data Mining Concepts, 12c Release 1, May 2017, chapter 1, What Data Mining Can and Cannot Do — predictive associations are not causal explanations, and results involve probabilities.
Understand what the information platform does
A transaction system records daily activity. A data warehouse integrates selected information for historical analysis, often through extraction, transformation and loading. Consistent identifiers and units make sources comparable. Preserving history supports trend analysis; it does not mean that loading, controlled corrections or refresh processes cease.
Business intelligence combines data, analytical processes and presentation tools to help users understand performance. Reports and dashboards communicate results; an executive information system concentrates on senior managers' information needs. Data discovery supports interactive exploration through filters and visual comparisons.
Online analytical processing, or OLAP, provides several views of measures across dimensions. Consolidation rolls detail into summaries; drill-down moves towards detail; slicing and dicing selects or rearranges combinations such as location, service and period. These operations change the view, not the underlying truth of the data.
Concept sources: Oracle Database Data Warehousing Guide, 12c Release 2, December 2020, chapters 1 and 7 and glossary entries for drill and slice and dice — warehousing, analysis and refresh.
Use mining and big data with a business question
Data mining searches for useful patterns, groups and predictive relationships. Preparation, model evaluation and interpretation remain necessary; software does not decide whether a relationship makes commercial sense.
Big data concerns datasets whose scale and characteristics require suitable analytical architectures. Volume, velocity and variety are familiar features; NIST also discusses variability. Sources can include structured transactions and less structured text or sensor records. More data creates opportunities alongside classification, storage, retention, privacy and quality challenges. Analysis can support strategy formation as well as implementation.
- State the business question
- Assess and prepare data
- Select an analytical method
- Test the findings
- Explain uncertainty
- Act and monitor outcomes
Concept sources: Oracle Data Mining Concepts, May 2017, chapter 1; NIST Big Data Interoperability Framework, Volume 1, Version 3, October 2019, sections 2–3, definitions and characteristics.