SOAD2021 Chap.3 Data Literacy and Public Data
Data Literacy and Public Data
Data literacy means more than reading a number. It requires knowing who or what was counted, how concepts were operationalised, when data were collected, what denominator was used and which people or events are absent. Public datasets are constructed records, not transparent copies of social reality.
Begin interpretation with provenance and metadata.
Identify the collecting body, purpose, unit, population, geography, time period and revision status. Then inspect rates, counts and percentages. A count describes volume; a rate relates events to a population at risk; a percentage depends on a clearly named denominator. Comparisons fail when definitions or denominators shift.
Responsible use also examines missingness, small cells, aggregation and uncertainty.
Group averages can conceal unequal experiences, while fine categories may threaten confidentiality or yield unstable estimates. Convert the final claim into four parts: observed pattern, comparison, plausible interpretation and limitation. That structure keeps data useful without claiming that description alone explains cause.
A sound interpretation begins before calculation.
Draw a unit map showing the event, person, household, service episode or geographic area represented by each row, because mixing units can create convincing but false comparisons. Check whether repeated contacts belong to the same person and whether the dataset counts eligibility, referral, attendance or completion.
After calculating, perform a sensitivity check: ask whether the pattern survives a plausible alternative denominator, an excluded group or a revised classification. Communication should preserve that reasoning. A chart title can state the observed pattern, but the accompanying text must name the population and period, explain an important limitation and avoid causal verbs.
When disaggregation is relevant, weigh the benefit of revealing unequal experience against unstable estimates and identification risk. The goal is evidence that supports a decision without making the dataset seem more complete than the social process it records.
What this chapter covers
- 01
Data provenance
- 02
Metadata
- 03
Counts and rates
- 04
Denominators
- 05
Missing data
- 06
Comparison
- 07
Responsible communication
Worked example · free
Worked example: interpret changing referral completion
- 1Calculate last year's completion proportion: 72 divided by 120 equals 0.60 or 60 percent.
- 1Calculate this year's completion proportion: 90 divided by 180 equals 0.50 or 50 percent.
- 1Distinguish the higher count from the lower proportion and state both observations.
- 1Request definition, population and missingness information before explaining the change or assigning cause.
Key terms
- Denominator
- A denominator is the reference total used to construct a fraction, rate or percentage. It determines the population to which the numerical statement applies.
- Metadata
- Metadata are information describing how data were defined, collected, classified, stored and revised. They are necessary for evaluating comparability and fitness for purpose.
- Missingness
- Missingness is the pattern and process through which expected observations are absent. Its consequences depend on why data are missing and whose records are affected.
Data Literacy and Public Data FAQ
Why can a count and rate move in opposite directions?
A count can increase because the underlying population or opportunity for events increased, while the rate decreases because the event grew more slowly than its denominator. Report both when they answer relevant questions, and never compare percentages without naming their denominators.
What should I check before comparing public datasets?
Check concept definitions, collection purpose, population coverage, unit of analysis, period, geographic boundary, classification changes and revision notes. Two tables with similar labels may not measure the same thing, and apparent change may be produced by a changed definition.
Does a visible trend show that a program caused change?
No. A trend is a descriptive pattern across observations. Causal interpretation needs a design that addresses timing, alternative explanations, selection, measurement changes and other influences. Use the trend to refine questions and comparisons rather than treating it as proof.
How should uncertainty be communicated?
Name the source of uncertainty and explain its decision consequence. Sampling error, small cells, missing records and uncertain classification are different problems. Give precise figures where warranted, but avoid extra decimal places that imply more measurement certainty than the data support.
Assessment move
Choose one public table and make a data biography: collector, purpose, unit, population, period, geography, definitions, missingness and revisions. Recalculate one rate from its numerator and denominator, then write one supported claim and one claim the data cannot support. Create a contrast by changing the denominator and explain why the interpretation changes.
Repeat with a chart until you can translate visual patterns back into precise numerical statements. Add a sensitivity note that identifies one plausible classification change and predicts whether it would strengthen, weaken or leave the observed pattern unchanged. Rewrite the chart caption so it names the unit, period, comparison and one interpretive limit without assigning a cause.
