Coverage, quality and privacy
4 min read
Lesson objectives
- Audit missingness and coverage, then select an appropriately aggregated release.
Please sign in to use all the tools
Once you sign in, the reading tools are yours: larger or smaller text, typeface, line spacing, focus mode, highlighting and bold, and notes saved to your account and gathered in My study.
Coverage, quality and privacy
Separate three quality questions. Did the source include the intended people? Are fields complete for those included? Do recorded values match the question's definition? If age is present in 95% of service register rows, this says nothing by itself about coverage of 95% of district residents. The first denominator is register rows; the second would be the district population. They can differ sharply. A blank value also differs from zero. Unknown age is not infancy, and no registered event does not always mean no event occurred if reporting is incomplete. Write numerator and denominator for each quality percentage.
Releasing fine cross tabulations of age, place and a rare characteristic can make individuals identifiable even after names are removed. This lesson cannot impose a universal cell suppression threshold. Follow the responsible data policy, minimize detail and balance analytic need against disclosure risk. A broader age band or geography may allow a safer display, while also concealing an important disparity. Record that tradeoff. Removing names alone is not a sufficient privacy argument. The invented exercise data do not authorize using actual personal records.
A compact quality log states the population definition, source, known or unknown coverage, variable completeness, rule for conflicting values and action before release. If 900 of 1,000 registered people have a recorded age, age completeness is 90% among register records. Calling it 90% of residents requires evidence that the register covers residents. Before filling missing values, look for a pattern: if blanks cluster in one district, an overall completion percentage may hide more serious bias in the age profile. Disclose measurement limits rather than silently imputing values.
Distinguish unit nonresponse from item nonresponse. In the first, an entire person or household is absent and source coverage may change. In the second, the person appears but a field such as age is blank. An age completion percentage among existing rows cannot repair absent people; they are missing from its denominator. If implausible ages come from a specific entry process, inspect the source before deletion. Removing a patterned group can distort the resulting distribution.
Lesson summary
Field completeness in a register is not population coverage; distinguish missing units from missing items and review small-cell disclosure.