An observational unit answers a deceptively practical question: what does
one row stand for? In Juniper's settlement extract, one row is one qualifying
invoice. days_to_settlement is a variable recorded about that invoice; the
number 31 is one observed value. Neither the column nor the value is the unit.
Grain changes the question
An accounting system may contain invoice headers, invoice lines, payments, cash-application events, customer accounts, and reporting periods. Joining those tables can produce several rows for one invoice. If an analyst averages the joined rows as though each were a separate invoice, invoices with more lines or payments receive more weight. Correct arithmetic then answers a different question.
A finance file has the same problem. One row might be a security, one daily return for a security, one portfolio-date position, or one trade. “Average return per row” has no stable meaning until the unit and weighting rule are known.
Audit the row before the statistic
Name the intended unit, identify its stable key, count duplicate keys, and ask whether any unit can generate multiple records. Then inspect whether the file mixes units—for example, invoice rows with customer-level credit limits or monthly balances. Aggregation may be appropriate, but it must be explicit.
Juniper's opaque invoice ID identifies the unit; it is not a quantitative measurement. If two rows share that ID, the source needs reconciliation before either row is counted. A clean mean cannot repair a duplicated case.
Put the concept to work
Understand this concept
- Distinguish the case represented by a row from the variables, values, identifiers, and repeated records attached to that case.
Analyze this concept
- Audit a business-data extract for its intended row grain, nested units, repeated cases, and duplicate-counting risk before summarizing it.
Learning resources
Choose a lesson, try an application, or inspect the sources behind this concept.
Build on these ideas
- Observational unit — Understand
To analyze this concept: Required. A grain audit applies the distinction between a case and the fields recorded about it.
Lessons
Worked examples and cases
Practice
Common mistaken ideas
Sources
Related concepts
Show 1 more related concepts
Use this idea next
- Observational unit — Analyze
Required level here: understand. Required. A grain audit applies the distinction between a case and the fields recorded about it.
- Outlier — Analyze
Required level here: analyze. Helpful. Duplicate and wrong-grain records are possible sources of extreme observations.
- Population — Understand
Required level here: understand. Required. A population is a set of a defined kind of observational unit.
Show 2 more next steps
- Sampling frame — Analyze
Required level here: analyze. Required. Unique-unit and duplicate checks depend on the intended row grain.
- Variable — Understand
Required level here: understand. Required. A variable is recorded about a defined observational unit.