Measurement Scales & Data Types
A measurement scale determines which relationships exist between values and which operations preserve their meaning. Data type determines how those values are represented, stored and processed. Neither is formatting: both constrain valid analysis.
Each scale grants a different set of mathematical permissions.
The familiar hierarchy is cumulative: ordinal values add order to identity; interval values add equal distance; ratio values add a meaningful zero. A richer scale supports more operations, but only when the property truly satisfies its assumptions.
Different or equal
Values label categories. Categories may be counted and compared for equality, but they have no inherent order or numeric distance.
device = mobile
result_type = organicVALID → MODE, FREQUENCY, PROPORTIONHigher or lower
Values have a meaningful order, but the distance between adjacent positions is unknown or unequal.
priority = low < medium < high
rank = 1st, 2nd, 3rdVALID → MEDIAN, PERCENTILE, ORDERMeaningful differences
Equal numeric intervals represent equal differences. Zero is an arbitrary reference point, so ratios are not meaningful.
index change: 40 → 50 = +10
50 is not 25% “more property”VALID → MEAN, SD, DIFFERENCEMeaningful ratios
Equal intervals combine with a non-arbitrary zero representing absence of the measured quantity.
400 ms = 2 × 200 ms
0 qualifying links = none observedVALID → ALL ARITHMETIC WITH CONTEXTThe operation must respect the structure of the scale.
This matrix shows the default permissions. Domain-specific models may justify additional operations, but the justification belongs to the model—not to the numeric appearance of the data.
Select a scale and operation. Test the claim before calculating it.
The engine demonstrates why valid syntax is not the same as valid measurement. Software can divide any two numbers; the scale decides whether the result has meaning.
Scale permission test
Choose a measurement scale and intended operation. The output reports the strongest statement supported by that combination.
Nominal values support equality, category counts, proportions and mode. Their codes do not support order, distance or ratios.
Scale describes meaning. Data type describes representation.
A ratio-scale duration may be stored as an integer, decimal or timestamp difference. A nominal category may be stored as text or an encoded integer. Storage type alone does not reveal measurement meaning.
present / absent
unordered states
ordered bands
integer events
proportion
interval
Every transformation preserves some meaning and discards something else.
Transformations are not neutral cleanup. Binning removes distance, normalization changes the reference frame and aggregation hides individual variation.
Page response times across repeated requests.
[180, 220, 240, 900] msUnits aligned; failed captures kept as explicit missing states.
valid n=4 / missing n=1Median and percentile preserve skew better than one mean.
median=230 / p95=801Values expressed against a declared baseline or eligible range.
z=(x−μ)/σContinuous distance is compressed into an ordered category.
fast / moderate / slowZero, absent, unknown and inapplicable are different data.
Collapsing these states can change denominators, bias averages and manufacture trends. Missingness must be modeled before calculation.
Measured absence
The procedure ran successfully and observed none of the defined quantity.
qualifying_links = 0
capture_status = completeNot established
Evidence is insufficient to decide whether the property is present or absent.
value = null
reason = unresolved_identityOutside the rule
The property is not defined for this object or condition.
value = N/A
reason = ineligible_objectNot observed
The intended measurement failed. Failure describes the procedure, not the object.
value = null
reason = capture_failedMore digits can display more certainty than the measurement contains.
Reporting precision should reflect resolution, repeatability and uncertainty. Decimal places are a formatting decision—not evidence of accuracy.
Choose displayed precision
The underlying illustrative estimate is 64.237 with uncertainty of ±4.8. Increase the decimal places and observe that uncertainty does not shrink.
One decimal aligns with the stated uncertainty. Additional digits would imply resolution not supported by the measurement.
Scale and type determine what each value can legitimately say.
These examples connect abstract scale theory to search, assets, links, retrieval and temporal observation.
Device class
Mobile, desktop and tablet are nominal categories.
Search position
Organic rank is ordered, but position gaps do not guarantee equal exposure differences.
Visibility index
A normalized score may behave as interval-like only within its declared model.
Response time
Milliseconds have equal intervals and a meaningful zero duration.
Entity coverage
A bounded proportion has a meaningful zero and a fixed eligible denominator.
Entity relation
A subject–predicate–object record is structural data rather than one scalar value.
Twelve checks before values enter analysis.
These controls preserve semantic meaning from acquisition through storage, transformation and reporting.
Choose scale from the property, not the available format.
State nominal, ordinal, interval or ratio assumptions.
Name category, count, proportion, duration or relation.
Use representation that preserves range and precision.
Store milliseconds, percent, count or index identity.
Declare valid values, impossible values and limits.
Separate zero, unknown, failure and not applicable.
Apply calculations supported by scale structure.
Make binning, normalization and aggregation visible.
Align displayed digits with uncertainty and resolution.
Do not merge values from incompatible scale models.
State what the scale cannot establish.
Scales and types, fully resolved.
What is a measurement scale?
A measurement scale defines the meaningful relations between values—identity, order, equal distance and true zero—and therefore constrains valid mathematical operations.
What is the difference between scale and data type?
Scale describes the empirical meaning of values. Data type describes their logical, structural or machine representation. An integer can encode a count, rank or category code, each with different meaning.
Can ordinal values be averaged?
A mean of ordinal codes requires an assumption that adjacent categories have sufficiently equal distances. Without that assumption, median, percentile and distribution are safer summaries.
Is a score from 0 to 100 a ratio scale?
Not automatically. A zero score may be a model boundary rather than absence of the property, and 80 may not represent twice as much as 40. The construction rule determines scale behavior.
Why is zero different from missing?
Zero records an observed absence of quantity. Missing means the value was not established. Treating missing as zero adds unobserved negatives and can bias the result.
Does converting text categories to numbers change their scale?
No. Numeric encoding improves storage or processing but does not create order, distance or ratios that the underlying categories do not possess.
Continue through the complete measurement system.
The next node tests validity: whether a measurement actually represents the property its interpretation claims.
Objects, properties, rules, values and uncertainty.
OPEN NODE →MSR / 02OBJECTMeasurement Objects & UnitsUnits of analysis, identity and measurable properties.
OPEN NODE →MSR / 03REPRESENTATIONMetrics, Indicators & ProxiesDirect values, derived measures and proxy limits.
OPEN NODE →MSR / 04RULEOperational DefinitionsTurning concepts into observable procedures.
OPEN NODE →Categories, order, distance, ratios and permitted operations.
CURRENT NODEWhether a value represents the intended property.
OPEN NODE →MSR / 07RELIABILITYMeasurement ReliabilityConsistency across repeated comparable conditions.
OPEN NODE →MSR / 08ERRORMeasurement Error & UncertaintyVariation, error sources, ranges and limits.
OPEN NODE →MSR / 09NORMALIZENormalization & ComparabilityMaking unlike observations responsibly comparable.
OPEN NODE →MSR / 10REFERENCEBaselines, Benchmarks & ThresholdsReference states and decision boundaries.
OPEN NODE →MSR / 11TIMETemporal Measurement & ChangeWindows, cadence, drift and comparable change.
OPEN NODE →MSR / 12SIGNALFrom Measurement to SignalWhen a measured difference becomes analytically relevant.
OPEN NODE →