Digest
The Four Levels of Measurement
Every statistic starts with a choice you rarely think about: what kind of scale is this number actually on? Nominal, ordinal, interval, and ratio data each support a different set of valid operations — mixing them up is one of the quietest ways an analysis goes wrong. Below, each scale is paired with a real dataset from a government or institutional source, not a toy example.
01 / 04 — Nominal
Nominal Scale
Nominal data is just names for groups — labels with no order and no numbers attached. Think "eye color" or "country." The only thing you can really do with it is count how many fall into each group.
Statisticians use it to answer "how many of each?" questions — counts, percentages, and the mode (most common group) — never an average.
The quick test
Could you rearrange these categories in any order without losing information? If yes, it's nominal.
U.S. Billion-Dollar Weather and Climate Disasters, by disaster type (2024)
Full year 2024
| Disaster Type | Separate Billion-Dollar Events (2024) |
|---|---|
| Severe Storm | 17 |
| Tropical Cyclone | 5 |
| Winter Storm | 2 |
| Flooding | 1 |
| Drought / Heat Wave | 1 |
| Wildfire | 1 |
Why this fits nominal data
- "Wildfire," "Flooding," "Winter Storm" are just names — none of them is naturally "before" or "after" another.
- The only valid math here is counting: 17 severe storms happened, 1 wildfire did. You can't average "Wildfire" and "Flooding."
- Swap the row order in the table above — nothing about the data changes. That's the tell.
Source: NOAA NCEI — Billion-Dollar Weather and Climate Disasters
02 / 04 — Ordinal
Ordinal Scale
Ordinal data can be ranked low to high, but the space between ranks isn't a fixed, equal amount. Think "1st, 2nd, 3rd place" — you know the order, not how far apart they finished.
Statisticians use it to compare rank and find the middle value (median) — but never a mathematical average, since the steps between ranks aren't guaranteed equal.
The quick test
Is there a clear order, but the gap between steps isn't a fixed, countable amount? Then it's ordinal.
Worst U.S. Drought Monitor category currently affecting each state
Data valid August 18, 2026
| State | Worst Category Currently Observed |
|---|---|
| New Mexico | D4 — Exceptional Drought |
| Texas | D4 — Exceptional Drought |
| Arizona | D2 — Severe Drought |
| Kansas | D2 — Severe Drought |
| California | D2 — Severe Drought |
| Iowa | D2 — Severe Drought |
| Georgia | D1 — Moderate Drought |
Why this fits ordinal data
- The categories have a real order: None is better than D0, which is better than D1, all the way to D4.
- But "D1 to D2" isn't the same size jump as "D3 to D4" — there's no ruler measuring the distance between drought categories.
- You can say Texas (D4) is worse off than Georgia (D1) — just not "how many times worse," the way you could with a plain number.
Source: U.S. Drought Monitor (NDMC / NOAA / USDA)
03 / 04 — Interval
Interval Scale
Interval data sits on a scale with equal, ruler-like steps, so adding and subtracting makes sense. But zero is just a point on that ruler, not "none of it" — so you can't say one value is a multiple of another. This is also where the name comes from: statisticians sort continuous data like this into equal-width "class intervals" (30–40°, 40–50°, and so on) and count how many observations land in each one.
Statisticians use it to average and compare continuous values meaningfully, and to bucket them into equal-width classes for a frequency table — but never as a ratio, since "zero" doesn't mean "nothing" here.
The quick test
Are the gaps between numbers even and measurable, but "zero" doesn't mean "nothing exists"? Then it's interval.
Central Park, NY — normal monthly average temperature (1991–2020), grouped into 10° classes
30-year normals period: 1991–2020
| Temperature Class (°F) | Months in This Class | Count |
|---|---|---|
| 30 – 40° | Jan (33.7°), Feb (35.9°), Dec (39.1°) | 3 |
| 40 – 50° | Mar (42.8°), Nov (48.0°) | 2 |
| 50 – 60° | Apr (53.7°), Oct (57.9°) | 2 |
| 60 – 70° | May (63.2°), Sep (69.2°) | 2 |
| 70 – 80° | Jun (72.0°), Jul (77.5°), Aug (76.1°) | 3 |
Bars touch — that's histogram convention, because the classes are continuous, equal-width slices of one scale (only possible because this is interval data), not separate categories.
Why this fits interval data
- Every degree Fahrenheit is the same size step, so it's fair to carve the year into equal 10° classes — 30–40°, 40–50°, and so on — the way you would with any interval data.
- 0°F isn't "no temperature," it's just a cold day in North Dakota — an arbitrary marker, not an absence. That's fine for building classes; it just means you can't say one class is "twice as warm" as another.
- Notice New York's months split almost evenly between the cold classes (Dec–Feb) and the hot ones (Jun–Aug), with spring and fall doing the bridging — that shape is the whole point of grouping into classes.
Source: National Weather Service, NWS Forecast Office New York, NY
04 / 04 — Ratio
Ratio Scale
Ratio data is like interval data, plus one thing: zero really means "none of it." That true zero is what makes ratios and multiples meaningful — 200 really is twice as much as 100. Like interval data, it can also be sorted into equal-width classes and counted — the classic ratio-scale example being "how many rivers fall into each size class."
Statisticians use it for the full toolkit — averages, ratios, percent change, coefficients of variation, and equal-width frequency classes — since every kind of arithmetic comparison holds up, all the way down to a true zero.
The quick test
Does zero mean a true, total absence of the thing being measured? Then it's ratio.
Real-time streamflow discharge at eight USGS gauging stations, grouped by size class
August 23, 2026 (provisional, subject to revision)
| Discharge Class (cfs) | Stations in This Class | Count |
|---|---|---|
| 10 – 100 | Peachtree Creek, GA (17.5); Guadalupe R., TX (89.2) | 2 |
| 100 – 1,000 | Missisquoi R., VT (157); Souris R., ND (197) | 2 |
| 1,000 – 10,000 | Connecticut R., CT (3,180); Colorado R., AZ (8,420) | 2 |
| 10,000 – 100,000 | — no stations in this range — | 0 |
| 100,000 – 1,000,000 | Columbia R., OR (106,000); Mississippi R., MO (174,000) | 2 |
These 8 stations happen to split neatly into small creeks and major rivers, with an empty class in between — a real gap this dataset shows, not a rounding artifact.
Why this fits ratio data
- 0 cubic feet per second means an actually dry streambed — a real absence of flow, not an arbitrary reference point like 0°F was.
- Because zero is real, ratios hold up: the Mississippi's ~174,000 cfs genuinely is about 9,900× Peachtree Creek's 17.5 cfs — a comparison that only makes sense with a true zero underneath it.
- The classes below span factors of ten precisely because a true zero lets you multiply your way up the scale — there's no equivalent "×10 class" you could build on the drought-severity ranks.
Source: USGS National Water Information System (NWIS)