Contents/ Part I · Foundations/ Chapter 5

Levels of Measurement

Not all numbers are created equal. The level of measurement tells you which math is meaningful, and saves you from "facts" like "40°C is twice as hot as 20°C."

⏱️ ~11 min read
🐍 Notebook included
📊 Chapter 5

In the Types of Data & Variables chapter we met nominal and ordinal categories. Now we sharpen the whole idea into the four classic levels of measurement, a ladder, often remembered as NOIR, where each rung carries more information and unlocks more math.

🪜
The four levels are Nominal, Ordinal, Interval, and Ratio. Each one adds a property: order, then equal gaps, then a true zero, and with each property comes a new operation you're allowed to perform.
1

The Measurement Ladder

Climb from the least to the most informative. Each step keeps everything below it and adds one new power:

1 · NOMINAL 2 · ORDINAL 3 · INTERVAL 4 · RATIO names / labels + order + equal gaps + a true zero more information →

Memory hook: N-O-I-R, Nominal, Ordinal, Interval, Ratio.

2

The Categorical Levels: Nominal & Ordinal

1Nominal

Pure labels with no order. You can only check if two values are the same or different.

  • Examples: blood type, country, gender, eye color
  • Allowed: = / ≠ (categorize, count)
  • Center: mode only
2Ordinal

Ordered categories, you know the rank, but not the distance between ranks.

  • Examples: S/M/L, 1–5 star ratings, education level
  • Allowed: =/≠ and </> (rank)
  • Center: median (or mode)
📏
Order without distance

With ordinal data you know 5 stars beats 4 stars, but not that the jump from 4→5 equals the jump from 1→2. That's why averaging ordinal codes can mislead; the median is safer.

3

The Numerical Levels: Interval & Ratio

Both have equal, meaningful gaps between values. The one thing that separates them is a true zero:

3Interval

Equal gaps, but the zero is arbitrary (it doesn't mean "none").

  • Examples: temperature °C/°F, calendar years, IQ scores
  • Allowed: + / − (differences are meaningful)
  • But NOT ×/÷: "twice as hot" is meaningless
4Ratio

Equal gaps and a true zero (zero means "none of it").

  • Examples: height, weight, age, income, Kelvin
  • Allowed: everything, incl. × / ÷ (ratios)
  • "Twice as heavy / half the income" all valid
🌡️
The true-zero test

Ask: does zero mean "none of the thing"? 0 kg = no mass → ratio. 0°C is just freezing water (there's still plenty of heat) → interval. This is exactly why 40°C is not "twice as hot" as 20°C, but 40 kg really is twice 20 kg.

4

The Whole Picture in One Table

LevelOrdered?Equal gaps?True zero?Valid centerExample
NominalNoNoNoModeBlood type
OrdinalYesNoNoMedianS / M / L
IntervalYesYesNoMeanTemperature °C
RatioYesYesYesMeanWeight (kg)

Level also guides the chart (tying back to the Types of Data & Variables chapter): nominal and ordinal data suit bar charts, while interval and ratio data suit histograms and box plots.

🧭
Why it matters

The level decides your valid average, the charts that make sense, and which tests and models you may use. Match the statistic to the level and your analysis stays honest.

🤔
Can you average ordinal data? (a real debate)

By the strict rules above, ordinal data gets a median, not a mean, because the gaps between "Low," "Medium," and "High" may not be equal. In practice, though, researchers routinely average rating scales (1 to 5 "Likert" items), and it often works fine. Treat the rules here as a strong default, not an unbreakable law: report the median when in doubt, but know that averaging ordinal codes is common and frequently defensible.

📝
A useful map, not the final word

This four-level scheme (from psychologist S. S. Stevens, 1946) is a great teaching ladder, but experts have long debated it. The same variable can even sit at different levels depending on how it is used, and a famous parable ("football jersey numbers") shows that context, not the numbers themselves, decides what's meaningful. Use it as a guide for picking sensible statistics, not a rigid rulebook.

🐍

Bring it to life in Python

The companion notebook tags each column with its level, proves why "twice as hot" fails the true-zero test, picks the right average per level, and maps the operations each level allows.

📓 View Notebook (code & outputs) ▶ Open in Colab ⬇ View / Download on GitHub

View opens the rendered notebook instantly (no setup). Open in Colab runs & edits it live in your browser. To run locally, install numpy, pandas, matplotlib and launch jupyter notebook.

🎓 Key Takeaways

  • NOIR ladder: Nominal → Ordinal → Interval → Ratio, each adding more information.
  • Nominal = labels (mode); Ordinal = order without distance (median).
  • Interval vs Ratio hinges on a true zero, only ratio supports "twice / half / ×".
  • 0°C ≠ "no heat" (interval), but 0 kg = no mass (ratio), the true-zero test.
  • Allowed operations are cumulative, and the level dictates your valid average, chart, and test.
5

Practice Challenges

Four short challenges on the four levels. Beginner-friendly, try them on paper or in Python.

1

Name the level

Give the level (N/O/I/R) for each: blood type, exam grade (A–F), temperature °C, weight kg, finishing place (1st/2nd/3rd), temperature in Kelvin, calendar year.

Hint: two of these are tricky, think hard about Kelvin vs. Celsius, and what year "0" means.
2

Which support "twice as much"?

Of temperature °C, height, income, IQ score, which allow a valid "twice as much" statement (i.e., are ratio scales)?

Hint: apply the true-zero test to each.
3

Pick the valid average

Which center (mode, median, or mean) fits each: favorite color, satisfaction (Low/Med/High), annual salary?

Hint: match the average to the level, mode, median, mean as you climb the ladder.
4

Spot the error

An analyst reports the "average zip code" as 60412 and the "mean satisfaction" as 3.4 (from Low=1…High=5). What's wrong with each, and what should they report instead?

Hint: what level is a zip code, really? And are the gaps between satisfaction levels equal?
Check your work

A fully-worked solutions notebook walks through all four challenges in the same visual style, try them yourself first, then compare.

📓 View Solutions ▶ Open Solutions in Colab ⬇ View / Download on GitHub
6

Quiz: Test Yourself

Eight quick questions to lock in the four levels. Answer them, hit Check Answers, and keep refining until you score 100%. Your progress is saved, so you can hop back to the chapter and return anytime.