Statistics › Section A · Sampling and data › Topic A.4
Levels of measurement
Knowing a variable's level of measurement - nominal, ordinal, interval or ratio - tells you which statistics are legal on it. This topic builds the four-level ladder and shows how to use it before touching any data.
Why levels of measurement matter
Before choosing a summary or a calculation, you have to know how your data were measured - because not every statistical operation is legal on every data set. The level of measurement is the classification that answers this, and it runs through four levels from lowest to highest: nominal, ordinal, interval and ratio. Each level permits what the one below permits, plus something new - nominal gives categories, ordinal adds a meaningful order, interval adds meaningful differences between numbers, and ratio adds a true zero that makes ratios meaningful. Classify first, calculate second: most mistakes with data types start by ignoring this ladder.
Nominal scale data
Nominal scale data are categories and labels with no inherent order: blood types, car colours, preferred music genres, yes-or-no answers. The word nominal comes from the Latin for "name", which is exactly what these data are. Any ordering you impose - alphabetical, or the order someone listed them in - is for convenience and carries no meaning: putting reggae before rock tells you nothing about reggae. What you may do with nominal data is count: frequencies, relative frequencies, and comparisons of those counts across groups. What you may not do is arithmetic on the categories themselves - there is no "average blood type".
Ordinal scale data
Ordinal scale data are categories with a meaningful order but no measurable distance between them. Race finishing places are the classic case: first beat second, but by how much? The gap could be a photo finish or a whole lap. Satisfaction ratings (excellent, good, fair, poor), heat levels on a menu (mild, medium, hot) and league rankings work the same way: the order is real, the differences are not numbers. Like nominal data, ordinal data cannot sensibly be averaged - "excellent minus good" has no value, and asking whether excellent is twice as satisfying as good is meaningless.
Interval scale data
Interval scale data are numbers with a definite order and meaningful differences, but no true zero. Temperature in Celsius is the standard example: 40 degrees is as warm above 30 as 30 is above 20, so differences are real arithmetic. But 0 degrees Celsius is an arbitrary convention, not an absence of temperature - minus 10 degrees exists and is colder. Because the zero is arbitrary, ratios are meaningless: a 30-degree day is not "three times as warm" as a 10-degree day. Calendar years work similarly: 2020 minus 2000 is a meaningful 20-year gap, but the year 2000 is not twice the year 1000.
Ratio scale data
Ratio scale data have everything interval data have, plus a true zero that means "none of the quantity". With a true zero, ratios become meaningful. Quiz scores out of 50 have 0 as the floor: a 45 is three times a 15, and that statement makes sense. The same holds for money, weight, length, time and counts: $0 is no money and 0 minutes is no waiting time, so a wait of 8 minutes really is twice a wait of 4 minutes. Test any variable for ratio level by asking whether "twice as much" and "half as much" make literal sense.
Worked example: classifying four variables
Classify each variable. a) A café's sauce labels (mild, medium, hot, inferno): ordered categories, but no measurable gap - ordinal. b) Daily high temperature in °C: differences meaningful, zero arbitrary - interval. c) Grams of sugar in a serving: 0 g means no sugar, so ratios work - ratio. d) Passengers' chosen airlines: names with no order - nominal. Check each answer with the ladder questions, in order: no order at all? Nominal. Order without measurable gaps? Ordinal. Gaps but no true zero? Interval. True zero? Ratio. Step through the four questions and the level falls out.
Rounding answers sensibly
Real data usually arrive rounded, and answers should respect that. The working rule: report the final answer with one decimal place beyond those in the original data, and round only at the end. Avoid rounding intermediate results - if you must, keep at least double the number of decimals the final answer needs, because small errors compound silently through a long calculation. If the data were recorded to one decimal place, report answers to two. In probability work especially, leaving a fraction unreduced - 12 out of 40 rather than 3 out of 10 - is often more informative.
Worked example: which claims make sense?
Judge each claim. a) "My call was on hold for 8 minutes, twice yesterday's 4." Hold time has a true zero, so the ratio works - meaningful. b) "Today's high of 30 °C is three times as warm as the 10 °C day." Celsius has no true zero; the claim confuses a difference with a ratio - meaningless. c) "Inferno sauce is half of medium." Sauce labels are ordinal; halving them is not defined - meaningless. d) "Vanilla got three times as many votes as chocolate." Vote counts are ratio-level numbers - meaningful. Check: whenever a claim says "twice" or "half", ask first whether the variable has a true zero.
Common mistakes to avoid
First, treating ordinal data as numbers: finishing places and star ratings are not quantities, and averaging them invents a precision the scale never had. Second, computing ratios on interval data: 30 degrees is not three times 10 degrees, because zero on that scale is a convention. Third, assuming digits make data high-level: a zip code is nominal even though it is written with digits. Fourth, skipping the classification step and reaching for whatever calculation is at hand - the ladder question takes ten seconds. Fifth, rounding intermediate results: round only the final answer, carried one decimal place beyond the data.
Key takeaways
Every variable sits on one of four levels. Nominal: named categories with no order - count them. Ordinal: order without measurable differences - rank them. Interval: meaningful differences but an arbitrary zero, like Celsius - subtract, but do not form ratios. Ratio: a true zero, as with money, mass, time and counts - everything works, including "twice as much". Each level permits all the operations of the levels below it, so classify before calculating. And round only the final answer, one decimal place beyond the data, never the intermediates.
Flashcards
Flashcard
How a data set is measured, which decides the legal operations
Test your understanding
Question 1 of 5
A café labels its hot sauces mild, medium, hot and inferno. These labels are at which level of measurement?
Question 2 of 5
Daily high temperatures recorded in degrees Celsius are at which level of measurement?
Question 3 of 5
Which claim is meaningful for data at the ratio level but not at the interval level?
Question 4 of 5
Scores on a 50-point quiz, where 0 is a possible score, are at which level of measurement?
Question 5 of 5
Data were recorded to one decimal place. Following the rounding rule, final answers should be carried to:
