Why does the ratio of two Celsius temperatures carry no meaning?
answer
- where the zero point came from
- convention versus absence of the quantity
- re-express the same reading in another unit
- affine shift versus pure rescaling
basics
~20 sCelsius is an interval scale whose zero is an arbitrary reference point, not the absence of heat, so only differences are meaningful and 20 divided by 10 says nothing. Kelvin has an absolute zero, so its ratios do hold.
solid answer
~50 sCelsius is an interval scale: equal gaps are meaningful, but zero was fixed by convention at the freezing point of water rather than at the absence of thermal energy. Ratios need a true zero as their anchor, and Celsius has none, so "20 degrees is twice as warm as 10" is not a statement about temperature. The cleanest proof is a unit change: 10 and 20 degrees Celsius are 50 and 68 degrees Fahrenheit, so the ratio moves from `0.5` to about `0.74` while the physical situation is unchanged. Kelvin is a ratio scale anchored at absolute zero, and there 10 and 20 degrees Celsius are `283.15` and `293.15` kelvin — a ratio of roughly `1.04`, which is the honest answer. Differences survive on an interval scale; ratios, percent change and geometric means do not.
go deeper
Be ready to say that Celsius zero is a chosen reference point, not an absence of heat, and that this is why differences are fine but ratios are not. One clean sentence is enough at this level.
Demonstrate it rather than assert it: convert the two readings to another unit and show the ratio changing while the physical situation does not. Then list which statistics the interval scale forbids.
Show you catch this in real reporting, where the trap is percent-change metrics computed on scores or indices whose zero is arbitrary. Explain how you would rebase such a metric onto a quantity with a true zero.
Own the metric definitions. Decide which organisational KPIs are allowed to be expressed as percentages or ratios, and require that anything reported as a ratio rests on a quantity with a genuine zero, so cross-team comparisons stay valid.
## The distinction in one line An interval scale has equal, meaningful gaps but a conventional zero. A ratio scale has equal, meaningful gaps *and* a zero that means none of the quantity. Everything else about interval versus ratio follows from where the zero sits. ## Why Celsius is interval Zero degrees Celsius is the freezing point of water at standard pressure — a useful landmark chosen by people, not a point where temperature stops existing. Negative values are perfectly ordinary, which is itself a signal: a scale that goes below zero cannot have a zero meaning absence. What *is* meaningful on Celsius is the gap. A rise from 10 to 20 degrees is the same amount of warming as a rise from 30 to 40 degrees, because the degree is a constant unit. So sums of differences, the arithmetic mean of a set of readings, and the standard deviation all behave sensibly. ## Why the ratio fails: run the unit change Ratios must not depend on the unit you happened to record the data in. Test it: - `10 degrees C = 50 degrees F`, `20 degrees C = 68 degrees F` (using `F = 1.8*C + 32`). - In Celsius the ratio of the two readings is `10 / 20 = 0.50`. - In Fahrenheit the same two temperatures give `50 / 68 = 0.735`. Nothing physical changed between those two lines; only the arbitrary origin moved. A quantity that changes when you re-express the same measurement in another valid unit is not a property of the measurement. Now do the same on Kelvin, whose zero is absolute: - `10 degrees C = 283.15 K`, `20 degrees C = 293.15 K`, ratio `= 1.035`. - Rankine, the other absolute temperature scale, is a pure rescaling of Kelvin, so it returns the identical `1.035`. That stability under a change of unit is what "the ratio is meaningful" actually means. ## The transformation view Stated generally, an interval scale is defined only up to a positive affine transformation `y = a*x + b` with `a > 0`. Order is preserved, equal differences are preserved, and ratios of differences are preserved — but the ratio of two raw values is not, because the offset `b` slides underneath it. A ratio scale is defined only up to a positive scaling `y = a*x`. There is no offset to slide, so a ratio of two values is invariant: `(a*x1) / (a*x2) = x1 / x2`. That single line of algebra is the whole reason ratio scales support ratios. ## Which statistics are affected On an interval scale you may still compute: - differences between values, - the arithmetic mean and the standard deviation, - correlations, ranks and percentiles. On an interval scale you may **not** meaningfully compute: - ratios of two values ("twice as warm"), - percent change from one value to another, - the geometric or harmonic mean, both of which are built on multiplication and therefore need the fixed zero, - any statistic that divides a value by another value on the same scale. A practical trap: the same reading, expressed in Celsius and in Fahrenheit, gives two different percent changes for one physical warming event, so any report quoting "temperature rose 15 percent" is quoting an artefact of the unit. ## A subtlety worth knowing Differences on an interval scale *are* themselves on a ratio scale. A temperature change has a genuine zero — no change — so a 10-degree rise really is twice a 5-degree rise, and that statement survives the conversion to Fahrenheit (18 degrees versus 9 degrees, still exactly twice). This is why change scores and deltas can be analysed with the full arithmetic toolkit even when the underlying readings cannot. ## Other interval-scale variables you will meet - Calendar year and clock time: the year 2000 is not twice the year 1000, and 20:00 is not twice 10:00, because the origin of both counts was chosen by convention. - Longitude: degrees east of a meridian that a committee picked. - Many standardised scores that are deliberately centred at an arbitrary value. In each case, differences and means are fine and ratios are nonsense. Contrast these with age, duration, mass, distance, count and Kelvin, where zero really does mean none of it, and every arithmetic statement is available. ## How to answer this in an interview Name the scale, name the reason (arbitrary versus true zero), then produce one concrete demonstration — the Celsius-to-Fahrenheit ratio flip is the fastest, because it converts an abstract rule into a two-line calculation that anyone can check.
- Which summary statistics stop being meaningful on an interval scale?Anything built on dividing one value by another: ratios, percent change, and the geometric or harmonic mean. Differences, the arithmetic mean, the standard deviation, ranks and correlations all remain fine, because they depend on gaps rather than on distance from a zero point.
- Is a difference between two Celsius readings itself on a ratio scale?Yes. A change in temperature has a genuine zero — no change at all — so a 10-degree rise really is twice a 5-degree rise, and that survives conversion: the same rises are 18 and 9 degrees Fahrenheit. This is why deltas and change scores support full arithmetic when the raw readings do not.
- Name an interval-scale variable outside temperature and say what it forbids.Calendar year. Gaps are meaningful — 1990 to 2000 is the same decade-long span as 2010 to 2020 — but the origin is a convention, so the year 2000 is not twice the year 1000 and percent change between years is meaningless. Clock time and longitude behave the same way.
saying these in an interview costs you the question
- Says zero Celsius means no heat is present
- Claims Fahrenheit is a ratio scale
- Computes percent change between two Celsius readings
- Thinks negative values are compatible with a true zero
- Cannot name a statistic that interval scales forbid