Here is the first problem in all of quantitative research: to do statistics, you have to talk to computers, and computers don't know what you mean by "are household labor tasks between husbands and wives equal?"
Statistical software runs numbers. If your question isn't numbers yet — and no real question starts as numbers — you have to do the translating. And AI won't rescue you from this. Ask a chatbot to "analyze my survey to find out if religion affects how people vote," and a good one hands the decision straight back: should I treat religion as unranked categories and compare voting across groups, or assign each religion a number and compute a correlation? Your income variable came in brackets — low, medium, high — should I treat those as evenly spaced numbers or as ranks with unknown gaps? Those are choices. They change the answer. Levels of measurement is the name for making them correctly.
There are three levels. I'll start with the one you already know.
Age in years. Income in dollars. Height in inches. GPA. Temperature. Two properties make these interval-ratio. First, they have a fixed order — 30 is more than 20. Second, the distance between any two points is equal and meaningful: the gap between 20 and 30 is the same size as the gap between 70 and 80. Because of that, arithmetic means something. "The average age of this class is 21.3" is a sensible sentence. You can ask how much more? and the answer refers to something real. (This is also why "I'm eleventy-six years old" means nothing — a number only works because it sits a known distance from other numbers.)
Education: high school diploma, bachelor's, master's, PhD. Is a PhD more education than a diploma? Obviously. Can you rank them? Obviously. So it looks like interval-ratio — there's an order, there are levels. But: is the jump from high school to bachelor's the same size as the jump from master's to PhD? One is four years of undergrad; the other can be seven years and a dissertation. The ranks are real. The distances between them are unknown. That's ordinal: you know which is higher, never by how much. Likert scales — strongly disagree to strongly agree — are the classic case. Agree is above disagree, sure. But is disagree→neutral the same distance as agree→strongly agree? How would you even know?
Now try this: "Rank the following religions from lowest to highest: Catholic, Protestant, Jewish, Muslim, Buddhist." Lowest to highest on what? Holiness? Popularity? Age? There's no dimension to rank them on — they're not more or less of anything, they're just different things. That's nominal: categories with no inherent order. Republican or Democrat. Catholic or Protestant. The only question a nominal variable can answer is same group or different group? Two rules make nominal categories usable: they must be mutually exclusive (every case fits exactly one) and exhaustive (every case has somewhere to go — which is what "Other" is for).
The diagnostic. Two steps, ten seconds, settles anything.
Step 1: Rearrange the categories. Is anything factually wrong? Male, female → female, male: nothing's wrong. Nominal — stop. Hot, warm, cold → hot, cold, warm: that's wrong, everyone knows hot > warm > cold. There's a built-in order — continue.
Step 2: Can you name the unit between the ranks? Age: the unit is a year, and a year is 365 days, and everyone agrees — the unit exists in the world. Income: a dollar is a dollar. Interval-ratio. Now a Likert scale: what's the distance between "disagree" and "neutral"? One… agreement? What is one agreement — how long does it last, how much does it weigh? You wrote 1, 2, 3, 4, 5 on the page, but there is nothing in the world those numbers are counting. The distance from "very happy" to "not happy" is not "two happinesses." No nameable unit → ordinal. If you can name the unit, it's interval-ratio. If you can't, it's ordinal.
Why this is worth ten seconds. Research question: does temperature affect ice cream sales? Measure temperature ordinally — cold, warm, hot days — and your best possible answer is "hot days sell more." Direction only. Measure it in degrees and you can say "every 1-degree increase adds $47 in sales." Direction and magnitude. Same concept, same question, same variables — the measurement choice determined what kind of answer was even available. The level of measurement also decides which statistics are legal: the mean of religion is nonsense, the mean of age is fine, and the entire logic of choosing the right statistical test starts exactly here.
One application before you go: nearly every human rating in an AI evaluation — helpfulness 1–5, preference scales, quality scores — is a Likert. Treating those as interval numbers is assuming "two happinesses" is a quantity. Sometimes it's a tolerable assumption. It is never a free one.