Town A reports 50 flu cases and Town B reports 200. B sounds worse until you learn that A has 1,000 people and B has 20,000. Divide each count by its population: A has 100050=0.05 cases per person, B has 20000200=0.01. Multiply by 1,000 to make the numbers friendlier: 50 cases per 1,000 people in A, 10 per 1,000 in B. Town A is five times as sick.
The multiplier changes how a rate reads, not which group comes out ahead. Rare events give tiny rates: 3 cases among 100,000 people is 0.00003 per person, which nobody can picture. Multiplying by 100,000 turns it into "3 per 100,000." Whatever multiplier you choose, use the same one for every group you compare.
The base matters as much as the division. To compare car crashes in two cities, dividing by population fits a question about residents' risk, while dividing by miles driven fits a question about how dangerous the roads are. Match the denominator to the people or things actually exposed.
Normalization is different from aggregation, which only adds up a year of crashes into one total; normalizing then divides that total by a base so it can stand next to another city's. It also differs from describing one group's makeup: "20% of City A's residents are children" is a share of one whole, while "3 thefts per 1,000 in A versus 6 in B" puts two groups on one scale.
Think of it like this. Normalizing is the unit price on a supermarket shelf. A 2 kg bag of rice for $6 and a 5 kg bag for $12 can't be judged by the sticker, but per kilogram they cost $3 and $2.40. Dividing by size reveals the real deal, just as dividing by population reveals the real rate.
Core idea. Normalization rescales different quantities to a common reference so you can compare them fairly.