But that's the thing. You normalize the rate, in this case using per thousands, to account for differences in populations of cars, and then you can directly compare the rates of how many accidents a particular brand has. Demographics matter when you want to understand who the drivers are.
Scenario: there are 20,000 Pontiac cars. The total number of accidents involving Pontiac drivers who were at fault (they caused the accident) was 45,000. 45,000/20,000=2.5 accidents per 1,000 for Pontiacs. Let's compare to 2,500 Teslas involving 7,500 accidents where the Telsa's drivers were at fault. You have 7,500/2,500 = 3 accidents per 1,000 for Teslas.
Even though Tesla's population was only 2,500 vs 20,000 Pontiacs you can normal the rate per 1,000. In this example, Tesla drivers had more accidents per population (in 1,000s). Now if you want to investigate WHO the drivers of those brands are you look at age, income, sex, etc.