Nearly every statistical fact anyone ever cites does exactly the same thing. When a fact says something like "13% of people are poor," they don't investigate every single person and determine whether they are poor. They create what is intended to be a representative sample, then they create a model that extrapolates the sample set to the entire population. They predict that their sample is representative and that a measure of the whole population would match the sample.

The only difference here is that there's no "poverty election day" where the entire population goes to register their poverty level. So we never actually get to see if the prediction was correct!