They’re simply extrapolating the percentage from tested samples.
Yes of course. But what is the sample size is my question. Is it large enough for high confidence? Are the samples collected randomly from around the country or just specific hot spots, etc? Lots of ways to manipulate data by selection bias among other biases.