But, in the past, when streaming wasn't a big deal, it would have been a much more accurate sample from which to extrapolate the total audience (or come close to that).
Like a poll, essentially. Instead of asking everyone the question, you ask a sample number of people and extrapolate based on that.
It still may be a fairly accurate sample for the viewership on regular tv (important for companies advertising in that format), but with the shift to online streaming, a sample like that would no longer have much value for measuring the overall audience.
In essence, using the Nielsen ratings now for total viewership would be like using an incredibly inaccurate poll that failed to take into account a major factor skewing the sample.
That can happen with regular polls also (hence why they can turn out to be very inaccurate).