Edit: maybe a better way to phrase this is that, without knowing criteria and scale, there's really no way to know if this is surprising and possibly informative, or utter nonsense and it's hard to find anything conclusive because they're trying to hide their horrifying abuse of statistics. I see the latter happen far more often than the former, which is why significant claims require significant evidence. So far this seems weirdly vague.
---
What I mean is that, depending on the criteria, their average network size could be 30 or 50,000.
A reduction of 17 from 50,000 is only 0.03%. That's overwhelmingly likely to be statistical noise and nothing else. But given enough dimensions to slice the data across, that could be the single strongest correlation, so it's the one they claim. You see that kind of thing all the time in bad data science.
The claim could fall anywhere between "shockingly strong correlation" and "utterly meaningless". So far I haven't been able to find anything that implies one way or the other.
---
What I mean is that, depending on the criteria, their average network size could be 30 or 50,000.
A reduction of 17 from 50,000 is only 0.03%. That's overwhelmingly likely to be statistical noise and nothing else. But given enough dimensions to slice the data across, that could be the single strongest correlation, so it's the one they claim. You see that kind of thing all the time in bad data science.
The claim could fall anywhere between "shockingly strong correlation" and "utterly meaningless". So far I haven't been able to find anything that implies one way or the other.