All we know is that from a sample of 20k scores those with Yahoo email addresses (which could well be 10k or 2k, we don't know) had a lower average than the rest. It says nothing about the whole population of Yahoo/Hotmail/etc. because we don't know how the email addresses were obtained.
It would be very easy to bias this sample by showing ads about finding your credit score to different demographics within each mail provider.
In addition, they don't show variability within the groups. The differences they find among the averages may not mean much in the context of the variations about the averages.
Furthermore, any statistically significant differences may be due to correlation with other factors. Perhaps domain name varies weakly with geographic location-- which the same source claims also predicts average credit scores.
Finally, their chart accentuates the differences by dropping the zero.
It would be very easy to bias this sample by showing ads about finding your credit score to different demographics within each mail provider.