All data is susceptible to manipulation, one of the issues in academia is the prevalence of positive results (i.e. suggesting both bias by editors and gaming editors by researchers), "data mining" is a serious issue in the social sciences (and why Big Data is overblown, correlation is not causation).
However, more data is better than less, including anecdotal data as long as it's valid data.
Measurement error is always a problem, especially if you don't know the potential bias or inaccuracy.
Corsi has value, xGF% has value, relative and adjusted measures have value.
They're not the be all and end all, but used together can provide a picture of how a player is doing.
More detail can be found with entry and exit stats, etc.
But there are left out variables, MacDonald has given up fewer goals on ice than would be predicted by his xGA, one season could be a fluke, an extended period is more likely a trend due to an unobserved variable.
Of course, right now MacDonald has horrible stats because, well, he's been horrible, a strong correlation there.
I think it's because his legs aren't in game shape and he's a step behind, but at 32 he could be approaching the career cliff.
The next few weeks will tell the tale, if he doesn't pick up his game, he could be gone by Thanksgiving.