Nintendo games have different decay curves, pricing and upside potential, among many other things. They are, to a large extent, consistent outliers. Outliers, if not accounted for in a data set, throw off whatever findings may be had and increase ranges of error. It's not my opinion, it's just how the math works.
Exactly. The metacritic score is a certainly imperfect quantitative surrogate for the actual unquantifiable factors that help a game sell. Like Andy says, the game doesn't sell because of the MC, it sells because of the reasons the MC is what it is.
In genres that have GOOD correlations between MC and sales you really only get to an r-squared of like .35, meaning that one can explain 35% of the variability in sales between like titles by difference in MC. Unfortunately, that leaves 65% that must be explained by differences in marketing, distribution, etc etc.
But really, best case, using all available quantifiable attributes, I've yet to see a genre where we can explain more than half of the variability of sales between titles.
So what does that mean? Huge ranges of error in forecasting based on unquantifiable factors.
It's why multivariate regression modeling breaks down in games and why if you get within +/- 10% on any forecast you threw a bullseye and more likely got lucky with the forecast than actually knew what was going to happen.
Anyways, long story short, pinpointing forecasts on the sales on a group of games on a consistent basis is extremely difficult.