The problem is that the sample TouringPlans has isn't representative. You need to randomize the sample from the total pool. TP, OTOH, has a self-selecting pool of their users.
E.g.,
@Disneyhead'71 recently revealed the Uni's Fallon ride has a very high Guest Satisfaction, but, you wouldn't know that from online commentary. This shows that the general public and frequenters of Disney fan sites can have different opinion.
Now, I wouldn't think TP's data is far off, but, it certainly has the ability to be far off in a few cases.