Are average online ratings of products (say out of 5) assumed to be on a linear scale as understood by the user? Suppose 2 washing machines A & B are rated as 3.5 and 4.0 (out of 5) respectively, vs another 2, C & D, that are rated as 4.2 and 4.7 respectively. Can I make some useful inference based on the differences among these average ratings on some portal like Amazon? In other words, can I say that B is as much "better" than A, as is D "better" that C?
Or is there some psychological/statistical/marketing theory about the difference in product ratings becoming a finer indicator as the average rating comes closer to the upper limit?
Assume that the number of ratings to each product is approximately equal and statistically significant (e.g. about 200 ratings for each product).
 A: This issue is commonly discussed using Stevens's idea of scales of measurement. The question becomes: a human-completed rating item is surely ordinally scaled, but is it interval-scaled? The consensus seems to be no, but there is considerable disagreement as to whether it's "close enough" to an interval scale to treat it as such. And not all rating scales are created equal, so some rating items might be close enough and some might not.
When you don't want to treat a rating item as interval-scaled, and you want to only interpret it ordinally, there are a number of approaches. Two popular ones are ordinal regression, in which you try to estimate the true distance between each pair of rating choices, and treating the item as nominally scaled, which means throwing out the order information but can make the analysis simpler and more flexible.
A short answer to your specific question is that no, it isn't realistic to presume that the difference in quality between A and B is equal to that between C and D, at least without data on how people rate washing machines on Amazon.
