,
6 tweets,
2 min read
Read on Twitter
I spend a lot of my day job hours thinking about numerical data and am increasingly fascinated about the inferences we can draw from even single charts. Ex: Based on these ratings, is this a good book? 
We could say that we can't tell because N=17 isn't a large sample, though I'm betting that in Amazon's data how a book looks after 15 responses isn't likely all that different than after 100.
The two most common answers are 5 stars and 1 star. This suggests at the least that if you read the book you won't be indifferent about it. There's a good chance you'll be moved to give the book 5 stars, but there's a non-trivial chance that the book will be truly terrible.
Amazon's average star rating is largely meaningless, but I find the distribution, particularly at the poles meaningful. If there's a novel that I'm intrigued by and I go looking at Amazon and see a 3-star average driven by a mix of 5-star and 1-star reviews, my interest goes up.
Certainly I'm more interested in a book that averages 3 stars through a mix of 5 star and 1 star reviews than a book where the vast majority of readers give the book 3 stars. The former book could be awesome if I share traits with the 5-star raters.
This is where the open-ended qualitative data in customer reviews comes in. I can which group my values better align with. The average by itself doesn't mean much, but there's usually enough data to make an informed individual choice if I dig deeper.
