Top critical review
on 11 June 2017
:) Numbers do not lie. Numbers do not tell the truth either.
Its people who read into numbers to lie and this is what these authors have done - knowingly or unknowingly.
They need to get their math straight first. In logic suppose we have P => Q , given Q we cannot say anything about P.
So when we see numbers and infer greatness, greatness => numbers and not vice versa, so seeing numbers there is really no absolute inference possible about the greatness. But then probabilistic inference is possible. For probabilistic inference a prior is needed and then one can arrive at a comparison by making use of the prior.
The problem with the analysts here is calibrating a function that artificially puts some people in the higher bracket and end up with wrong calculations. if X is a random variable, one can always evaluate a function f(X) that is such that when X is high f(X) is low and viceversa. so if i am a little careful in choosing a function I can make any conclusion by cutting and pasting a few functions f, g, h etc together to formulate my favorite batsmen as greater than the others.
And many naive and mathematically unsophisticated people would fall for it and take it to be some kind of elite analysis , while in actuality its simply ignorance garbed as knowledge.
marred by standard management misnomers like "it is a team sport" [ya it is, but you do not evaluate players based on team success. ] , "ultimately victory matters" [ya it does but do you have a negative impact when one player plays it really well and others spoil the game by failing the match ] ... these are management jargon without meaning ... how about taking a negative impact ? invert all scores, make them -X if score is X and change loss to victory and vice versa and calculate your impact index. you shall get another statistical quantity ... how about summing these to see how players figure out ?
in any case subjectivity comes in through two interesting by routes : one is the impact index function thats calculated , the other is through what statistical attributes contribute and what do not. this is very big issue. in machine learning its called feature selection and many cases the features selected are wrong or their interrelationship is misread and here its glaringly obvious. If you are looking to create a start up here is my advice: try deep learning for representation learning and you shall get a much better approach.
On the whole the logic is weak and meaningless ! I am convinced t hat if you try averages and variance etc you would get better analysis than this.