Towards a new theory of statistics — full data vs summary
Statistics started as the art of summarizing large amounts of data, with an additional goal of allowing comparisons between datasets.
Fischer– the ‘sufficient statistics’, often two numbers which, combined with an assumption, are sufficient to fully describe a dataset (i.e. assume normality and measure mean and variance).
Statistics as data compression.
The need to summarize is a concession to the structure of our minds. A computer, with a different structure, can easily hold millions of points in its working memory. ( like B.B: Mesquita’s use of game theory to predict negotiation outcomes).
How, then, should we do stats? In the end we DO need a summary, a conclusion.