« A Cricketer Leaves The Crease | Main | Gordon Brown: The Personality Thing (Again) »

March 19, 2007



Well, as someone with a degree in statistics, I have to express my profound disagreement with the statement that "Statistics are just numbers. They are completely value-free". No statistic, absolutely and categorically, is value-free. Let me say that again, in case some one wasn't listening: No statistic is value-free.

Values are embedded in every stage of the generation of every statistic:

(a) In the definitions of underlying concepts -- such as what constitutes a "cot-death";

(b) In the means of collection -- such as administrative processes which may involve varying degrees of reliability and self-interest -- see just about any police statistic;

(c) In the aggregation and calculation processes -- we will almost certainly get different results if we draw the regional boundaries differently, as any political gerrymanderer can tell you;

(d) In the publications policy -- ask the UK Government why they want to retain the right to preview official statistics before publication;

(e) In the interpretation -- why do we use 5% or 1% significance levels for statistical hypothesis testing? These particular threshold levels encode particular values regarding the desirability or otherwise of accepting or rejecting new ideas.

There is no such thing as a value-free statistic.

The comments to this entry are closed.