Re: Fallacy of Hypothesis Testing

From: passerby (passerby_at_yahoo.com)
Date: 02/26/05 

Date: Sat, 26 Feb 2005 14:05:22 GMT

As above, given H is the null hypothesis and E is the "rejection region".

Given we have E, we want to know the following probability
P(H | E) = P(E | H) P(H) / [P(E | H)P(H) + P(E | ~H)P(~H)]

But in reality, we don't have any info on P(H) and P(~H)... and
basically we want to show whether we believe P(H)=1...

So, what can we do?

illywhacker wrote:
> The vague notion of '(largely) contradicts' is taken into account
> precisely if one calculates the probability of the hypothesis H given E
> rather than E given H:
>
> P(US | NY) = P(NY | US) P(US) / [P(NY | US)P(US) + P(NY | ~US)P(~US)] =
> 1
>
> as it should. This does not change the fact that hypothesis testing
> frequently does not do this, and hence is flawed in exactly the manner
> suggested.
>
> illywhacker;
>

Relevant Pages

  • Re: A problem for complete beginners
    ... 'with replacement' or <´without replacment'. ... go back to the <fundamental concept should be the ... Probability is NOT fundamental: it comes from the fact that ...
    (sci.stat.math)
  • Re: Fallacy of Hypothesis Testing
    ... to consider the priors. ... > illywhacker wrote: ... >> precisely if one calculates the probability of the hypothesis H given E ... This does not change the fact that hypothesis testing ...
    (sci.stat.math)
  • Re: What is the probability of this?
    ... Hey there illywhacker, ... Maybe one should compute the probability that both ... or more of the OP's friends have telephone ...
    (sci.stat.math)