Re: Larry Athy and His Logical Analyses.

"Larry Athy" <[EMAIL PROTECTED]> wrote in message
news:[EMAIL PROTECTED]
> On Tue, Dec 2, 2003, 11:44am (CST+6), JMB posted here again and
> continued to demonstrate that he does not understand that the
> statistical method I used in my paper can be used with different types
> of problems.  He simply insists on disagreeing with things about which
> he understands nothing.
> ===================
> I had written: ++ The "r" value gives us some idea as to how well my two
> groups are related to one another. The "p" value takes "N" into account
> and thus gives a more meaningful value of relative relationship. In the
> even that they are not related, "r"= 0 and "p" =1. A "p" > .05 has no
> meaning in this type analysis. ++
> ---------- 
> He, now: ~~ Something Larry has made up. Williams doesn't give this
> method. Nobody does. In the Williams book, the method given is to assume
> a null hypothesis of r = 0. Then you work out what the chances are of
> getting the r value that you actually got if the real r should be 0.
> Getting a p < .01 means that there should be less that one in 100 chance
> of getting your r value if the true r was 0. It does not mean that there
> is over 99% correlation between the sets. Williams never claims that the
> p value means what Larry claims it means, and I can't even see anywhere
> that Williams says something that could be honestly confused with such a
> meaning. ~~
> ----------
> +++ Note that he says "Getting a p < .01 means that there should be less
> than one in 100 chance of getting your r value if the true r was 0".
> That means the same thing as "Getting a p < .01 means that there should
> be more than 99 in 100 chances of not getting your r value if the true r
> was 0".

Okay, Larry is doing okay up to here.

>  Thus the true r is not 0 with a 99% probability.

Still doing okay.

>  Thus they
> would almost certainly be closely related.

This is where Larry goes off track.  Just because r is not equal to 0 does
not mean that the sets are closely related.  The true r could be .5, giving
a determination of 25%, which is a poor correlation.

> - Note that in my paper the HP-1/Bi1 relationship has an r =.659, t =
> 3.155, and p = 01.  Thus they are almost certainly closely related by
> what JMB is quoting. +++

No, what that is saying is that you can be 99% certain that r is not 0.  It
does not mean you can be 99% certain of a correlation.  It does not even
mean that you can be 99% certain the the r=.659 is correct, although that
would be assumed.  An r=.659 would give a determination of 43%, which is a
poor correlation.

>         Regards,  Larry Athy, P.E.
>
>
>
>

-- 
John Byrne
www.iol.ie/~archaeology
To email me use the feedback form on the website.
The address attached to this post is just a spam trap.