From: [EMAIL PROTECTED]
> Actually I'm just looking at Ian Taylor's submission to the HAC
> report and there is a huge mistake in it, he says that Killias
> did a Spearman rank order correlation on the data he collected
> that shows a connection between firearm ownership and homicide
> rates, but that's not accurate, Killias did a Pearson correlation.
> I did a Spearman correlation for the Dunblane Public Inquiry and
> it doesn't show a statistically significant positive correlation.
I looked at the set of data produced by the Home Office in some detail.
Steve is quite correct if you perform a Spearman correlation, the result is
in no-way significant.
Depending on which data items you include you can get a positive correlation
using a Pearson correlation coefficient but it is highly unstable. Leave one
or two countries out of the equation and you get wild results from a nearly
certain correlation coefficient of +1.0 to virtually no correlation at 0.0.
A strong correlation would be insensitive to such considerations.
The result presented by Killias and echoed by the Home Office is entirely
spurious. It is a well known problem with small data sets that you cannot
get a meaningful correlation. Given any two completelty random data sets you
can get a positive correlation with small data sets. Prof. Horrocks
illustrated the sheer stupidity of this by correlating homicide and car
ownership - you will actually get a stronger correlation.
I can back this up by supplying the original data to whomever is interested.
At the end of the day correlation by itself does not imply causation which is
the single biggest flaw of Killias' argument. Marriage statistics used to
show a strong correlation with springtime, nothing to do with a young man's
fancy but with a full years married man's tax allowance!
Cybershooters website: http://www.cybershooters.org
List admin: [EMAIL PROTECTED]
____________________________________________________________
T O P I C A -- Learn More. Surf Less.
Newsletters, Tips and Discussions on Topics You Choose.
http://www.topica.com/partner/tag01
