Here is a query from someone I met in a hallway a couple weeks ago.
I'm not sure I even understand the question. If anyone out there
recognizes anything familiar in this scenario, you might respond to
JMC -- not to me or the list.
----- Forwarded message from Jennifer Mary Collins -----
The problem is that the data I use, which is hurricane numbers, deals
with small numbers. Hence they are poisson distributed, when one
considers the other criteria to be poisson.
If I have pressure, 5 strongest pressure years from a 26 year
data set, and 5 weakest pressure years and calulate the average number
of hurricanes in each group.
I have group a, with an average of say 4
group b 0.4
The stdev on each is 1.23 an 0.55 repectively.
I calculated the ratio of a/b to be 10:1
I am trying to work out the significance of this ratio. I did this in
three ways,
1/ see if the Stdev on each number overlap to the other average
or even the other stdev on the average, and reject it if it does. In
this case I can accept it as the stdev's do not overlap to one standard
deviation.
or
2/ I worked out a propogation of errors on the ratio.
where Ratio= Ratio+-error
error= (stdev of hurricanes (groupa)/root(n))squared/(average(group
a)squared) + (stdev of
hurricanes(groupb)/root(n))squared/(average(group b)squared)
so is the method I did above ok? but then I was going to ask, what level
of error is significantly ok, and what is unacceptable. I got this
equation from a book by Bevington, but I think this may be for Gaussian
distributions.
3/ Another equation I have found is in a book by Rothman and Greenland,
'Modern Epidemiology'.
sqrt(1/Nprh + 1/Nprl)
where N is total number
where prh prl are for High pressure years and low pressure years
respectively,
I get
sqrt(1/20 + 1/2)=sqrt(0.55)=0.74
Then you test if this is significantly different from unity.
I am aware that this also requires numbers to be greater than 10.
So my problem is that I want to look at ratios from two numbers and see
if they are significant, but I need to account for small numbers, i.e.
not Gaussian distributions. Are you able to suggest an appropriate
technique?
I look forward to hearing from you and hope that you do not mind me
asking you this question.
Many Thanks
Jennifer
---------------------------------------------------------------
Jennifer Collins Mullard Space Science Laboratory (MSSL)
[EMAIL PROTECTED] University College London
Tel +44 1483 204151 Holmbury St. Mary
Fax +44 1483 278312 Dorking, Surrey, RH5 6NT, UK
---------------------------------------------------------------
----- End of forwarded message from Jennifer Mary Collins -----
--
_
| | Robert W. Hayden
| | Department of Mathematics
/ | Plymouth State College MSC#29
| | Plymouth, New Hampshire 03264 USA
| * | Rural Route 1, Box 10
/ | Ashland, NH 03217-9702
| ) (603) 968-9914 (home)
L_____/ [EMAIL PROTECTED]
fax (603) 535-2943 (work)
