Re: statistical test help for beginner please

On 26 Jul 2006 07:42:23 -0700, ViPF.Tool@xxxxxxxxx wrote:

Hi,

i would appreciate it if someone with knowledge of statistics could
read over my problem and maybe can suggest an appropriate test.

i am running experiments that often result in a contingency table like
this:

V=v V!=v
--------------------------------
U=u | 2 | 17
U!=u | 60 | 80900
--------------------------------

with a null hypothesis that (u,v) are independent;

usually i would apply the Chi-Square test but due to the small values i
end up with expected values below 1 which screws up the whole Q
calculation. Further research suggested that Fisher's exact test should
be the method of choice for observed values below 10, but calculating
the factorial of 80900 breaks my machine.

is there any other statistical test that is appropriate?

The nature of the problem with small expectations, less than
5 or less than 1, is that the distribution is not so "exactly"
equivalent to a chisquared. I haven't seen this written up, but
I imagine you might get 6.0 when it ought to be 3.0, say.

The Yates-corrected chisquared test is ordinarily a good
approximation to the FET. Again, I don't know when it might
break down. However, I get a X^2= 151.8, which seems
pretty unquestionably extreme.

Someone else mention OR -- I expect that the Odds ratio
test is going to have troubles with accuracy for a cell of 2.

The practical implications of seeing cell 1 equal to 2 instead
of zero may be rather limited....


--
Rich Ulrich, wpilib@xxxxxxxx
http://www.pitt.edu/~wpilib/index.html
.

Relevant Pages

  • Re: Is this guy familiar with Sean Pitman?
    ... mathematical calculation shows that some routine process is impossible, ... The odds are obviously 100% because common descent is supported by ... to support your religious position. ... lie with statistics, attempting to create controversy where none ...
    (talk.origins)
  • Re: The Philosophy of War
    ... If you cant do a simple rote calculation how can you be expected to ... I can see how you might conjecture that I used the red herring, ... With the proper understanding of statistics, ... You don't understand statistics and probabilities. ...
    (rec.martial-arts)
  • Re: Is this guy familiar with Sean Pitman?
    ... mathematical calculation shows that some routine process is impossible, ... lie with statistics, attempting to create controversy where none ... the notion of common descent still stands. ... god of the gaps. ...
    (talk.origins)
  • Re: Is this guy familiar with Sean Pitman?
    ... mathematical calculation shows that some routine process is impossible, ... functional system beyond a given level of functional complexity? ... lie with statistics, attempting to create controversy where none ... You forget, John, that common descent isn't a "mechanism". ...
    (talk.origins)
  • Re: Is this guy familiar with Sean Pitman?
    ... mathematical calculation shows that some routine process is impossible, ... The odds are obviously 100% because common descent is supported by ... to support your religious position. ... lie with statistics, attempting to create controversy where none ...
    (talk.origins)