Re: Statisically Insignificant


Lou Pecora wrote:
In article <1143742050.924655.162320@xxxxxxxxxxxxxxxxxxxxxxxxxxxx>,
robert.dodier@xxxxxxxxx wrote:

z wrote:

I suppose you could ask, given X lottery tickets total, how many would
you have to buy to have your probability of winning be statistically
signficant, i.e. a less than alpha chance, where alpha = .05 or
whatever. Large number.

You could ask that, but it would be entirely beside the point.
The point being, how many tickets must you buy in order to expect
to do better than breaking even. If you work out the details, you'll
find that statistical signficance doesn't enter the solution anywhere.

For what it's worth,
Robert Dodier


I see what you are getting at, Robert (Hi,it's me :-)). The issue of
stastical significance is brought up in a lot of data analysis
(including my own) and lately I've been wondering how to replace it with
the more sensible risk or "payoff" analysis you are promoting. However,
I haven't a clue how to do it abstractly. Eg. You are testing for
correlations between time series and you use the null hypothesis that
they are completely independent (is that the Bartlett choice? I
forget.). So then you reject the null at the 95% level (I think that's
a one-sided test -- although take what I say with a large grain since I
am not a statistician). Anyway, why 95%? Why not 99%? Thus, my
realization that a "risk analysis" is really needed to determine the
alpha level. But then what is the risk or payoff for time series
correlations? If it's a medical decision, you might want 99.9999%
assuredness. If it's just to show there's some common signal 70% might
do.

Bottom line: Do you know of any way to assess risk or payoff
abstractly. I realize this might be almost a ludricrous question, but
maybe there is some aspect of risk analysis that can be extended in some
meaningful way to such abstract, data-analysis situations.

-- Lou Pecora (my views are my own) REMOVE THIS to email me.

Seems to me, the big problem with statistical significance as the final
determinant is its dependence on sample size. All these huge studies
that find some ridiculous "statistically significant" correlation, vs.
all these small studies that find something more logically compelling,
but it's got a p-value of .06, so the headlines all say "new study
finds having more blubber than average walrus does not affect health".

.

Relevant Pages

  • Re: Statisically Insignificant
    ... the more sensible risk or "payoff" analysis you are promoting. ... am not a statistician). ... maybe there is some aspect of risk analysis that can be extended in some ...
    (sci.stat.math)
  • Re: Test of two Independent Proportions: How to Do It RIGHT.
    ... requires a method to assess the Probability of Type I Error or alpha. ... using an un-prespecified level of alpha but through the use of p-value, ... -- Reef Fish Bob. ... As the comment in The American Statistician points out, ...
    (sci.stat.math)
  • Re: alpha stable distribution
    ... >this is my first post at this group and I am not statistician. ... >groud radar as alpha stable distribution. ... I suggest you find out what a stable distribution is. ...
    (sci.stat.edu)
  • Re: PERT analysis
    ... Can it be used in place of @Risk et al? ... I'm not a statistician, but it general terms, ... it accounts only for the probability of each individual task's uncertainty ... Find a good risk analysis tool and forget PERT. ...
    (microsoft.public.project)
  • Re: PERT analysis
    ... I'm not a statistician, but it general terms, ... it accounts only for the probability of each individual task's uncertainty ... what the more sophisticated risk tools do. ... Find a good risk analysis tool and forget PERT. ...
    (microsoft.public.project)