RE: sample size & effect size

["Henkle, Gretchen" <HENKLGA@xxxxxxxxxxxxxxx>]

Hi Bill,
Duncan Lindsay at UCLA Social Welfare (www.sppsr.ucla.edu - faculty) may
answer this.

-----Original Message-----
From: bill higgins [mailto:bill_higginsus@xxxxxxxxx]
Sent: Wednesday, May 14, 2003 12:15 PM
To: Child Maltreatment Researchers
Subject: sample size & effect size


I have seen people calculate effect sizes when
comparing means or in regressions, but can these be
calculated in chi-square analyses?  If so, how?  Is
there a web site with this formula?

Bill




--- GTooman@xxxxxxxxxxxxxxxxxx wrote:
> It's good to keep in mind what significance levels
> are for: to determine
> whether the same results could be arrived at via
> chance.  The likelihood
> that a result could be arrived at by chance 5% of
> the time is considered a
> "gold standard" of social science research, but may
> not always be the most
> important statistic to look at.  Effect size with
> large cohorts is, I think,
> more important, given that many frequently used
> statistics are sensitive to
> sample size.  The converse is also true: a
> pronounced effect, significant at
> only, say, p = .10 may be considered important if
> the effect size (variance
> accounted for by the effect) is high -- sometimes,
> this is referred to as
> the "medical effect size," because often in medical
> studies with small
> samples the effect is so pronounced that trials are
> suspended so that the
> control group can receive the treatment immediately
> -- such was the case
> with the initial aspirin trials for prevention of
> subsequent heart attack,
> if I remember right.
> Statistical significance is a good tool in the
> toolbox.  But only one, and
> has limited value.
> 
> Gregory Tooman
> Research Associate, Children's Services
> American Humane
> 63 Inverness Drive East
> Englewood, CO 80112
> p: 303-925-9471; fax: 303-792-5333
> http://www.americanhumane.org
> 
> 
> > -----Original Message-----
> > From: John M Price PhD [mailto:jmprice@xxxxxxxxxx]
> > Sent: Friday, May 09, 2003 8:52 AM
> > To: Child Maltreatment Researchers
> > Subject: Re: sample size and sig levels
> > 
> > 
> > 
> > One should look also to the effect size to see if
> it is 
> > meaningful in any
> > way.  When you have a large sample, you will find
> the small 
> > parts to any
> > relationship, but one has to evaluate that in
> light of the whole
> > issue/system under study.  An r of .05 can be
> significant, 
> > indicating the
> > covariance is reliable, but it only accounts for
> one quarter of one
> > percent (.0025, .25%) of the variance in the two
> variables.
> > 
> > In preventing sudden failure of a jet engine on a
> commercial 
> > airline, that
> > might be important.  In preventing the same failue
> in an auto 
> > engine where
> > gravity isn't such a killer, well, maybe not.
> > 
> > On Thu, 8 May 2003, bill higgins wrote:
> > 
> > > I recently read an article that had an extremely
> large
> > > sample (population study of health care).  The
> > > differences looked small but were statistically
> > > significant.  should there be adjustments made
> for
> > > extremely large samples?
> > > 
> > > bill
> > > 
> > > 
> > > __________________________________
> > > Do you Yahoo!?
> > > The New Yahoo! Search - Faster. Easier. Bingo.
> > > http://search.yahoo.com
> > > 
> > 
> > -- 
> > John M. Price, PhD                                
>     
> > jmprice@xxxxxxxxxx
> > Life: Chemistry, but with feeling!      |      PGP
> Key on 
> > request or FTP!
> > Comoderator:
> sci.psychology.psychotherapy.moderated          
> > Atheist# 683
> > 
> 


__________________________________
Do you Yahoo!?
The New Yahoo! Search - Faster. Easier. Bingo.
http://search.yahoo.com