Most definitely, yes!
Best wishes,
Simon
----- Original Message -----
From: Dyn Parry BSc(Hons) SRCh MChS <[log in to unmask]>
To: <[log in to unmask]>
Sent: Friday, July 28, 2000 5:34 PM
Subject: Re: Interpreting an ICC (type 2,1) from SPSS
> Dear Tony,
> Why are the CIs so rarely reported? and should we be reporting confidence
> intervals in our research, to compliment p-values.
>
> Cheers
> Dyn
>
> ----- Original Message -----
> From: Anthony Redmond <[log in to unmask]>
> To: <[log in to unmask]>
> Sent: Friday, July 28, 2000 1:26 PM
> Subject: Re: Interpreting an ICC (type 2,1) from SPSS
>
>
> > Hi Craig
> >
> > -----Original Message-----
> > From: Craig Payne <[log in to unmask]>
> >
> >
> > >Anyone with some knowledge of ICC's??????. I got the following from
> SPSS -
> > >I have rechecked it several times and can't work out how to interpret
the
> > >ICC with a negative value (-1.5523) and greater than 1
> > (95%CI: -8.5186-0.7043)
> >
> >
> > The ICC is a measure of the agreement between observations **in excess
of
> > that which would be expected by chance**. I am sure you know this.
> > In this context in ICC of 0.0 would mean 'no better than chance' and
the
> > values increase from here.
> > Theoretically ICCs would therefore only range from 0 -1 but in practice
> the
> > ranges are much greater than this. Lahey et al report that for 2,1 the
> > theoretical range is actually +1 to -(minus) infinity. For the ICC 2,k -
> > which I think is the 'average' measure correlation you report, the
> possible
> > range is +infinity to -infinity ! (Didn't want to hear that eh?) it is
not
> > merely an SPSS specific artifact.
> >
> > See
> > Lahey MA. et al . Intracalss correlations:: there's more than meets the
> eye.
> > Psychol Bull 1983. 93:3; 586-595
> > which deals with this in detail.
> >
> > A couple of key points re your particular results
> > If you look at the 95% CIs for most ICCs, they are depressingly wide -
> hence
> > they are rarely reported in the literature. The CIs are so rarely
reported
> > I'll bet you cannot find more than a couple of cases where they have
> > actually been reported in a paper looking at the reliability of a
measure.
> > If you look at your own raw data though - you might find a few cases
where
> > the 95% CI includes -1 or other sub zero values.
> > In theory though the ICC cannot be less than **unless the ANOVA from
> which
> > it is derived is also insignificant**.
> >
> > In the Lahey paper they suggest that "a negative ICC can only be
obtained
> if
> > the BMS is less than the EMS or WMS." For the two-way ANOVA this would
> mean
> > that the overal sig of the (F score component of the ) ANOVA table would
> be
> > ***insignificant***. ie if the ANOVA is not significant then the ICCs
are
> > not significant either. Again, very few papers report the F score or the
> > overall sig of the ANOVA so this does not spring to mind unless you are
> > looking for it.
> >
> > I don't claim to know this in depth, your best bet is to look up the
Lahey
> > paper. A few others mention it in passing too. Bland and Altman, Portney
> and
> > Watkins and many other critics of the ICC have levelled similar
crtiticism
> > at this particular measure. Such is the price you pay for having such a
> > convenient 'single' measure of agreement.
> > In a nutshell it means that in your specific case the agreement is no
> better
> > than that which would be expected through chance alone -indeed it is
> > possibly worse - but you cannot be sure becaue the ANOVA in
insignificant.
> > (also check the 2,1 v the 2,k aspect)
> >
> > BTW I hope these are not FPI figures we are talking about !!!!! - you
know
> > what I mean.
> >
> > ...and yes, I had to look this up.....
> >
> >
> > All the best
> > Tony
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
>
>
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|