Yes, but … with all due respect – that wasn’t the question he should really be addressing.
Robert Newcombe.
From: Martin Bland [mailto:[log in to unmask]]
Sent: 25 July 2018 15:28
To: Robert Newcombe <[log in to unmask]>
Cc: [log in to unmask]
Subject: Re: correlation coefficient
That is fair enough, Robert, but the original question was about a lack of statistical significance.
Martin
On 25 July 2018 at 12:40, Robert Newcombe <[log in to unmask]<mailto:[log in to unmask]>> wrote:
In answer to Martin's comment - it is true that in terms of achieving p<0.05, a non-parametric correlation measure is useless here. Nevertheless, my stance is that quantities such as correlation coefficients, odds ratios, etc. etc., should be regarded primarily as measures of effect size - which are sample-size-free and can be interpreted as point estimates - rather than as a route to obtaining a p-value.
-----Original Message-----
From: A UK-based worldwide e-mail broadcast system mailing list [mailto:[log in to unmask]<mailto:[log in to unmask]>] On Behalf Of Steve Nyangoma
Sent: 25 July 2018 11:50
To: [log in to unmask]<mailto:[log in to unmask]>
Subject: Re: correlation coefficient
Could resampling methods be used to generate more data to confirm evidence?
On Wed, Jul 25, 2018 at 10:08 AM Martin Bland < [log in to unmask]<mailto:[log in to unmask]>> wrote:
> Non-parametric is useless here, as with fewer than 6 observations all
> possible values of either Spearman's or Kendall's have P>0.05.
>
> This is a case where the question of whether there is good evidence of
> a relationship between two variables with only 4 observations always
> has the answer "no, you would need more data."
>
> Martin
>
> On 25 July 2018 at 06:24, Juanita Hatcher <[log in to unmask]<mailto:[log in to unmask]>>
> wrote:
>
> > I agree. Non parametric may be better, but one is still very
> > restricted
> by
> > a very small data set. Look at the data.
> >
> > Juanita Hatcher
> >
> > > On Jul 24, 2018, at 2:33 PM, Robert Newcombe
> > > <[log in to unmask]<mailto:[log in to unmask]>>
> > wrote:
> > >
> > > Exactly! The first step is to plot a scatter diagram. Also, on a
> dataset
> > as small as this, there is no real positive assurance that
> > parametric assumptions hold, and the correlation is particularly
> > sensitive to distributional assumptions. So I suggest a
> > non-parametric correlation rather than a parametric one anyway. With
> > the proviso that this could
> well
> > be exactly 1, without this finding being anything to get excited about.
> > >
> > > Robert Newcombe.
> > >
> > >
> > > -----Original Message-----
> > > From: A UK-based worldwide e-mail broadcast system mailing list
> [mailto:
> > [log in to unmask]<mailto:[log in to unmask]>] On Behalf Of John Bibby
> > > Sent: 24 July 2018 21:28
> > > To: [log in to unmask]<mailto:[log in to unmask]>
> > > Subject: Re: correlation coefficient
> > >
> > > Please look at the data, not the summary statistics. JOHN BIBBY
> > >
> > > On Tue, 24 Jul 2018 at 20:01, Martin Bland <
> > > 000017e8e212eb29-dmarc-
> > [log in to unmask]<mailto:[log in to unmask]>> wrote:
> > >
> > >> You are correct, the value of the correlation coefficient which
> > >> would be significant with 4 observations is 0.95.
> > >>
> > >> Martin
> > >>
> > >> On 24 July 2018 at 17:57, paaveen jeyaganth <[log in to unmask]<mailto:[log in to unmask]>>
> wrote:
> > >>
> > >>> Dear allstat ,
> > >>> i have 4 data point i did a pearson correlation end up with r=
> > >>> 0.8919 p= 0.1081 why is that it's not significant since it's
> > >>> high correlation 0.89 because of sample size??
> > >>>
> > >>> Thanks
> > >>> Paaveen
> > >>>
> > >>> You may leave the list at any time by sending the command
> > >>>
> > >>> SIGNOFF allstat
> > >>>
> > >>> to [log in to unmask]<mailto:[log in to unmask]>, leaving the subject line blank.
> > >>>
You may leave the list at any time by sending the command
SIGNOFF allstat
to [log in to unmask], leaving the subject line blank.
|