[R] Using t tests

[Mike Marchywka]

( after getting confirmation of lack of posts try again, LOL )

----------------------------------------
> From: marchywka at hotmail.com
> To: r-help at r-project.org
> Subject: RE: [R] Using t tests
> Date: Sun, 10 Jul 2011 10:13:51 -0400
>
>
> ( sorry if this is a repost but I meant to post to list and never received
> any indication it was sent to the list, thanks asking for comments about approach
> to data analysis).
>
>
> > From: marchywka at hotmail.com
> > To: jim at bitwrit.com.au; gwanmesia at aol.com
> > CC: r-help at r-project.org
> > Subject: RE: [R] Using t tests
> > Date: Sun, 10 Jul 2011 07:35:32 -0400
> >
> >
> >
> >
> >
> > ----------------------------------------
> > > Date: Sat, 9 Jul 2011 18:40:43 +1000
> > > From: jim at bitwrit.com.au
> > > To: Gwanmesia at aol.com
> > > CC: r-help at r-project.org
> > > Subject: Re: [R] Using t tests
> > >
> > > On 07/08/2011 07:22 PM, Gwanmesia at aol.com wrote:
> > > > Dear Sir,
> > > >
> > > > I am doing some work on a population of patients. About half of them are
> > > > admitted into hospital with albumin levels less than 33. The other half have
> > > > albumin levels greater than 33, so I stratify them into 2 groups, x and y
> > > > respectively.
> > > >
> > > > I suspect that the average length of stay in hospital for the group of
> > > > patients (x) with albumin levels less than 33 is greater than those with
> > > > albumin levels greater than 33 (y).
> > > >
> > > > What command function do I use (assuming that I will be using the chi
> > > > square test) to show that the length of stay in hospital of those in group x is
> > > > statistically significantly different from those in group y?
> > > >
> > > Hi Ivo,
> > > Just to make things even more complicated for you, Mark's suggestion
> > > that the length_of_stay measure is unlikely to be normally distributed
> > > might lead you to look into a non-parametric test like the Wilcoxon (aka
> >
> > ( please correct any of the following which is wrong, but note that
> > the discusion is more interesting and useful with details of your goals )
> > I'm curious why people still jump to setting arbitrary cutoff points,
> > in this case based on what you happen to have sampled, rather than
> > try to find a functional relationship between the two parametric
> > variables? Generally the thing that separates likely cause from
> > noise is smotthness or something you can at least rationalize
> > in terms of physical mechanisms. If your question relates
> > to the reprodiciblity of a given result (" well this experiment showed
> > hi and low were significantly different on hospital stays, maybe the next
> > experiement will show the same ") you'd probably like to consider
> > the data in relation to possible causes. I'd not sure your disease process
> > would know about your median test results when patients walk in. BTW,
> > what is terminating the hospital stay, cure death or insurance exhaustion?
> > This sounds like you are just trying to reproduce something that is already
> > in the literature:cutoff is on the low side of normal and often hypoprotein
> > is suspected of being bad, that the higher group would be usually expected to do better no? Although
> > I suppose this could have something to do with dehydration etc but the point
> > of course is that data interpretation is difficult to do in a vacuum.
> >
> >
> >
> >
> >
> >
> >
> >
> > > Mann-Whitney in your case) test. You will have to split your
> > > length_of_stay measure into two like this (assume your data frame is
> > > named "losdf"):
> > >
> > > albumin_hilo <- albumin > 33
> > > wilcox.test(losdf$length-of-stay[albumin_hilo],
> > > losdf$length_of_stay[!albumin_hilo])
> > >
> > > or if you use wilcox_test in the "coin package:
> > >
> > > albumin_hilo <- albumin > 33
> > > wilcox_test(length_of_stay~albumin_hilo,losdf)
> > >
> > > Do remember that the chi-square test is used for categorical variables,
> > > for instance if you dichotomized your length_of_stay into less than 10
> > > days or 10 days and over.
> > >
> > > Jim
> > >
> > > ______________________________________________
> > > R-help at r-project.org mailing list
> > > https://stat.ethz.ch/mailman/listinfo/r-help
> > > PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
> > > and provide commented, minimal, self-contained, reproducible code.
> >
>