Message-ID: <55675228.2060206@gmail.com>
Date: 2015-05-28T17:36:40Z
From: Duncan Murdoch
Subject: Issue with 95% CI using MASS{confint}
In-Reply-To: <CACBvRXTxZV3mnTPW1hT=F581Q-vNZ2CWcpGCJcNqu8F1nmXBEQ@mail.gmail.com>
On 28/05/2015 1:11 PM, Jennifer Sweatman wrote:
> Hi all,
>
> Thank you for your responses. Unfortunately, setting level=0.9500001 or
> 0.949999 did not work for me. I followed your code, Duncan, and some of my
> tau values are less than 2. I don't know how to interpret this
> information, however. Can you point me in the direction to a reference?
This is probably discussed in the MASS book, but I don't have a copy at
hand. I learned it from Bates
and Watts' Nonlinear Regression, Ch 6, Graphical Summaries of Nonlinear
Inference Regions. The
idea is that the tau parameter should have roughly a N(0,1) distribution
if the parameter is
correct, so the confidence region corresponds to the set of parameters
where tau is in the
central 95% of that distribution, i.e. +/- 1.96. If tau never goes
outside that range, then you can't limit the parameter.
Duncan Murdoch
>
> Thanks,
>
> Jenn
>
> On Thu, May 28, 2015 at 11:22 AM, Duncan Murdoch <murdoch.duncan at gmail.com>
> wrote:
>
> > On 28/05/2015 9:08 AM, sami.toppinen at kolumbus.fi wrote:
> >
> >> I encountered the same problem today. It seems to occur only when you
> >> use exactly 95 % as the confidence level. For example:
> >>
> >> > confint(g1)
> >> Waiting for profiling to be done...
> >> 2.5% 97.5%
> >> a0 1.257512 1.330881
> >> KP NA 163.862932
> >> > confint(g1, level = 0.95)
> >> Waiting for profiling to be done...
> >> 2.5% 97.5%
> >> a0 1.257512 1.330881
> >> KP NA 163.862932
> >> > confint(g1, level = 0.950001)
> >> Waiting for profiling to be done...
> >> 2.5% 97.5%
> >> a0 1.257512 1.330881
> >> KP 125.814534 163.862996
> >> > confint(g1, level = 0.949999)
> >> Waiting for profiling to be done...
> >> 2.5% 97.5%
> >> a0 1.257512 1.330881
> >> KP 125.814709 163.862802
> >>
> >
> > That's very strange. It would be nice to see a reproducible example.
> >
> > Duncan Murdoch
> >
>
>
>