Re: Which truncation is best?
145 views
Skip to first unread message
Eric Rexstad
Jun 3, 2014, 3:12:08 PM6/3/14
to distance...@googlegroups.com
All good questions Delyth:
As you probably know, there is no perfect answer to the question "what
is the proper truncation distance"?
Happy you consulted the archives to discover what the monotonicity
warning was about and that you took action to truncate detections that
caused there to be a rise in detections at large distances.
The heart of the matter is what to do about "parameter 2 is at a
bound". That warning is cause for concern because it suggests that the
hazard rate function that fitted best was one in which the beta
coefficient (pp 46-47 of Buckland et al. (2001)) was as small as
Distance permitted. That is usually an indication of an unreliable
estimate.
Look at many of the example histograms in Chapter 4 of Buckland et al.
(2001): Fig 4.1, 4.4, 4.5; all have bars that exceed 1.0. Remember the
histograms are of a sample of perpendicular distances. Distance is
attempting to fit data that best describe the process by which
detectability diminishes as a function of perpendicular distance. Do
not be terribly concerned about occasional bins above 1.
One diagnostic you have not mentioned is goodness-of-fit. Include that
consideration in your decision process regarding model choice.
On 21/05/2014 15:11, Delyth Williams wrote:
> > Dear all,
> >
> > I have run DISTANCE analysis on my data set and I have a few queries regarding which truncation is best. I have set the data filter to discard the top 10% of observations and the model with the lowest AIC and delta AIC was the Hazard rate with simple polynomial expansion. However, I did get a couple of warnings, including the "Parameter two is at a lower bound" and "Parameters are being constrained to obtain monotonicity".
> >
> > Going through the archives, I found the discussions concerning the second warning and I saw that the last bar on the histogram was higher than the previous one, so I re-truncated the data to "discard all observations beyond 24m". Re-running the analysis, the model with the lowest AIC and delta AIC was now the uniform function with the cosine polynomial expansion and it ran without any warnings. However, the first bar on the histogram exceeds 1.0 detection probability which seems a little bizarre to me!
> >
> > The global density estimates do not differ a huge amount (2.917individuals/Km2 for the hazard rate and 2.167 individuals/km2 for the uniform) but I am wondering which is the better one to select?
> >
> > Many thanks, I appreciate your time and advice!
> >
> > Delyth
> >
> > --
> > distance-sampling list home page:
> > http://www.jiscmail.ac.uk/lists/distance-sampling.html
> > To unsubscribe, go to the home page and click "Join or leave the list",
> > or email jisc...@jiscmail.ac.uk with the following message body:
> > leave distance-sampling
> > Before posting to the list, read the list guidelines at
> > http://www.jiscmail.ac.uk/files/distance-sampling/
> > distance-sampling.welcome
> > Any questions, email the list owner at
> > distance-sam...@jiscmail.ac.uk
> > --
> >
--
Eric Rexstad
Research Unit for Wildlife Population Assessment
Centre for Research into Ecological and Environmental Modelling
University of St. Andrews
St. Andrews Scotland KY16 9LZ
+44 (0)1334 461833
The University of St Andrews is a charity registered in Scotland : No SC013532
--
distance-sampling list home page:
http://www.jiscmail.ac.uk/lists/distance-sampling.html
To unsubscribe, go to the home page and click "Join or leave the list",
or email jisc...@jiscmail.ac.uk with the following message body:
leave distance-sampling
Before posting to the list, read the list guidelines at
http://www.jiscmail.ac.uk/files/distance-sampling/
distance-sampling.welcome
Any questions, email the list owner at
distance-sam...@jiscmail.ac.uk
--
As you probably know, there is no perfect answer to the question "what
is the proper truncation distance"?
Happy you consulted the archives to discover what the monotonicity
warning was about and that you took action to truncate detections that
caused there to be a rise in detections at large distances.
The heart of the matter is what to do about "parameter 2 is at a
bound". That warning is cause for concern because it suggests that the
hazard rate function that fitted best was one in which the beta
coefficient (pp 46-47 of Buckland et al. (2001)) was as small as
Distance permitted. That is usually an indication of an unreliable
estimate.
Look at many of the example histograms in Chapter 4 of Buckland et al.
(2001): Fig 4.1, 4.4, 4.5; all have bars that exceed 1.0. Remember the
histograms are of a sample of perpendicular distances. Distance is
attempting to fit data that best describe the process by which
detectability diminishes as a function of perpendicular distance. Do
not be terribly concerned about occasional bins above 1.
One diagnostic you have not mentioned is goodness-of-fit. Include that
consideration in your decision process regarding model choice.
On 21/05/2014 15:11, Delyth Williams wrote:
> > Dear all,
> >
> > I have run DISTANCE analysis on my data set and I have a few queries regarding which truncation is best. I have set the data filter to discard the top 10% of observations and the model with the lowest AIC and delta AIC was the Hazard rate with simple polynomial expansion. However, I did get a couple of warnings, including the "Parameter two is at a lower bound" and "Parameters are being constrained to obtain monotonicity".
> >
> > Going through the archives, I found the discussions concerning the second warning and I saw that the last bar on the histogram was higher than the previous one, so I re-truncated the data to "discard all observations beyond 24m". Re-running the analysis, the model with the lowest AIC and delta AIC was now the uniform function with the cosine polynomial expansion and it ran without any warnings. However, the first bar on the histogram exceeds 1.0 detection probability which seems a little bizarre to me!
> >
> > The global density estimates do not differ a huge amount (2.917individuals/Km2 for the hazard rate and 2.167 individuals/km2 for the uniform) but I am wondering which is the better one to select?
> >
> > Many thanks, I appreciate your time and advice!
> >
> > Delyth
> >
> > --
> > distance-sampling list home page:
> > http://www.jiscmail.ac.uk/lists/distance-sampling.html
> > To unsubscribe, go to the home page and click "Join or leave the list",
> > or email jisc...@jiscmail.ac.uk with the following message body:
> > leave distance-sampling
> > Before posting to the list, read the list guidelines at
> > http://www.jiscmail.ac.uk/files/distance-sampling/
> > distance-sampling.welcome
> > Any questions, email the list owner at
> > distance-sam...@jiscmail.ac.uk
> > --
> >
--
Eric Rexstad
Research Unit for Wildlife Population Assessment
Centre for Research into Ecological and Environmental Modelling
University of St. Andrews
St. Andrews Scotland KY16 9LZ
+44 (0)1334 461833
The University of St Andrews is a charity registered in Scotland : No SC013532
--
distance-sampling list home page:
http://www.jiscmail.ac.uk/lists/distance-sampling.html
To unsubscribe, go to the home page and click "Join or leave the list",
or email jisc...@jiscmail.ac.uk with the following message body:
leave distance-sampling
Before posting to the list, read the list guidelines at
http://www.jiscmail.ac.uk/files/distance-sampling/
distance-sampling.welcome
Any questions, email the list owner at
distance-sam...@jiscmail.ac.uk
--
Message has been deleted
Delyth Williams
Jun 3, 2014, 3:19:20 PM6/3/14
to distance...@googlegroups.com
Dear all,
I have run DISTANCE analysis on my data set and I have a few queries rega=
rding which truncation is best. I have set the data filter to discard the=
top 10% of observations and the model with the lowest AIC and delta AIC =
was the Hazard rate with simple polynomial expansion. However, I did get =
a couple of warnings, including the "Parameter two is at a lower bound" a=
warning and I saw that the last bar on the histogram was higher than the=
previous one, so I re-truncated the data to "discard all observations be=
yond 24m". Re-running the analysis, the model with the lowest AIC and del=
ta AIC was now the uniform function with the cosine polynomial expansion =
and it ran without any warnings. However, the first bar on the histogram =
s/Km2 for the hazard rate and 2.167 individuals/km2 for the uniform) but =
I have run DISTANCE analysis on my data set and I have a few queries rega=
rding which truncation is best. I have set the data filter to discard the=
top 10% of observations and the model with the lowest AIC and delta AIC =
was the Hazard rate with simple polynomial expansion. However, I did get =
a couple of warnings, including the "Parameter two is at a lower bound" a=
nd "Parameters are being constrained to obtain monotonicity".
Going through the archives, I found the discussions concerning the second=
warning and I saw that the last bar on the histogram was higher than the=
previous one, so I re-truncated the data to "discard all observations be=
yond 24m". Re-running the analysis, the model with the lowest AIC and del=
ta AIC was now the uniform function with the cosine polynomial expansion =
and it ran without any warnings. However, the first bar on the histogram =
exceeds 1.0 detection probability which seems a little bizarre to me!
The global density estimates do not differ a huge amount (2.917individual=
s/Km2 for the hazard rate and 2.167 individuals/km2 for the uniform) but =
Reply all
Reply to author
Forward
0 new messages