entry data without distances vs transects with no observations

[Ana María Prieto]

Dear distance-samplers,

I have a data set of one survey, single obsever and line transect study. For several transects I have no observations, and then I leave the space in the distance column empty, as the user's guide describes.
In the guide it says also that is possible to include observations in which the distance is missing, apply data filter to estimate a global detection function using only observations that do have a distance value, and then estimate density as desired -global or stratified- using all observations. However if any distance value is written, and it was the only observation of a given transect, how can I entry the data in Distance in order to differentiate between transect with no observation and observation with missing distance?

Thanks in advance!

Best,
Ana

[Eric Rexstad]

Ana

The trick is for detections without distances, you provide a fictitious distance.  However, you do not want this fictitious distance to be used in the fitting of the detection function (hence use the data filter with a criterion that specifies certain detections--with fictitious distances) are not to be used in the analysis.

You then perform a second analysis of the data, this time without the filter such that all data are included in the analysis, but detection probability is provided as a multiplier, and not computed in this second analysis (you specify a uniform key with no adjustments).

Note the additional caveats in this users guide description:

"Note an important assumption here is that the missing distances are missing at random – for example it will not work if you are less likely to record the distance for objects farther from the line or point.  For this reason, the first approach is probably safer.  Note also that this approach won’t work if you use stratification – in that case you’ll need more than one multiplier (one for each stratum), and will have to calculate the global density estimate by hand. "

The advice is to discard observations with missing distances unless there are exceptional circumstances.

--
You received this message because you are subscribed to the Google Groups "distance-sampling" group.
To unsubscribe from this group and stop receiving emails from it, send an email to distance-sampl...@googlegroups.com.
To post to this group, send email to distance...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/distance-sampling/af3442f5-f61e-4821-906f-63f4a2a53abc%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

-- 
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

[Ana María Prieto]

Dear Eric,

thanks for your answer and advice. My detections without distances are completely random so I proceed to calculate density as you suggested. Because my intial idea was anyway to have a global detection function, i used the multipliers with a uniform key, and specified global detection and stratified density.  It worked out, the model run well and i did't get any warnings. However, I'm still getting very high values for the coefficient of variation, which was my motivation for increasing the sample including those detections for which i didn't have a distance. The smallest value that I get is 30%. I don't really know what else I can do to improve estimation of density with a dataset in which most of the strata have several transects with zero detections.

I'll appreciate very much any further idea or hint regarding this issue.

Best regards,

Ana.

2017-12-11 14:13 GMT+01:00 Eric Rexstad <er...@st-andrews.ac.uk>:

Ana

The trick is for detections without distances, you provide a fictitious distance.  However, you do not want this fictitious distance to be used in the fitting of the detection function (hence use the data filter with a criterion that specifies certain detections--with fictitious distances) are not to be used in the analysis.

You then perform a second analysis of the data, this time without the filter such that all data are included in the analysis, but detection probability is provided as a multiplier, and not computed in this second analysis (you specify a uniform key with no adjustments).

Note the additional caveats in this users guide description:

"Note an important assumption here is that the missing distances are missing at random – for example it will not work if you are less likely to record the distance for objects farther from the line or point.  For this reason, the first approach is probably safer.  Note also that this approach won’t work if you use stratification – in that case you’ll need more than one multiplier (one for each stratum), and will have to calculate the global density estimate by hand. "

The advice is to discard observations with missing distances unless there are exceptional circumstances.

On 08/12/2017 15:49, Ana María Prieto wrote:

Dear distance-samplers,

I have a data set of one survey, single obsever and line transect study. For several transects I have no observations, and then I leave the space in the distance column empty, as the user's guide describes.
In the guide it says also that is possible to include observations in which the distance is missing, apply data filter to estimate a global detection function using only observations that do have a distance value, and then estimate density as desired -global or stratified- using all observations. However if any distance value is written, and it was the only observation of a given transect, how can I entry the data in Distance in order to differentiate between transect with no observation and observation with missing distance?

Thanks in advance!

Best,
Ana

--
You received this message because you are subscribed to the Google Groups "distance-sampling" group.

To unsubscribe from this group and stop receiving emails from it, send an email to distance-sampling+unsubscribe@googlegroups.com.
To post to this group, send email to distance-sampling@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/distance-sampling/af3442f5-f61e-4821-906f-63f4a2a53abc%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

-- 
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

--

Ana María Prieto Ramírez
PhD student

Departement of Conservation Biology

Center for Environmental Research UFZ, Leipzig

Zoologisches Forschungsmuseum Alexander Koenig

University of Bonn

[Eric Rexstad]

Ana

It may very well be the case that there is great uncertainty in your estimates of lizard density and a CV of 30% is the best that can be achieved under the design you implemented.  Animals that have clumped distributions lead to high variability in encounter rates between transects; some transect hit animal clumps and other transect do not hit clumps.  Higher levels of replication are needed for such populations.

If you implemented a systematic (rather than completely random) placement of transects AND there is a trend in lizard density through your study area, you might produce a slightly better estimate of encounter rate variance by using estimator S2 or O2 under "advanced encounter rate variance"  (see screenshot).  Read the caveats regarding specific data setup requirements for the O2 estimator.

Read the users guide section entitled "Advanced analytic variance estimation in CDS" in chapter 8 on conventional distance sampling analysis.  As noted, this is an advanced topic.