New features and changes in Density 3.2

General
Input
Habitat masks
Output
Closed population estimation
Inverse prediction to estimate density
Changed defaults for inverse prediction
Open population analysis
Simulation (Power analysis)
Trap builder
Monte Carlo tests for serial correlation
Bugs fixed in Density 3.2
Known bugs and warnings
Minor new features
Discontinued options
Implementation in progress
Acknowledgements

General

· Lower memory requirements
· Now possible to open more than one instance of Density at a time
· Text file viewer now has search and edit functions
· Simple Distance interface for ‘trapping webs’ and ‘passive transects’
· Simple Capture interface
· Changed defaults for several settings – see separate heading below

Input

·  Input data may include blank lines and comments (preceded by # or ;)
· Coordinates of trap locations may now take real (non- integer) values
· Non-spatial capture data allowed (closed population analysis is restricted to estimation of population size).  Only the first three columns of data are read (SessionID, AnimalID, Occasion).
·

Option to import data in ‘capture history’ format  (Data | Import capture histories)  

· The format for target data has been simplified  
· Data filters allow analyses to be restricted to a subset of capture data, and sessions or occasions to be pooled:
o       Sessions  e.g. ‘[1-3]’ to pool data from three different sessions
o       Occasions e.g. ‘[1 3] 4 7’ for three new occasions 1¬(1+3),      2¬4,      3¬7.
o       Captures Include only captures whose Notes field contains the given text (or exclude these  captures if the text is preceded by ‘!’). Case insensitive.

Habitat masks

· Habitat masks may be imported as polygons in ESRI shapefiles
· Preview (Options | Habitat mask)  
· Warning when mask not compatible with trap layout
·   Interface with MS Paint for editing bitmap masks
· Habitat mask now applied in PowerAnalysis (but needs more work)

Output

· Fields printed on output now chosen by the user
· Flat or Stacked tabular output formats
· Tables optionally tab-delimited for easy cut & paste to Excel
·  Improved printing of text files (margins, orientation, selected text)

Closed population estimation

· Profile likelihood confidence intervals for ML N-hat
· Graphs of profile likelihood for population size MLE
· Graphs of fitted distributions of Pi for model M(h) (2-point, beta)
· Default asymmetric (lognormal) confidence intervals for all non-ML estimators
· AIC values for MLE (N)
· Additional estimator: beta-binomial for model Mh (Dorazio & Royle 2003)
· Automatic Capture model selection on ‘Read data’ (result written to log file)
· Closure test of Otis et al. (1978)
· All estimates of N are subject to an upper bound set in Options | Miscellaneous
· Explicit handling of losses on capture (‘Ignore’, ‘Discard’, ‘Add back in’, see Options | Miscellaneous)

Inverse prediction to estimate density

· Root pooled spatial variance (RPSV) more robust spatial measure than d-bar
· Choice of experimental design: fractional factorial or full factorial or simplex
· Number of centre points controlled by the user
· Number of replicates controlled via max(CV(yi-bar))
· Two-phase model fitting (refinement of initial estimate with different design)
· Check on planarity, and indication of maximum expected bias
· Transformations to improve planarity and reduce bias in fitted model
· Confidence ellipses for pairs of parameters
· SE adjusted for spatial variance
· Inferred boundary strip width W

Changed defaults for inverse prediction in Density 3.2

·  Auto g0 adjustment 1.0 (previously 1.0 in version 2.1, 0.5 in version 3.1)
·  Population dispersion  Poisson (previously Even)
·  Home range measure RPSV (previously dbar)
·  Experimental design Full factorial (fractional in version 3.1)
·  Centre points in design 3 (previously zero)
·  Maximum replicates Phase I 2000 (500 in version 3.1)  

Open population analysis

Density performs simple multi-session analyses with a variety of models. These include  the closed-form Jolly-Seber methods described by Pollock et al. (1990) (‘Direct’), maximisation of the Cormack-Jolly-Seber likelihood (‘CJS’) or the Pradel (1996) likelihood with γ parameterisation (‘Pradel gamma’), and reduced-parameter versions of the CJS model.

First check Options | Miscellaneous | ‘Enable multi-session analysis’ and then select the desired model under Options | Open population. Note that when Model = ‘Direct’ estimates of survival f and seniority γ are not constrained to lie between 0 and 1. Pradel’s γ and the derived measure of rate of population change (λt) are available both from the Direct method (capture histories are automatically reversed to get γ), from the ‘Pradel gamma’ method, and when a supplementary ‘reversed CJS’ model is used in tandem with CJS to estimate γ.

 Bootstrap confidence intervals may be calculated. See Options | Open population for bootstrap n. The a-level is controlled on the Options | Output page (e.g. a = 0.05 for 95% confidence interval). Bootstrap confidence intervals override the default type in Options | Output.

Simulation (Power analysis)

Density simulates trapping by applying a specified detection model to a population with known parameters. Version 3.2 expands the range of simulation options.

·        New detection models
o       Non-spatial simulation
o       Random walk detection
·        New sources of variation
o       Heterogeneous g(0), σ (LogitNormal, Beta, Lognormal)
o       Preview of univariate distributions (double click on mean or CV)
o       Heterogeneous trap-specific g(0) (Advanced)
o       Learned response to capture at particular sites (General - more).
o       Interference by non-target species or processes (Disturbance)
o       Clustered populations (Neyman-Scott 2-D point process)
·        Distance interface for ‘trapping webs’ and ‘passive transects’
·        Better control of simulations
o       Trapping & Estimation stack (allows crossing of population & trap options)
o       Save of capture data with metadata in a binary file for later reanalysis [disabled 22 12 04]
o       ‘Sleep’ tool button to leave simulations running in the background (Windows priority ‘below normal’) and continue with other work
·        Reporting
o       Many new output fields
o       Approximate local density of realised 2-D pattern now recorded (TrueD)
o       Histograms of any output field
o       Open population estimates for multi-session simulations

Trap builder

The trap builder creates trap layout files from building blocks with a standard geometry – e.g. trapping grids, webs or lines. New options are:

· Trapping units replicated across the landscape at even x- and y-spacings
· Array rotation – independent rotation of each trapping unit
·  File import – import a user-defined unit geometry from an existing trap layout file
· Labelling options – flexible trap numbering for grids and webs
· ‘Passive transects’ – automatically generate field designs for method of Lukacs, Franklin and Anderson (In press)

Monte Carlo tests for serial correlation of capture locations

Serial correlation invalidates the ‘mean recapture distance’ (d-bar) measure of home range. Two tests are implemented.

·   Pooled version of Schoener’s statistic (t2/r2) (cf Swihart & Slade 1985, 1987)
·  Proportion of recaptures in same trap (P(zero))

Bugs fixed in Density 3.2

· Erroneous SE of ML estimates for model Mb (Zippin) [Zippin still not 100% reliable cf CAPTURE]
· CV(capture probability) plotted alongside contour display was inaccurate
· Radial histogram of recapture directions gave erroneous results

Known bugs and warnings

·   Filters occasionally give spurious 'no data' message – please report details
·   Large values of g(0) (g(0) > 0.75) may cause problems – please report details
·   ‘Power analysis’
o       ‘Save captures’ is not fully tested [currently disabled altogether 10 Dec 04]
o       ‘Random walk detection’ cannot be used with individual heterogeneity in g(0), sigma etc.
o       Clusters do not persist between sessions in open-population simulations
o       Habitat masks may not always function as expected
o       Some statistics not fully implemented (Cov%, RMSE, RB%)

Minor new features and changes

· 'Simulate one sample' popup menu option under ‘Animals’ tool button  
· Split capture display (Options | Miscellaneous; display of captures uses different colour for later occasions)
·  Movement summary
o       Histogram of distances
o       Improved directional histogram
·  Graph predicted pdf of Pxy given trap layout and detection function (g(0), σ) (Use ‘Pxy contours’ toolbutton to map Pxy and right-click on ‘Capture probability’ plot for PDF option).
· Ran3 pseudorandom number algorithm of Press et al. (1989) (optional)
· Import capture data in compressed XY format (Session AnimalID x1y1x2y2x3y3...xtyt)
·  Import or export capture histories in MARK input format (used also by M-SURGE)
· Data tools do not overwrite old files
·  Windows priority class may be altered within Density (Windows 2000, XP)
· Tools | Spot simulations more flexible & robust
·  Coefficients of inverse prediction multivariate linear model output to log
·  Terminology: 'sample' is replaced with the more standard 'occasion'
·  More detailed reporting in log file
· MMDM (Compute ‘mean maximum distance moved’)
·  Simulations for inverse prediction may take account of known variation in detection effort on each occasion (see Options | Input to specify effort).
·  Choice of hazard or probability parameterisation for detection models

Discontinued options

·   ‘Last solution’ is no longer offered as an initial-values option
·  The Monte Carlo test for serial correlation of locations described by Efford (2004) has been replaced by two other Monte Carlo tests that appear to be more sensitive. One uses the proportion of recaptures at the same site (Pzero) and the other uses an adaptation of Schoener's t2/r2 statistic.

Implementation in progress

· Area detectors (encounters are not restricted to fixed points (traps), but occur with uniform probability throughout a polygon)
·  MLE for density (with Dave Borchers).   

Acknowledgements

Many thanks to Deb Wilson, Dave Ramsey, Ed Debevec, and Grant Norbury for bug reports and ideas, and to Dave Fletcher for statistical support.