www.phidot.org

Hi Everyone,

This is a first time post by a novice user of DENSITY. I carried out removal trapping of ship rats using a trapping web (5 webs) of 81 traps (16 spokes, 5 traps per spoke, 20m trap spacing, 1 centre trap). Number of rats caught each night declined over time (5 nights for each session). I am more interested in estimating density for each session than overall (but am happy to pool detection probabilities across sessions). I realise this is not the best way to model animal density, as there is not really much to model, but this is all I have. In DENSITY I use closed population with the Huggins estimator.

Everything works quite well...until I add sex as a covariate, which either increases confidence intervals, or makes capture probability = 1.0, or the model fails completely (see annotated output below, based on 2 webs). I had thought that time and sex (males and females have different home range sizes) should be included, but looking at the AIC and the confidence intervals, it seems excluding time and sex is "better". I also ran models with sex as session, which resulted in similar N-hat values, but wider CIs, given that my data became more sparse for each session.

Does anyone have any ideas on why sex might be throwing the model out so much? And can anyone please help me with understanding the effects of time=none and time=time specific? I understand these are probably novice questions, but any help is greatly appreciated!

I have started looking at possible R script, but I have a long way to go before I figure all of that out (actually, can secr handle removal trapping?)

Code: Select all

Trap type         : Single kill
Input format      : TrapID
Habitat mask      : None
Session filter    : ALL
Occasion filter   : ALL
Capture filter    : ALL
Area units        : ha
Confidence level  : 95% (alpha = .05)

CLOSED POPULATION METHODS
  Relative upper bound Nhat : 100
  Losses on capture         : Ignore
  Confidence interval  Nhat : Profile likelihood
  Warning : PLI available only for full MLE estimators (not linear-logit model, jackknife etc.)
            Default interval type is Lognormal

HUGGINS LINEAR LOGIT MODEL
Behavioural response      Removal

TIME EFFECT-NONE, NO COVARIATES - logit (p) = beta0
Session NCapture NAnimal  CPAIC  CPNhat SE.CPNhat LC.CPNhat UC.CPNhat   CPphat
    1       41      41    122.66  47.83      5.59     42.68     68.82   0.3225   
    2       51      51    104.60  52.40      1.54     51.24     59.04   0.5959   

TIME SPECIFIC, NO COVARIATES - logit (p) = beta0 + beta_t[j]
Session NCapture NAnimal  CPAIC  CPNhat SE.CPNhat LC.CPNhat UC.CPNhat   CPphat   
    1       41      41    126.80 50.14      6.68     43.54     73.94   0.3043   
    2       51      51    107.89 55.14      4.04     51.83     71.56   0.4204   

TIME EFFECT-NONE, SEX AS COVARIATE - logit (p) = beta0 + beta1.zi[i,1]
Session NCapture NAnimal CPAIC CPNhat SE.CPNhat LC.CPNhat UC.CPNhat   CPphat
    1       41      41   124.06  48.95      7.53     42.66     79.17   0.3186   
    2       51      51     NA    NA        NA        NA        NA       NA   

TIME-SPECIFIC, SEX AS COVARIATE - logit (p) = beta0 + beta1.zi[i,1] + beta_t[j]
Session NCapture NAnimal  CPAIC  CPNhat SE.CPNhat LC.CPNhat UC.CPNhat   CPphat 
    1       41      41    128.08 41.00      0.00     41.00     41.00   1.0000
    2       51      51      NA    NA        NA        NA        NA       NA

Just thought I'd include a sample of my data layout also:

TRAP LAYOUT

A1 18.5 -7.7
A2 37.0 -15.3
A3 55.4 -23.0
A4 73.9 -30.6
A5 92.4 -38.3
B1 14.1 -14.1
........

CAPTURE DATA (session, rat ID, occasion, trapID, sex [1=male,2=female])
1 r1 5 A2 1
1 r10 1 E5 1
1 r11 2 F3 1
1 r12 1 F4 2
1 r13 2 F5 1
1 r14 3 F5 1
1 r15 4 F5 2
1 r16 5 F5 2
1 r17 1 G2 1
1 r18 1 G4 2
1 r19 2 G4 1
........

Hi
I can't say anything definite. It's sad you used a trapping web as that tends to inflate heterogeneity from differential access to traps (the web is designed for high capture probability near the centre), and as you have only a kill sample you cannot model that heterogeneity. Density 5.0 does not estimate density from this sort of data. Presumably you were thinking of a distance analysis?
Murray

Hi Murray,

Many thanks for your reply. Yes originally I was going to implement many trapping webs and use distance analysis, but alas the project shifted focus (for various other reasons) and I ended up only trapping at 5 sites (in different areas and at different times), so the paucity of data now does not lend itself to DISTANCE either. Although, I'm not sure it ever would have as the detection probabilities increase as one moves away from the web centres, because of the high number of animals trapped on the outside rings.

It looks like I may be out of luck then, unless there are any other ideas?

I do need to report these results in some format, so perhaps I'll just have to be very careful in how I present them.

Some additional info:
On all webs, by the final night the number of captures were generally zero or close to zero (compared to c. 30-40 captures on night 1), so it 'seems' like most of the rats in the area were trapped. Linear regression of nightly catch vs cumulative catch have ok fits (R squared 0.7 - 0.9) and these yield almost identical density estimates as those produced by DENSITY using the Huggins estimator. To be honest I'm not sure what this says about the data, if heterogeneity between traps is not being accounted for.

Just as an additional point - if I tell DENSITY the trapping was in a grid layout rather than a web layout, the estimates are identical and the same issues persist when trying to add sex as a covariate.