pcount- abundance and beta estimates

[sulman]

hi all-
I want to look at shorebird abundance using three survey covariates. There are 2 seasons, 3 plots, and 4 veg class.
For 4 of my species the top model was 'season.vegp' and I was able to have predicted estimates for each plot*vegetation.class interaction.
For 2 species the top model was the global. The beta estimates are different, but the predicted abundance is the same at each of the 12 plot and vegetation.class combintations
For the global model, is there a way to get abundance estimates at each plot and vegetation combination? maybe not. what am I missing?
any guidance is greatly appreciated.
thanks for your time.
-Sadie

> grye.abun.global <- pcount (~season+plot+vegetation.class ~1, grye.abun, K=50)

> (ms1 <- modSel(grye.fitL))
                         nPars    AIC      delta   AICwt cumltvWt
global               8         793.74     0.00 1.0e+00     1.00
season.vegp   14        807.40    13.65 1.1e-03     1.00
season             3         831.22    37.48 7.3e-09     1.00
plot                  4         864.53    70.79 4.3e-16     1.00
vegetation.class  5     876.24    82.49 1.2e-18     1.00
null                 2         896.09    102.35 6.0e-23     1.00

> summary(grye.abun.global)

Call:
pcount(formula = ~season + plot + vegetation.class ~ 1, data = grye.abun,
    K = 50)

Abundance (log-scale):
 Estimate    SE    z P(>|z|)
    0.351 0.174 2.02  0.0436

Detection (logit-scale):
                  Estimate    SE     z  P(>|z|)
(Intercept)         -0.981 0.310 -3.16 1.55e-03
seasonS             -2.510 0.374 -6.71 1.97e-11
plotM               -0.806 0.279 -2.89 3.90e-03
plotN               -2.269 0.540 -4.20 2.65e-05
vegetation.classB    0.494 0.290  1.71 8.77e-02
vegetation.classD   -1.335 0.521 -2.56 1.04e-02
vegetation.classE   -1.604 0.590 -2.72 6.57e-03

AIC: 793.745
Number of sites: 270
optim convergence code: 0
optim iterations: 35
Bootstrap iterations: 0
> predict(grye.abun.global, newdata=new.data, type="state")
   Predicted        SE    lower    upper
1   1.420812 0.2473426 1.010078 1.998565
2   1.420812 0.2473426 1.010078 1.998565
3   1.420812 0.2473426 1.010078 1.998565
4   1.420812 0.2473426 1.010078 1.998565
5   1.420812 0.2473426 1.010078 1.998565
6   1.420812 0.2473426 1.010078 1.998565
7   1.420812 0.2473426 1.010078 1.998565
8   1.420812 0.2473426 1.010078 1.998565
9   1.420812 0.2473426 1.010078 1.998565
10  1.420812 0.2473426 1.010078 1.998565
11  1.420812 0.2473426 1.010078 1.998565
12  1.420812 0.2473426 1.010078 1.998565

[Jeffrey Royle]

I think you can do this with the raneff() function -- check that out.

andy

[Richard Chandler]

Hi Sadie,

How are you treating season? This model assumes population closure.

Also, can you show a summary of grye.abun? I'm not sure if you only have 3 sites, or if there were multiple sites within each plot. If there were only 3 sites, then I don't think this model will work very.

In general, if you want estimates of abundance at each site, you can do something like:

re.global <-  ranef(grye.abun.global)
plot(re.global)
bup(re.global, "mode")
confint(re.global)

Or, if you want predictions, you need to create a new data.frame in which each row has the covariate combinations of interest, and then use predict. But in your case, it doesn't appear as though you included any covariates of abundance (only detection), so expected abundance will be constant among sites.

Richard

_____________________________________
Richard Chandler, post-doc
USGS Patuxent Wildlife Research Center
301-497-5696

From:	sulman <sadie...@gmail.com>
To:	unma...@googlegroups.com
Date:	04/12/2012 08:41 AM
Subject:	[unmarked] pcount- abundance and beta estimates
Sent by:	unma...@googlegroups.com

---

[sulman]

Thanks for the comments awhile back. I have come up with some more
questions/concerns that I hope you can help me with.
Here is the summary you wanted to see:

> summary(grye.abun)
unmarkedFrame Object

270 sites
Maximum number of observations per site: 6
Mean number of observations per site: 3.42
Sites with at least one detection: 30

Tabulation of y observations:
0 1 2 3 4 5 6 7 8 10 11 15 <NA>
883 21 7 2 1 2 1 1 1 1 2 1 697

Site-level covariates:
season vegetation.class plot
F:153 A: 89 L: 98
S:117 B:106 M:114
D: 38 N: 58
E: 37
>
…………
regarding treatment of season…..
Maybe I am not dealing with this properly. I have Spring and Fall, and
would like to look at both abundance and detectability in each season,
at each of the plot*vegetation.class combinations. My set-up is
this….I have 270 ‘sites’ which are transects. Each of these ‘sites’
have up to 6 ‘visits’- which are separate point counts completed in
the same day. 153 of the ‘sites’ were visited in the Fall, and 117 in
the Spring . I was going with pcount because my ‘visits’ definitely
occur within closed population, but maybe not? Should I be using
pcountOpen in a migration survey situation?
--------
I did use the predict() to show abundance at each
plot*vegetation.class combination. I have read many threads, and I am
wondering if I am interpreting these predicted numbers correctly. Is
1.65397 the estimated mean density at Plot L, Veg A ? This number does
seem accurate from what I saw during surveys.

> new.pred=predict(grye.abun.global, newdata=new.data, type="state")
> temp=data.frame(plot=newplot,veg=newveg, est=new.pred$Predicted, SE=new.pred$SE, l95=new.pred$lower, u95=new.pred$upper)
> temp
plot veg est SE l95 u95
1 L A 1.65397289 0.39550502 1.03510751 2.6428427
2 L B 1.77604147 0.43172772 1.10290832 2.8600050
3 L D 0.47121677 0.22610536 0.18398658 1.2068557
4 L E 0.43670425 0.23315824 0.15336538 1.2435049
5 M A 1.27118367 0.32515513 0.76998116 2.0986331
6 M B 1.36500117 0.31010597 0.87448816 2.1306500
7 M D 0.36216015 0.17507513 0.14041503 0.9340879
8 M E 0.33563507 0.17781203 0.11882822 0.9480147
9 N A 0.31576743 0.15397029 0.12142715 0.8211431
10 N B 0.33907209 0.16220366 0.13276952 0.8659358
11 N D 0.08996212 0.05772674 0.02557774 0.3164151
12 N E 0.08337318 0.05591909 0.02239331 0.3104091


However, the estimates of detection are WAY to low.
This is of concern, because the detections are very low…. I was
surveying 100m radius area on a mudflat. My detection of a greater
yellowlegs in the field should be very close to 1, if not 1; whereas
detection is showing 0.07 for spring, and 0.23 for fall. (if I
interpret that correctly)

> plogis(coef(grye.abun.global, type="det"))
p(Int) p(seasonS)
0.23487372 0.07275806

Maybe I am calculating the detection incorrectly?
I wonder if low detection in pcount (but shouldn’t be in actual
survey) has something to do with these species being gregarious, and
the ‘high’ counts of 10, 11, 15, etc may correlate with flocks
(clusters). As of a June 2010 thread, cluster size was not a part of
unmarked. Has this changed?

Also- there was a thread on schooling coho salmon and concern on non-
independence. Because shorebirds may be in small flocks (2-15 on my
field site) during migration, maybe I am seeing this with my data?
i.e. do I need to think about non-independent detection (Martin et
al. 2011), and possibly looking at a beta-binomial mixture model as
opposed to N-mixture model?
maybe pcount is not best to use for estimating abundance/detection on
migrating shorebirds in 2 different seasons.

Thanks for any thoughts. I am concerned I may have an overestimation
of abundance, and of all the things I have read, I am not sure what
exactly could be my problem.

[Dan Linden]

Sadie, remember that detection probability is a function of detectability and availability.  Your survey may afford perfect detectability (i.e., if the bird is there, you will definitely detect it), but if the birds are not always available to be detected then estimates of detection probability will be less than 1 (much less than 1 if availability is low).  If this is the case, your abundance estimates are more accurately interpreted as the number of birds using the sites (as opposed to occupying the sites).  In which case, the model is suggesting that there are more birds using the sites than a naive estimate from the survey data would indicate.