unmarked single season occu interpretation help

223 views
Skip to first unread message

Kelly

unread,
Jul 30, 2019, 11:17:52 AM7/30/19
to unmarked
Hello, 

I am a master's student in the crux of thesis crunch time. I have been working on intpretting the results of my code, and am finding some troubling numbers. Due to studying elusive species over the short duration of a master's, these numbers aren't going to be friendly to begin with. But they seem completely off for some of the species I am analyzing. 

In addition, general interpretation--e.g. how literature discusses these values, is not innately obvious. 

Data: I have site and observational covariates; have to run each possible combination (assigned as fm_#). Then a model selection table is run in order to find the "best fit" for each species. 

For this example, the "best fit' is fm_6:
Call:
occu(formula = ~Habitat_classification + Basal_area_hectare ~ 
    Trap_days, data = umf)

Occupancy:
            Estimate     SE      z P(>|z|)
(Intercept)   0.4568 1.9765  0.231   0.817
Trap_days    -0.0115 0.0201 -0.571   0.568

Detection:
                                                            Estimate   SE              z            P(>|z|)
(Intercept)                            -                9.061  17.1896       -0.52714 0.598096
Habitat_classificationLess Disturbed   -8.118   881.9583    -0.00921 0.992656
Habitat_classificationMore Disturbed 12.423  17.2109        0.72180 0.470416
Basal_area_hectare                            -0.136    0.0408       -3.32993 0.000869

AIC: 86.63837 

As you can see, these values are ridiculous. This is the only run that has come out with values of this sort. For other species who have fm_6 as their best fit model, their values are still not this high. This species is particularly challenging to detect; however, these high SE values are concerning. 

Additionally, I am hoping to receive clear insight on how to report these values. Based on the literature, it seems common place to report the k, AIC (or AICc in my case), delta AIC, psi, SE psi, chi sq, p value, and c-hat. I have the test statistics from the Mackenzie Bailey package; it is the psi and SE psi values that I believe I'm not interpreting correctly. Thanks for your help! 

Jim Baldwin

unread,
Jul 30, 2019, 11:32:28 AM7/30/19
to unma...@googlegroups.com
Here's an opportunity to show my continuing ignorance of wildlife biology:  it seems to me that the variables for occupancy and detection reversed.  Trap_days would seem to be more logically associated with detection rather than occupancy.

Jim

--
You received this message because you are subscribed to the Google Groups "unmarked" group.
To unsubscribe from this group and stop receiving emails from it, send an email to unmarked+u...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/unmarked/07e2dd2c-f5dd-4352-a4df-7c3b7513e060%40googlegroups.com.

Kelly

unread,
Jul 30, 2019, 2:44:31 PM7/30/19
to unmarked
Thanks Jim. That was my thought originally too and I was instructed to place it as such. It never hurts to play a little trial and error to see if moving it to detection would be more effective. 
To unsubscribe from this group and stop receiving emails from it, send an email to unma...@googlegroups.com.

Jim Baldwin

unread,
Jul 30, 2019, 3:50:31 PM7/30/19
to unma...@googlegroups.com
OK.  I'm a statistician so I have to ask such questions.  (And if one's major professor/supervisor insists on certain things, I don't have any constructive suggestions about that.)

So under the assumption that the code is fine one can look at the model results.  For occupancy one can estimate the probability of occupancy for various numbers of trap days, say 0 and 14.  That gives estimates of

0.6122548 = 1 - 1/(1+exp(0.4568))  (for zero trap days)
0.5734155 = 1 - 1/(1+exp(0.4568 - 0.0115*14))  (for 14 trap days)

So there isn't much of a range meaning the estimate of the probability of occupancy is around 0.6 (and I'm assuming that maybe 14 days is a maximum amount of trap days).  And the variable "Trap_days" is not even close to being statistically significant so I'm a little surprised that "Trap_days" made it into the top model.

For detection:  I'm assuming that Habitat_classification:Less Disturbed, Habitat_classification:MoreDisturbed, and Basal_area_hectare are all site variables (i.e., they don't change over visits).  I don't know what the range of basal_area_hectare values are but my guess is that the numbers range from 0 to 1000.  Plugging in that range of values doesn't help me much as here are 6 combinations for the per visit detection probabilities:

Basal_area_hectare
  Less Disturbed 
More Disturbed
0
0.720
1.000
100
0.000
1.000
1000
0.000
0.000

With such wide ranging detection probabilities and not knowing the actual ranges of basal area per ha and seeing that the large difference Less Disturbed vs More Disturbed parameters are not statistically significant, I wouldn't be able to say much about the detection results.  I know I sound like a broken record but probably looking at the actual data is also necessary to check on the results.  (Determining the correlation matrix of the parameter estimates might be suggestive that something is wrong.)

Jim


To unsubscribe from this group and stop receiving emails from it, send an email to unmarked+u...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/unmarked/82925e1e-c5e5-4709-96ef-0e1f5e897777%40googlegroups.com.
Message has been deleted

K.

unread,
Jul 30, 2019, 4:33:05 PM7/30/19
to unmarked
Perhaps I should have provided more information in my initial post, apologies. Your questions are welcomed. 

This is a 120 sample period, verified in 7-day increments. There are 45 site locations. The variables (habitat_classification, basal_area_hectare (basal area per hectare), elevation, trees_per_hectare) are all site-specific covariates. Those are unique to each site but static. 

Regarding the stats themselves--due to small sample size of cryptic species, they aren't going to be of statistically significant value for this study. Just means that this study needs to be replicated for a longer period of time! :)

 Mostly my concern is that the values for this specific species are so outrageous I think there is user error on my end regarding the coding that may need to be corrected. Here is the basic lines of coding thus far:

"##read in the data
sevenday <- read_csv("sevenday.csv")
#View(sevenday)
#example from Fiske and Chandler 2011 "unmarked: An R package for Fitting Hierarchical Models of Wildlife Occurrence and Abundance"
#M=45 J=(120/7)=18 visits over 7 day increments 

##had to remove cameras with no site covs or test stats won't run correctly

y <- sevenday[,26:43]#rows of Perhaps I should have provided more information in my initial post, apologies. Your questions are welcomed. 

This is a 120 sample period, verified in 7-day increments. There are 45 site locations. The variables (habitat_classification, basal_area_hectare (basal area per hectare), elevation, trees_per_hectare) are all site-specific covariates. Those are unique to each site but static. 

Regarding the stats themselves--due to small sample size of cryptic species, they aren't going to be of statistically significant value for this study. Just means that this study needs to be replicated for a longer period of time! :)

 Mostly my concern is that the values for this specific species are so outrageous I think there is user error on my end regarding the coding that may need to be corrected. Here is the basic lines of coding thus far:

"##read in the data
sevenday <- read_csv("sevenday.csv")
#View(sevenday)
#example from Fiske and Chandler 2011 "unmarked: An R package for Fitting Hierarchical Models of Wildlife Occurrence and Abundance"
#M=45 J=(120/7)=18 visits over 7 day increments 

##had to remove cameras with no site covs or test stats won't run correctly

y <- sevenday[,26:43 ]#rows of spp detection "dates;" (7 day camera trap visit)
y

##set up Site and Observed Covariates
siteCovs <- as.data.frame(sevenday[,2:7])
obsCovs= list(cov2=sevenday[,26:43])
siteCovs
obsCovs

##get output on data
umf <- unmarkedFrameOccu(y = y, siteCovs = siteCovs, obsCovs = obsCovs)
summary(umf)


fmnull <- occu(~ 1 ~ Trap_days, umf) 
fmnull

fm1 <- occu(~ Habitat_classification ~ Trap_days, umf)
fm1

fm2 <- occu(~ Elevation ~ Trap_days, umf) 
fm2


fm3 <- occu(~ Basal_area_hectare ~ Trap_days, umf) 
fm3

fm4 <- occu(~ Trees_per_hectare ~ Trap_days, umf) 
fm4

fm5 <- occu(~ Habitat_classification + Elevation ~ Trap_days, umf) 
fm5

fm6 <- occu(~ Habitat_classification + Basal_area_hectare ~ Trap_days, umf) 
fm6

... continue to fm15...

Cand.mods <- list("fm1" = fm1, "fm2" = fm2, "fm3" = fm3, "fm4" = fm4, "fm5" = fm5, "fm6" = fm6, "fm7" = fm7, "fm8" = fm8, "fm9" = fm9, "fm10" = fm10, "fm11" = fm11, "fm12" = fm12, "fm13" = fm13, "fm14" = fm14, "fm15" = fm15, "fmnull" = fmnull)

aictab(cand.set = Cand.mods, second.ord = TRUE)


##compute observed chi-square
obs <- mb.chisq(fm6)
obs
##round to 4 digits after decimal point
print(obs, digits.vals = 4)


##compute observed chi-square, assess significance, and estimate c-hat
obs.boot <- mb.gof.test(fm6, nsim = 3)
##note that more bootstrap samples are recommended
##(e.g., 1000, 5000, or 10 000)
obs.boot
print(obs.boot, digits.vals = 4, digits.chisq = 4)"

Based on instructor recommendation of trap_days being the null, and reallocating it to a detection covariate with no success, that doesn't appear to be the issue but was worth the try. I'm certainly not well rehearsed on R, and this is my first really attempt to use unmarked as a tool, so I presume that the coding is wrong on my end. 

I hope this helps!  
Reply all
Reply to author
Forward
0 new messages