using the 'predict' function in unmarked

[Tim Henderson]

Hi, I'm having trouble using the predict function in unmarked.

I've run all my models, and this is the best model (using AIC)

 fm14<-occu(~ habitat ~ survey+habitat, data=STQ_um)

Now I want to find out the detectability and the occupancy estimate. So I've done this:

#detectability

newData  <-  data.frame(habitat = factor(c("DryRF","DrySchl","GrassyWL","Open","WetSchl"), levels=c("DryRF","DrySchl","GrassyWL","Open","WetSchl")))

pred <- predict(fm14, type="det", newdata=newData, appendData=TRUE)

pred

and I get my output values

> pred

   Predicted         SE      lower      upper  habitat

1 0.08584068 0.01966684 0.05433488 0.13304480    DryRF

2 0.11718511 0.01108717 0.09714015 0.14072169  DrySchl

3 0.12577125 0.01312243 0.10222061 0.15381811 GrassyWL

4 0.10369650 0.01648434 0.07554840 0.14073555     Open

5 0.03873907 0.01038252 0.02280158 0.06507448  WetSchl

Great! (I think).

Now I want to do the same, but with occupancy estimate.

I'm am unsure on how to calculate this when I have two covariates on occupancy (survey+habitat). If it was just one covariate (habitat for example), I would just change "det" to "state" and it would work. But if I do it with survey+habitat as covariates, it says 'object survey not found'. 

Does anyone know what I do here to get occupancy estimate??

Thanks!

[Tom Bruce]

Hey,

Nice to see someone else working in Australia.  My first word of caution would your detection probabilities look quite low especially in the wet sclerophyll and dry rainforest, this could potentially make any reliable modeling challenging.  Have you run a goodness of fit test on the global model to check for overdispersion etc? 

Now for the occupancy estimate, it should be easy enough 

Here is some example code where I'm comparing how habitat and understory vegetation influence occupancy.

test<-occu(~Effort ~Habitat+Low_veg,species)# Example model

preds_syn<-data.frame(Habitat=Habitat, Low_veg=Low_veg) #Predictors in data frame

Hab_LV_Psi<-predict(test,type="state",newdata=preds_syn,appendData=T) #create dataframe with predictons and variables tied together.

print(Hab_LV_Psi)# print the list. 

   Predicted         SE     lower     upper Habitat     Low_veg

1  0.5804043 0.08381513 0.4133561 0.7308570     Hab_1 -0.32180924

2  0.5844523 0.08142305 0.4216476 0.7306983     Hab_1 -0.25678236

[Tim Henderson]

Thanks Tom, 

I have edited as per your example, however I get this error:  

Error in data.frame(habitat = habitat, survey = survey) :   arguments imply differing number of rows: 5, 3.

So what I did was to merge the two: 

newData1<-data.frame(habitat=habitat)

newData2<-data.frame(survey=survey)

total <- merge(newData2,newData1)

pred <- predict(fm14, type="state", newdata=total, appendData=TRUE)

pred

and output is:

Does this seem correct to you??

Cheers!

[Tom Bruce]

Yeah as far as I can tell that should be what you're expecting as you're basically predicting the state against two categorical variables right? That means you'll only have one value per combination of survey and habitat type.

One thing to bear in mind could be that if your surveys are at different times you might want to look into a different occupancy model type as the single-season assumes that the population is closed. However, if you're expecting to see a change in your species population between surveys then maybe a multi-season occupancy model might be more appropriate. Anyway, just a thought, keep on plugging on looks like you're making progress.

[Divya Vasudev]

Dear Tim, 

Since what you need is all combinations of your two categorical covariates, one way to get this would be to use the function expand.grid. 

new.data = expand.grid(survey = levels(STQ_um$survey), 

                                       habitat = levels(STQ_um$habitat))

That should give you a data.frame with all combinations of survey and habitat that are present in your data, which can then be fed into the predict function. 

best wishes,

Divya

----------------

Divya Vasudev, Ph.D.

Conservation Initiatives

conservationinitiatives.org/divya-vasudev

DST–INSPIRE Faculty

Member – IUCN Connectivity Conservation Specialist Group

Member – IUCN Section on Small Apes

Alumna – University of Florida and WCS–NCBS Master's Program in Wildlife Biology and Conservation, Bangalore, India

--
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/c85f640a-3500-406d-bf57-70b2a6eedb6cn%40googlegroups.com.

[Daniela M-Holland]

Hi there,

Are you by any chance looking at occupancy of spotted-tail quolls?  In case you are, I am working on northern and would be great to have a chat ! 

[Tim Henderson]

Thanks Divya!

Would you happen to know how to do this if one of the covariates were a continuous variable? For example, lure age (which increase by occasion length) + habitat.

Kind regards, 

Tim

To view this discussion on the web visit https://groups.google.com/d/msgid/unmarked/CABcVe0yQY_YcubM1JWib0TqvWrqC6-PS_fGRKscwg_D8M_n65w%40mail.gmail.com.

[Divya Vasudev]

I think this should work: 

new.data = expand.grid(lure = seq(min(STQ_um$lure), max(STQ_um$lure), length.out = 50),

                                       habitat = levels(STQ_um$habitat))

This should give you a range of 50 lure values, from the min to max based on your data, for each habitat type. You can change the number of lure values, or any other parameters as per how you want to predict. 

best wishes,

Divya

To view this discussion on the web visit https://groups.google.com/d/msgid/unmarked/CACUmFw3%3D_79winep%2B5A18znf76eQq2bUZbMpsJxeaZqsaCGQGg%40mail.gmail.com.

[Tim Henderson]

Hi Tom, 

With regards to your second comment, I'm trying to decide whether single-season of multi-season is more appropriate. Dates for my 3 surveys are S01=May/June, S02=Aug/Sept, S03=Nov/Dec, all in the same year. Based on the ecology of my study animal (quoll), their mating season is May-July/Aug, post-mating (i.e female denning) is Sept-Nov, then Juvenile dispersal is Nov-onwards. Therefore, we would expect changes in the population such as male die-off post-mating (e.g. S02), and new juveniles in S03. So based on that, I would think multi-season is the one to go with. However, my question is less related to the effect of survey on occupancy, but rather occupancy between habitat types and treating survey periods as  'replicates/pseudo-replicates' to compare occupancy for each habitat type between survey periods (e.g. do each surveys have a similar pattern on one habitat type having higher occupancy than the others, but occupancy overall decrease from S01 to S03). I'm less familiar with multi-season, though it does seem much more complex, especially given the question I want asked. So at the moment my choices are: do I do a single-season occupancy analysis for 1)  each individual survey, 2) all surveys combined, using survey as a covariate, or 3) multi-season.

If you (or anyone else) has some extra advice on this that would be appreciated!

Thanks,

Tim