Looking for recommendations for analysis of small mammal data

129 views
Skip to first unread message

Valerie Steen

unread,
Nov 21, 2024, 10:36:46 AM11/21/24
to unmarked
Hello all,

I was given a small mammal dataset to analyze and I've never analyzed this type of data. I'm wondering what an appropriate or good model would be. 

Sampling details: 93 sites were sampled over 9 years (2014-2022). 8 - 12 plots were sampled per year and a given plot was sampled just once in the 9 years. At each site, 3 transects were sampled (in 3 directions, e.g., at 60, 180, and 270 degrees, from a plot center) and each transect had 10 - 20 trapping stations (varies from year to year) with a snap trap, a sherman trap, and a pitfall trap. Traps were checked 3 consecutive nights and they recorded the species trapped, or whether the trap was sprung, bait taken, open, or etc. 

Constraints/questions: I need to do this analysis quickly (over about two days; thus posting here versus hierarchical modeling) and I just view this as exploratory. There aren't any particular questions with this data; I'd like to offer something about occupancy or abundance, site to site variability, and year to year variability. There's some year to year noise given different yearly effort/skill in species identification (e.g., shrew spp. in some years, versus identified to species level in others). Any suggestions on what to look at here would be welcome as well. I have one habitat covariate and a handful of detection covariates (date, weather, etc.). There are also some morphometrics for the small mammals such as weight, body length, etc. 

Thanks in advance,
Valerie

Jeffrey Royle

unread,
Nov 21, 2024, 5:54:02 PM11/21/24
to unma...@googlegroups.com
hi Valerie,
 I'm afraid that getting something done in a couple days will probably be a stretch since there seems to be a lot of aspects to your data. That said, if the individual animals were marked then this would be an ideal situation for a spatial capture-recapture model.   It's not clear if they were marked though.
 Was the sampling for each grid repeated on consecutive days?  Or were the traps set, checked, and then pulled?   If the latter, then this is probably best viewed as a logistic regression (or possibly Poisson regression depending on if detections are aggregated to sites).  Without replication you won't be able to fit occupancy or related models.

 Its curious that a new set of sites were surveyed in each year so that there are no replicate samples of any given site.  That's not a typical design and seems limiting from the standpoint of understanding animal population structure or dynamics.

 Sorry I can't give more specific advice at this point.
regards
andy

 
 

--
*** Three hierarchical modeling email lists ***
(1) unmarked (this list): for questions specific to the R package unmarked
(2) SCR: for design and Bayesian or non-bayesian analysis of spatial capture-recapture
(3) HMecology: for everything else, especially material covered in the books by Royle & Dorazio (2008), Kéry & Schaub (2012), Kéry & Royle (2016, 2021) and Schaub & Kéry (2022)
---
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 visit https://groups.google.com/d/msgid/unmarked/dc73cd36-2ee0-4cf4-92c2-39b50479f90bn%40googlegroups.com.

Jeffrey Royle

unread,
Nov 21, 2024, 5:57:59 PM11/21/24
to unma...@googlegroups.com
hi Valerie,
 ok I missed the detail that the traps were run for 3 consecutive nights.... so then: were individuals marked?   If so then capture-recapture is the way to go.  If not then you could fit occupancy models perhaps using "traps" as sites.  (how far apart were the traps?). 
 There would be some spatial dependence to consider for neighboring traps perhaps.  At any rate a "basic occupancy" model could be made to run in relatively short order
 (sorry for misreading your email and my errant initial response)
regards
andy


On Thu, Nov 21, 2024 at 10:36 AM Valerie Steen <valeri...@gmail.com> wrote:
--

Valerie Steen

unread,
Nov 22, 2024, 10:15:20 AM11/22/24
to unma...@googlegroups.com
Hi Andy,

They had a code for 'recapture', but few captures were marked as such. Thus, I don't think they reliably tracked this. Traps were placed 15-m apart. Thank you for the occupancy model idea at the trap station level. I hadn't thought of that, but that might be the way to go! I could probably use the ubms package to allow me to include a random effect of plot. However, I can't wrap my head around how to accommodate the multiple trap types for each station, and to complicate that even more, this differed by year where in 2014-2017 each station had a pitfall, sherman and snap trap and from 2018-2022 snap traps were no longer used. Any thoughts here? 

And, yes, agreed about the deficiency in the sampling design for looking at temporal variability. I plan to give them that feedback. 

Thanks for your reply and help!
Valerie



--
Valerie Steen, PhD
Ecologist
Pronouns: she / her
 

Reed Scott

unread,
Nov 22, 2024, 11:11:35 AM11/22/24
to unma...@googlegroups.com
Hi Valerie and Andy,

For a case where you were using multiple traps at the same site, if you're using an occupancy model, I might suggest using the multi-method AKA multi-scale model. I don't think multi-scale models are currently available in unmarked, but you could consider using PRESENCE or Mark. It sounds like if you surveyed different sites each year, it might make the most sense to run a separate occupancy model for each year? i'm not quite sure how to handle the multi-year aspect of the data as you've described it.

If you decide that a multi-scale model is the way to go, and need help implementing it PRESENCE or Mark, I'd be happy to help.



--
Reed Scott (he/him/his)

Valerie Steen

unread,
Nov 23, 2024, 10:24:34 AM11/23/24
to unma...@googlegroups.com
Hi Reed,

Thanks for that suggestion, and the offer to help! I'll let you know.

Best,
Valerie

Jeffrey Royle

unread,
Nov 23, 2024, 11:32:01 AM11/23/24
to unma...@googlegroups.com
hi Reed,
  the new unmarked function goccu will fit the multi-method type of models I believe.  Ken just added this over the summer, so check it out.  One thing I'm not sure about with Valerie's data is that the methods are not sampling independently. For example, there is probably a negative association between getting caught in a snap trap and getting caught in a live trap when population densities are low.  In this case maybe the multi-state occupancy models would be more suitable (occuMS I think). 
regards
andy


Reply all
Reply to author
Forward
0 new messages