Program PRESENCE - FAQ

FAQ (frequently asked questions) concerning program PRESENCE.

Program PRESENCE - FAQ

Postby egc » Tue Nov 06, 2007 9:33 am

Many of the questions concerning PRESENCE, especially those coming from 'new users', will have been asked before. While you can always use the search facility to find answers (with a little work), to make it more efficient to find answers to common questions, we've posted this FAQ (which we will regularly edit, update and tweak). Answers to some common 'technical' questions are at the bottom.

1. Where can I get PRESENCE? - the primary source for PRESENCE is

[http://www.mbr-pwrc.usgs.gov/software/doc/presence/presence.html

2. What kind of computer do I need to run PRESENCE? - Ideally, a hot-rod PC - there is no such thing as 'too much computer' when using PRESENCE (or any software application, in fact). Minimally, a 2.0 GHz (or faster) PC running Windows 7 or newer, with 1 GB RAM or more (more RAM is always better). Although written for Windows, you can also run PRESENCE on a Mac, or under Linux, using a good Windows-emulator (details to come...)

3. how can I learn to use PRESENCE? - At present, the best way to learn how to use PRESENCE (and occupancy modelling in general) is by attending one of the workshops that are regularly offered (and posted on this forum). In addition, you can/should also consult the following 'online documentation':

http://www.uvm.edu/envnr/vtcfwru/spread ... upancy.htm

4. PRESENCE or MARK? - Both PRESENCE and MARK offer considerable facility in analyzing occupancy data. At present, we'd prefer that your general occupancy questions be posted under the appropriate software section (i.e., if you have an occupancy question, and use PRESENCE, then post your question under the PRESENCE section, and vice versa for occupancy questions for MARK users).

5. I have a question, but suspect it might have been asked before. Rather than ask again, anywhere I can look to see if its already been addressed? - Congratulations and thank you for your consideration - asking a question that has already been answered is a sure way to get 'flamed'. You should approach the decision as to whether or not to post a question in 2 steps:

(1) make sure you have thoroughly read and understood the documentation for PRESENCE that is available first - this is the 'RTFM' request (RTFM = Read The F****** Manual). If you're sure can't find your answer there, then

(2) search the forum

6. OK, I've RTFM, and searched the forum, and still need help. What next? - Easy - post your question here. But, several things to keep in mind. First, resist the initial temptation to email Darryl MacKenzie, or Jim Hines (or anyone else) directly. Even if your thesis/project is due by the end of the week, resist. Post your question on the forum as the first step. And, it is recommended that you post your question (or, questions) in small, discrete pieces - again, resist strongly the temptation to post a several hundred line question containing multiple sub-questions. Once again, even if your thesis/project is due by the end of the week, resist. Many of the people who might want to help will look at a multi-line, multi-element question and figure they have better things to do with their time (and not answer you at all).

7. the literature related to this stuff (i.e., analysis of occupancy data) is pretty daunting - where should I start? - the number of papers in this area isn't large - yet, but some of them are pretty technical.

Fortunately, there is a key, very accessible starting point: run (don't walk) to your nearest bookstore (or, in the modern age, fire up the website for your favorite online bookstore) and order yourself the following:

Occupancy Estimation and Modeling: Inferring Patterns and Dynamics of Species Occurrence by Darryl I. MacKenzie, James D. Nichols, J. Andrew Royle, Kenneth H. Pollock, Larissa L. Bailey, James E. Hines

Frequent Technical Questions

Q: I'm having problems understanding/constructing design matrices in PRESENCE. Is there somewhere I can get some help with this?

A: Yes. First, you need to make sure you understand the basic concepts behind design matrices - a decent book on linear models would help. A good starting point that is fairly general, are sections 6.1 - 6.2 in Chapter 6 of the MARK book

As far as the mechanics of using the design matrix in presence, the following has several examples presented in 'tutorial' form:

http://www.proteus.co.nz/PRESENCE_Examples.zip

Q: I get a warning about model convergence, what does that mean?

A: In order to obtain the maximum likelihood estimates that are reported, PRESENCE uses a numerical algorithm that iteratively tries different parameters values searching for new values that provide a likelihood value that is higher than the previous set of values. Essentially the algorithm gets the new values by adding or subtracting a small amount from the current values. Once the new parameter values no longer increases the value of the likelihood and the parameter values themselves are unchanging (to a certain number of significant digits), then the algorithm is said to have converged. An analogy is if you think of the parameters as latitude and longitude, and the likelihood function as elevation, then the algorithm is trying to find at what point is elevation the highest by starting at some arbitrary starting point, then always moving uphill. Some of the reasons for getting this warning are if the likelihood is very flat in that region of the current parameter values (i.e., the algorithm has found a plateau) or if there are a lot of local maxima (i.e., a jagged mountain range).

Now this warning is outputted if the approximate number of significant digits to which the algorithm does converge is less a specified value (the default is 7). Now there are no guarantees, but in our experience if the reported number of significant digits is:

>3 then the results are probably ok
between 1 and 3, treat them with caution
<1, don’t believe them

Things you can do to try and resolve the issue are:

1. try different initial or starting values (the default is 0 for all parameters) and see if you get the same results. If you do, that’s good, if you don’t then check whether you have the same likelihood value or not. If the likelihood value is identical, but some parameter estimates are different, than could suggest some parameters are confounded (you don’t have the required data to estimate them separately).

2. try a simpler model. You may have too many covariates in your model that you’re trying to estimate, and not enough data.

3. try a different parameterisation of the same model.

Q: I get a warning about the variance-covariance matrix, what does that mean?

A: This warning is usually caused by beta parameter estimates that have a large absolute value (e.g., <-20, or >20), as this translates to real parameter estimates that are very close to the bounds of allowable values; 0 or 1. Firstly, all the standard errors that are reported when you get this warning are meaningless; ignore them. However, if you don’t get the convergence warning as well, then the point estimates and likelihood value may still be valid.

Things you can try are:

1. identify which real parameter is essentially 0 or 1 and use the ‘fix parameter’ option to constrain it to be that value.
2. if using covariates, try a different covariate standardisation.
3. if using a categorical covariate, try using a different category as the ‘standard’.
4. try a simpler model.
5. try a different parameterisation of the same model.

Q: Why doesn’t PRESENCE give me a real parameter estimate for all sites?

A: To avoid overly long outputs, PRESENCE has been developed such that it will only output a real parameter estimate if it is different from the last one outputted. So, if psi estimates are only given for sites 1, 2, 5 and 8, then: 2 is different to 1; 3 and 4 must be the same as 2; 5 must be different from 2 (and 3 and 4); 6 and 7 must be the same as 5; and 8 must be different from 5 (and 6 and 7).
egc
Site Admin
 
Posts: 168
Joined: Thu May 15, 2003 3:25 pm

Return to FAQ

Who is online

Users browsing this forum: No registered users and 1 guest

cron