If you do not find here what you were looking for, check (again) the E-SURGE user’s manual. Alternatively, you might want to have a look to the ‘Gentle introduction to MARK’ written by Evan Cooch and Gary White. Lastly, do not hesitate to post your question on the E-SURGE forum, we’ll do our best to help you (make sure you have checked the Archives before asking your question, the answer to your problem might well have been already given).

1. I’m not sure what’s in the states, I need a refresher.

>> No problem! The number of states always includes the state dead, and you must account for that when giving the number of states using “Modify”. This state always appears in last position in the transition matrices. If you just have single state live recaptures (CJS model sensu lato) you will then have two states, state 1 being “alive”, and state 2 “dead”. If you have 3 sites, then you will have 3 + 1 = 4 states. Note that in GEPAT, the state "dead" is always removed from the initial state vector as individuals are all alive when marked.

2. I’m not sure what’s in the observations or events either, I need a refresher.

>> You must remember that the events correspond strictly to the codes in the capture histories (one code = one event) and that they are numbered starting from event 0 (0 traditionally corresponds to "not-observed"). You must account for that when giving the number of events using “Modify”, e.g. if you have 2 events: “observed breeding” and “observed not-breeding”, coded as 1 and 2 in the encounter history, you will have 3 events in E-SURGE. There cannot be less than 2 events (simple capture-recapture). Lastly, note that in E-SURGE the event at the time of the first encounter (mark) is considered as different from the following ones (recaptures): this is specified as 'firste + nexte' in GEMACO.

3. I can’t make sense of the row and columns in the transition matrices.

>> For transitions, the rows are the departure states (from) and the columns the arrival states (to). As a result all transition matrices are row stochastic (i.e. have row sums equal to 1).

4. How can I handle recoveries?

>> For handling dead recoveries or mixture of live and dead recoveries, the state “dead” is insufficient, as an individual stays in this state whereas recovery can only happen in the year of death. You have to use one state “newly dead”. Procedures for ring-recovery models and a mixture of recaptures and recoveries are described in Lebreton et al. (1999) and Choquet (2008). For several sites, the mixture model is described in Duriez et al. (2009).

5. How can I handle heterogeneity in the detection process?

>> You can have > 2 states (i.e. at least 2 states + the dead state) and just 2 events (not captured, captured) when you are running a survival model with heterogeneity in detection (as a two-level mixture model). The states can then be without exchange (diagonal survival-transition matrix) and are then alike hidden groups (Cubaynes et al. 2010). They can also have some transitions, i.e. heterogeneity changing over time (Pradel 2009).

6. It’s been a while since I’ve used E-SURGE, and I’m not sure how to code in GEPAT. Can you remind me the basics?

>> To specify a parameter that will be estimated (i.e. that will be assigned an effect in GEMACO), you can use any letter (that may appear as a Greek character). In a Pattern matrix, "-" means the parameters corresponding to this cell will always be set to 0; "*" means 1 - the sum of all the other parameters on the same row. In all matrices, there must be one and only one “*” per row, as the parameters in a row always sum to 1. In a row, if you want to fix a parameter to 1 and all the others to 0, use "*" for that focal parameter and "-" for all the other ones. If you want to relate mortality m to a covariate (such as an estimate of relative harvest h, as m=b x h, according to the theory of exploited populations), use a letter for the transition to the dead state and “*” for the corresponding survival probability in the same row. With the identity link and no intercept you get your relationship in GEMACO with a syntax such as “time*x” (Choquet 2008). When you have several steps, after entering the size of the matrix, use the default matrix options (diagonal, full or empty matrix) as a starting point to write your own matrix. Keep in mind that the number of rows at a step must be equal to the number of columns in the previous step. Also, the number of rows of the matrix at the last step must be equal to the number of states.

7. I get confused when handling age classes. Can you help me?

>> First things first. If you have K occasions, then you can specify any number of age classes A < K. In transitions, the age classes you will refer to are for intervals and will be 1, 2, ..., A-1, A = "A+" i.e. {A, A+1 ,..., K-1}.In E-SURGE, age classes always start from 1. If individuals are marked at birth (age 0), the first age class (age 1) in the output file actually corresponds to age 0. For instance, the survival estimates obtained for age class 1 in the output file is the survival probability from birth to age 1, the estimate for age class 2 corresponds to the survival probability from age 1 to 2, etc. For transition between states (e.g. recruitment), it all works like individuals were making the transition at the beginning of an occasion. For instance, let us assume that individuals are marked at birth and none of them recruit before 3 years old. In GEMACO, for the recruitment part, you should code a(1:2,3,...,K-1) and fix the probability of a(1:2) to zero in IVFV. In the output file, the recruitment probability estimates for age class 3 is the probability that a juvenile will be breeder at 3 years old.For events, the age classes you will refer to are for occasions and will be 0,1, 2 ,.., A, but will be denoted as 1, 2,…, A+1. This is sometimes problematic when you want to include age effects in recapture probabilities because firste is age(1), and the typical age(1) (used in M-SURGE or MARK) is now age(2). If you collapsed age in the « modify » option for example to 3 age classes and you want a model in which recapture varies between first year and older individuals you have to write: firste+nexte.[a(2,3:4)].

8. When I launch E-SURGE, I only get the black DOS window and nothing appears.

>> When it is the first time you use E-SURGE, you’ll have to be patient as it can take some time before the main window pops up (less than one minute though). If it closes suddenly, It may also be related to you not having the appropriate administrator rights or because you need to run it as an administrator for the first time (right click option on the E-SURGE icon). Also, make sure that you save all your files (data, session, …) in a directory for which you have the rights to read and write.

9. I run lots of models and I would like to resume my session and run one of these models. How do I do that?

>> In E-SURGE, just choose to open an existing session. You will be asked to re-enter the data file if the session directory has changed. Once your session is opened, make sure that the data characteristics are correct (number of states, events, groups, age classes). The list of models you have run before should be displayed in the box in the bottom-right. Click on the target model, then, click on the ‘retrieve’ button. At this point, if the ‘retrieve’ button is not active then either click on GEPAT or on the model status and exit. E-SURGE will use the excel file to retrieve the model (GEPAT, GEMACO, IVFV). At this point, if you want to re-run your model, just click on “RUN”.

10. Are the characteristics of my dataset (number of capture occasions, age classes, …) remembered when I re-enter my session?

>> No, you have to set adequate values using the “Modify” button every time you reenter your session.

11. What are the advantages of the Headed format?

>> From an Excel sheet, you can name each column. So, from your Excel data, simply enter “H:” in headed cell for each occasion and “S:” for the sample size of each capture history (= 1 if each row corresponds to the encounter history of a single individual). If you have quantitative or qualitative variables, use “COV:” or “$COV:” respectively. You can enter a label for that covariates after the “:” that you will be able use when specifying the effects of your model. You can use characters for the levels of qualitative covariates (‘male’ and ‘female’ for a variable labeled ‘sex’ for example). E-SURGE will automatically create the associated shortcut with as many levels as entered in the data file. In GEMACO, you will be able to use “sex” in the sentence to have different estimates for each level of your covariate. Using this format, you no longer have to use binary columns for each level of the covariate.

12. How do I check whether my parameters are estimable?

>> Use the diagnostic tool incorporated in E-SURGE that will tell you if any parameter cannot be estimated: see the E-SURGE 1.8 user’s manual for more details – sections 6.7 and 7.2 and Choquet and Cole (2012).

13. Are there some tricks to be more computationally efficient?

>> Based on our experience, the ‘EM (20) + Quasi-Newton’ procedure can be more effective than the Quasi-Newton algorithm, especially regarding local minima. However, the EM(20) will not be available when using Unequal Time Interval (UTI). It also sometimes helps to speed up the convergence to use the estimates of a previous simple model as initial values for a more complex model. To do so, use the “From Last Model” option in “Advanced? numerical/Initial Values”. If you are using UTI, make sure that the option is checked in the Setting menu and that you are applying UTI for the correct steps. If you change the GEMACO phrases and you are using the “From Last Model” option, the IVFV will propose you Initial Values on the scale of real number. Click on the “R” button to show Initial Values on the biological scale, i.e. on [0; 1]. All parameters fixed to 0 and 1 in the previous model will have value 0.25 and 0.75, respectively. Make sure to change the value on the biological scale and to check the “fixed value” case.

14. How to avoid local minima?

There is no definitive answer to this question. We highly recommend using the “Multiple Random Initial Value” and the ‘EM (20) + Quasi-Newton’ options to run our model starting from different sets of initial values. E-SURGE will pick the minimum deviance with the associated estimates.

15. I’d like to enter initial values myself. How do I do?

>> If you have only a few parameters for which you want to assign your own initial values, then you can do it directly in the IVFV window just before running your model. Otherwise, go to the IVFV step (enter your sentence in GEMACO, etc.), then open the IVFV menu. Click on FILE and save the IVFV file. You can now open this file containing the initial values assigned by default by E-SURGE in a text editor (TextPad is useful). The 3 numbers on the first line correspond to the number of parameters for Initial State, Transition and Event, respectively. A line indicates the kind of parameters you’re dealing with. The first column indicates whether the parameter is fixed on the real scale (value 1), on the biological scale (value 2) or not (value 0). When you have modified the initial values, save your file again. In the IVFV window, go back to the File menu, load you IVFV file, check if everything is fine and exit. You can now run your model.

16. Do you have examples of shortcuts in GEMACO that I may use for my own purpose?

>> Péron et al. (2010) provide R code to generate shortcuts for age effects when individuals enter the study at various ages. Choquet (2008) provides examples as well as some theoretical considerations.

17. I’m stuck with covariates. Can you help me please?

Most often, the error comes from the fact that the number of values in the covariate text file does not correspond to the number of levels of the effects. For example, let us assume we want to specify an effect on survival with the sentence t*x, then you need k-1 values for the covariate where k is the number of occasion. Note also that when including environmental covariates (e.g. temporal covariates), if one value of a covariate is repeated for example for 2 years, only the estimate of one year will appear in the « reduced set of parameters » in the excel file. Eventually, if you have to use individual covariates and you’d like to get confidence intervals for the biological parameters to which you applied these covariates, you’ll have to do some calculations on your own that are detailed in the document ‘Slides: fixed effects with individual covariates’ provided in E-SURGE.

18. Self-teaching is my thing. Are you aware of papers using E-SURGE that have provided details for implementing their models?

A list of applications of the multievent framework by topics is given here. Below we refer to papers and appendices providing details on the implementation of their models.

·       Random effects: Gimenez and Choquet (2010); Choquet and Gimenez (2012); Choquet et al. (2013).

·       Mixture models (heterogeneity in detection / survival probabilities): Cubaynes et al. 2010; Péron et al. 2010.

·       Experience: Desprez et al. 2011; Pradel et al. 2012.

·       Trap-dependence: Crespin et al. (2008); Pradel and Sanz-Aguilar 2012.

·       Multistate model: Cubaynes et al. 2011; Sanz-Aguilar et al. 2012.

·       Models with uncertainty on state assignment: Gimenez et al. 2012

·       Mixture of recaptures and recoveries with tag loss: Juillet et al. 2011

·       Live and dead encounters, mortality causes and tag loss: Tavecchia et al. 2012.

·       Telemetry and mortality causes: Devillard and Bray 2009

·       Sex uncertainty: Genovart et al. 2012.

·       Exploiting indirect information and reproductive skipping behavior: Sanz-Aguilar et al. 2011.

·       Transience and reproductive strategy: Genovart et al 2013.

REFERENCES

Choquet R. (2008). Automatic generation of multistate capture recapture models. The Canadian Journal of Statistics 36: 43-57.

Choquet, R., et al. (2009). Program E–SURGE: a software application for fitting multievent models. Modeling Demographic Processes in Marked Populations. D. L. Thomson, E. G. Cooch and M. J. Conroy. Berlin, Germany, Springer. 3: 845-865.

Choquet, R. and Gimenez O. (2012). Towards built-in capture-recapture mixed models in program E-SURGE. Journal of Ornithology. 152 (suppl 2): 625-639

Choquet R. and Cole D. (2012) A Hybrid Symbolic-Numerical Method for Determining Model Structure. Mathematical Biosciences 236: 117-125

Choquet, R., Sanz-Aguilar, A. Doligez, B., Nogué, E., Pradel, R., Gustafsson, L., Gimenez, O. (2013) Estimating Demographic Parameters in the Wild with Dependence among Individuals within Cluster. Methods in Ecology and Evolution. 4: 474-482

Crespin L., Choquet R., Lima M.A., Merritt J.F. and Pradel R. (2008). Is heterogeneity of catchability in capture-recapture studies a mere sampling artifact or a biologically relevant feature of the population? Population Ecology, 50: 247-256.

Cubaynes, S. Pradel, R. Choquet, R. Duchamp, C. Gaillard, J.-M., Lebreton, J.-D., Marboutin, E., Miquel, C., Reboulet, A.-M., Poillot, C., Taberlet, P. and O. Gimenez. (2010). Importance of accounting for detection heterogeneity when estimating abundance: the case of French wolves. Conservation Biology 24:621-626.

Cubaynes, S., P. F. Doherty Jr, E. A. Schreiber, R. W. Schreiber and O. Gimenez (2011). To breed or not to breed: seabirds response to extreme climatic events. Biology Letters 7: 303-306.

Desprez, M., Pradel, R., Cam, E., Monnat, J. Y., & Gimenez, O. (2011). Now You See Him, Now You Don't: Experience, Not Age, Is Related to Reproduction in Kittiwakes. Proceedings of the Royal Society B-Biological Sciences, 278, 3060-3066.

Devillard, S. and Y. Bray (2009). Assessing the effect on survival of natal dispersal using multistate capture-recapture models. Ecology 90(10): 2902-2912.

Duriez O., S.A. Saether, B.J. Ens, R. Choquet, R. Pradel, R.H.D. Lambeck, M. Klaassen. (2009). Estimating survival and movements using both live and dead recoveries: a case study of Oystercatchers confronted with habitat change. Journal of Applied Ecology, 46(1), 144-153.

Gauthier, G. & Lebreton, J.D. (2008). Analysis of band-recovery data in a multistate capture-recapture framework.The Canadian Journal of Statistics36 : 59–73

Genovart M, Pradel R, Oro D (2012) Exploiting uncertain ecological fieldwork data with multi-event capture-recapture modelling: an example with bird sex assignment. Journal of Animal Ecology 81:970-977.

Genovart M, Sanz-Aguilar A, Fernandez-Chacon A, Igual JM, Pradel R, Forero MG, Oro D (in press) Contrasting effects of climatic variability on the demography of a trans-equatorial migratory seabird. Journal of Animal Ecology. 10.1111/j.1365-2656.2012.02015.x

Gimenez, O.and R. Choquet (2010). Incorporating individual heterogeneity in studies on marked animals using numerical integration: capture-recapture mixed models Ecology. 91: 951-957.

Gimenez, O., Lebreton, J.-D., Gaillard, J.-M., Choquet, R. and R. Pradel (2012). Estimating demographic parameters using hidden process dynamic models. Theoretical Population Biology. 82: 307-316

Juillet C, Choquet R, Gauthier G, Pradel R (2011) A capture-recapture model with double-marking, live and dead encounters, and heterogeneity of reporting due to auxiliary mark loss. Journal of Agricultural, Biological and Environmental Statistics 16:88-104.

Lebreton, J.-D., et al. (1999). Competing events, mixture of information and multistrata recapture models. Bird Study 46(suppl.): 39-46.

Péron, G., Crochet, P.A.C., Choquet, R., Pradel, R., Lebreton, J.-D. and O. Gimenez. (2010). Capture-recapture models with heterogeneity to study survival senescence in the wild. Oïkos 119: 524-532.

Pradel R, Sanz-Aguilar A (2012) Modeling Trap-Awareness and Related Phenomena in Capture-Recapture Studies. PLOS One 7:e32666.

Pradel R, Choquet R, Béchet A (in press) Breeding experience might be a major determinant of breeding probability in long-lived species: the case of the greater flamingo. PLOS One.  

Sanz-Aguilar A, Tavecchia G, Genovart M, Igual JM, Oro D, Rouan L, Pradel R (2011) Studying the reproductive skipping behavior in long-lived birds by adding nest-inspection to individual-based data. Ecological Applications 21:555-564.

Sanz-Aguilar A, Béchet A, Germain C, Johnson A & Pradel R. (2012). To leave or not to leave: survival tradeoffs between different migratory strategies in the Greater Flamingo. Journal of Animal Ecology 81: 1171-1182.

Tavecchia G, Adrover J, Muñoz Navarro A, Pradel R (2012) Modelling mortality causes in longitudinal data in the presence of tag loss: application to raptor poisoning and electrocution. Journal of Applied Ecology 49:297-305.