Ecological Circuitry Collaboratory

MODELING WORKSHOP HELD WITH TONY STARFIELD, 26-30 AUGUST 2002

CONNECTING IDEAS AND DATA IN MODELS WORKSHOP WITH IES STAFF, JANUARY 23-25, 2003

MODELING WORKSHOP HELD WITH TONY STARFIELD, 26-30 AUGUST 2002

Tony Starfield gave an introduction to modeling and provided a methodology of modeling using simple models and rapid prototyping. For many of us, this served as a bridge between conceptual and quantitative models, illustrating that elegant models need not be (and perhaps cannot be!) complicated. What follows is a summary of the main points of this workshop and his heuristics for modeling:

• Models are built for particular purposes, and the utility of a model has entirely to do with how well it works for the purpose for which it was built; they do not have to be right nor do they have to give you a perfect answer.
• To build a model, you have to simplify the real world into a model world. Both quantitative and qualitative knowledge about a system get incorporated into the model world. Keep in mind that modeling "answers" are about the model world.
• Simple models can provide a bridge between the world of data and the world of models. Such models are elegant because they are simple conceptually though they may involve large numbers of computations.
• Computers can be a laboratory for clarification of research questions and needs in data collection as well as for exploration of problems where sufficient data collection is impossible. Spreadsheets can be used for the construction of simple models (both spatial and not). When states of a system and changes in those states need to be explored, frame-based models can be quite helpful. These models both track changes within a system and contain rules for when changes within one state (frame) of the system are sufficient to cause a change to a different state (frame). Qualitative information about systems can be incorporated into computer-based models with quantitative elements by making qualitative information into semi-quantitative state or integer variables.
• Ecological models really cannot be "validated" (how could enough data ever be collected to know the true relationships among elements in the real world?), but model verification (which for Tony includes reality checks on model structure, assumptions, proper functioning on the computer, parameter choice, and sensitivity analyses) is essential.
• Collaborative projects can be a good way to learn new things, but collaborative models can be challenging because of the different perspectives among participants. Tony emphasized (1) the importance of suites of simple models with clear objectives, (2) not having models or model components that are inscrutable, (3) working hard at communication, (4) balancing what you know and what you don't, and (5) being humble-everyone is working outside her or his own area of expertise.

Heuristics for modeling (in general)

1. keep it simple, but not too simple (too simple=trivial)
2. use prototypes
3. if in doubt, leave it out
4. make assumptions
5. but keep a list of them
6. look for upper and lower bounds
7. be sure of your spatial and temporal scales (if you are having a problem, try changing your spatial or temporal scale)
8. think yourself into the problem
9. anticipate your output (what and how graph?)
10. plan for a sensitivity analysis
11. find a suitable notation
12. cut through Gordian knots (don't get caught in detail)
13. consider salami tactics (don't try to do it all at once...take smaller steps to a goal)
14. plan to explain your model (a deterrent to making it too complicated)
15. keep going! persevere
16. don't be discouraged by poor data-think your way around it

top

CONNECTING IDEAS AND DATA IN MODELS
WORKSHOP WITH IES STAFF, JANUARY 23-25, 2003

This workshop focused on philosophy of science and modeling, the structure of specific models, progress in individual student projects, and discussion of our group project. Matt Van de Bogert, Maria Uriarte, and Mike Pace, staff at the Institute of Ecosystem Studies, offered examples of models from their work with special emphasis on the motivation behind the models and some of the mechanics of connecting ideas and data in different kinds of models. Some of the main points of the discussions were:

• All science is based on models, some quantitative, others conceptual. We are more likely to recognize assumptions and data needs if we can translate our conceptual models into either a pictorial (e.g., box-and-arrow) or quantitative model. Students were encouraged to build their conceptual models more completely and then to build the math and programming piece-by-piece. Programs such as Matlab can facilitate programming and matrix manipulation.
• Scientific investigation of the world includes both designing experiments to test hypotheses about the way the world works and by trying to make valid, useful inferences about reality from available data. While the former approach is familiar to all, the latter approach is powerfully exemplified in likelihood modeling in which multiple models can be compared against each other with respect to how well they explain the data.
• In a likelihood approach, probability distributions are used first to find parameters for a given model and then to compare alternative formulations of the model with respect to how well they fit the data given a penalty for extra parameters. Additionally, the likelihood approach allows confidence bounds to be calculated for each parameter. In their most useful formulation, a nested series of models of increasing complexity are built so that the simplest models can be derived from the more complicated ones by setting terms equal to zero. Models are sequentially tested as to how well they fit the data and then are compared against each other using a likelihood ratio test. In these models, it is best that parameters have some biological meaning so they can be interpreted with respect to system behavior. More detail about likelihood models is given in Hilborn and Mangel's The Ecological Detective (1997 ).

top