Troubleshooting the model

Modeling is an experimental approach in itself. You should be encouraged to try different things when problems arise. Make a small list of the problems you see in the simulation results. Ask yourself which model parameters you should adjust – and in what way you should adjust them – to improve your simulation. Which of these changes might have adverse consequences to the model's fit? Then adjust a parameter and see what happens. Often you will be surprised by the results. Ask yourself why things worked out as they did. Only after experimentation will you begin to learn the ropes. Modeling is open­ended. There is no one way to do it.

Look for balance in how your simulation matches the data you have for comparison. Do not try to mimic the history of the population to the nth degree. Though it may be possible to get the simulation ever closer to the observed data, remember that all data contains bias, regardless of quality. Trying to pin things down exactly may be counterproductive. Balance is more important.

If you cannot get a good fit between simulated and observed values, you should review the basic assumptions of the model. A common problem with models like POP-II is the assumption of negligible net migration to or from the population or herd unit. If you cannot meet this assumption, then you may need to reorganize your data using different herd unit boundaries. In any event, POP-II only contains selected components from the ecological setting and may not be appropriate for every situation.

If you can get a good fit between simulated and observed values, you should be relentless in making sure how good you think the simulation is. It is prudent to look at the "Simulation Details Report" to see if the details reported agree with your expectations. In particular, look at the age distribution of the harvest. These should agree with your data or expectations. If they do not, try further adjustments. If they do agree, proceed to the next step. Don't be a lazy modeler!

Next, review the simulations from several different populations. Are there any major differences in the parameters? If so, why? Maybe there is a good reason, maybe not. Remember that any previous models may be in error. Does this simulation make you want to re-evaluate prior models you have built?

Continue  

  Previous (1 of 3)