Home » Uncategorized » 3. If no budget and/or personnel concerns, how would you like to address these challenges or issues as mentioned above?
Home » Uncategorized » 3. If no budget and/or personnel concerns, how would you like to address these challenges or issues as mentioned above?
A number of experimental design methods already exist in the modeling literature. The question is which of them fit into ABM experimentation. Or perhaps we need a new approach?
As of now, every report on ABM has its own structure or lack thereof. Sometimes we know the distributions of inputs, sometimes we don’t. I read ‘we used average value for input’ – average from what? why average? why not sample the whole distribution? How did you generate your sample? Why this number of model runs? etc.
There are a lot of ad hoc assumptions. I’d like to have some framework so that, when I read a paper, I can focus on the problem, rather then figuring out the design of simulation used in this particular project.
If I had the money and personnel, I’d start from reviewing what’s out there, summarizing candidate approaches, implement them, and test on a number of archetypal problems that use ABM to identify the best-practice(s). We need a standard protocol for ABM experimentation that has been rigorously tested. Just like there are protocols of experimentation in the physical lab, there should be a protocol for experimentation in ABM. More on the topic when we meet …
Related to Arika’s comment, comparability of findings across different ABM efforts remains a challenge. There are so many case-based ABM studies that are useful for their specific application, but which do not connect to broader theory to facilitate building on prior knowledge and advancing theory. This could be related to a standardized experimental design for ABMs, but it also relates to a larger conversation about what are the ‘grand objectives’ for ABM use. With its bottom-up, process-based perspective, ABM is a unique approach and I think we can ask more of it than from some other modeling approaches.
I think comparison studies would be useful. Answering the same question(s) from different angles by using different ABM approaches from the different domains. This way we can learn about the opportunities that are out there there. I would then host a multi-disciplinary website (e.g. wiki) where people could upload their models and in this way contribute to the comparison. Other modelling approaches could be added in the future to unearth how all these approaches complement each other.