14 Realism
There are several reasons why models for policy will tend to be more complex than models developed for scientific research.
In science, the questions are narrow and the study designs are focused enough to form tests of ideas that have clear outcomes.
When models are used to guide policy, the same level of scientific rigor should be applied, but weighing all the evidence requires a broder synthesis.
To build models that can guide malaria policies, the models must be realistic enough to be compelling, and they ought to reflect the knowledge and experience accumulated over years of studying and controlling malaria.
Policy advice should be checked for consistency across studies.
The models must be complex enough to serve many purposes all at once.
To weigh tradeoffs, policy requires broad, synthetic models that allow for comparisons across subject matter domains.
To carry a conversation forward, the models used to guide discussions will need to retain a memory of what has been learned already, so they will tend to add features and grow more complex. Given the uncertainty, policy should be based on model swarms that propagate the uncertainty. The predictions of those models must be specific enough to be proven wrong, so that over time some of the models can be trusted over others. The same models can be used to identify which missing data would have the greatest impact on a policy, and ideally, studies can be conducted to gather this data. Over time, the advice should shift from generic advice to specific advice as more evidence is gathered. This is, in a nutshell, how daptive management works.
It might take a lot of work to build a model that has been fit to all the evidence describing malaria in a management unit over the recent past, and it might cut against the instincts we have as scientists to add all that realism, but it’s worth it to make the effort if it helps communicate with malaria managers.
In designing a software solution to the problem of building realistic models, we designed a framework for building models and a toolbox to build model swarms that would address the concerns of malaria programs. In the chapters that follow, we’ll show the features of this framework by constructing examples. Even if we’re principled about adding complexity, a cost of doing so is computational complexity. That is something the software was designed to manage. For the moment, we thus want to set aside concerns about realism vs. abstraction, about parsimony, and about error propagation, and we want to simply ask the question of how to build models with the features we want.
This chapter is an overview of the historical development of malaria models and an introduction to the toolbox. We’ll cover the same material in much greater detail in the chapters that follow, and we’ll construct examples using exDE and MicroMoB.