Can Agent-Based simulation models help us to improve services in complex WASH systems? Practitioners in the water, sanitation and hygiene (WASH) sector use a variety of modelling tools to guide them in understanding and improving service delivery. Examples include financial modelling in spreadsheet models, graphic information system-based (GIS) modelling for geographic mapping of infrastructure and conceptual flow modelling in a sanitation system. These tools are powerful in their respective area of interest. However, in this blog I advocate for the use of a complementary modelling tool that will help us to understand and analyse complex social interactions in WASH: an Agent-Based Modelling (ABM) tool. ABM can help practitioners to:
- diagnose the system;
- explore the effects of policy interventions; and
- discuss with partners and clients how the theory of complex systems affects them
This article was originally published on ircwash.org. To view the original post, click here.
By Felix Knipschild
Policy Interventions in WASH
In a previous IRC blog I described WASH as a Complex Adaptive System (CAS). This perspective taught us that a system concerned with the delivery of water or sanitation services is complex. Complexity theory investigates how relations between parts in a system result in a collective, observable behaviour. We can translate this to interactions and relationships between donor organisations, governments, service providers, technical infrastructure and water resources – the parts of the system – that result in a certain level of service delivery – the collective, observable behaviour.
How stakeholders and people involved in service delivery react to each other and to a policy intervention is very difficult to predict. (Policy) interventions can result in different outcomes due to unforeseen and unexpected reactions and interactions between the people, organisations and governments involved. This creates a major challenge for improving service levels.
In an ideal situation we are able to test any intervention we plan beforehand, turn back the time if the intervention does not go according to plan, and try a new intervention. Time and time again, until we get it right. Unfortunately, this is not possible. The best we can do, is make a well-educated guess of the effect of a policy. We can support this guess with calculations, by consulting stakeholders and by piloting the policy in an isolated environment. ABM is a method that investigates and anticipates the interactions between people and organisations. Furthermore, ABM can be used to test an intervention, turn back time, and test another intervention. The premise of ABM can be summarised as follows:
1. ABM is a tool for diagnosis
In an ABM a modeller determines relations between agents. These agents can be anything – a person, a hand pump, an organisation or a country – and form the key entities in an ABM. The agents are given a set of decision rules. Based on the defined relations and decision rules the agents interact with each other, resulting in some form of observable behaviour.
The core of an ABM is nothing more than lines of code. This means that the real system as we observe it has to be translated into simple code. To do so, you have to be very precise in defining relations. This demands a thorough understanding of your system of interest. It demands a clear understanding of relationships between stakeholders, their motives, service delivery models, budget and information flows, material availability, et cetera. Therefore, this system analysis phase is often done in consultation with stakeholders and based on thorough desk research.
2. ABM is a policy playground
Once completed an ABM creates a playground for policy analysis. The ABM enables us to test a certain implementation of a policy and simulate what happens over a period of time (depending on the time frame of your interest: minutes, days, months, years). Once tested, we can turn back time, change a parameter in the policy implementation, and run it again. And again. And again.
Often the relations between stakeholders or the decision rules that are given to the stakeholders hold one or more stochastic variables. This means that we define an interaction based on chance: if A happens, the chance is 70% that B will happen as well. With many stochastics present you cannot trust one outcome of the ABM. But it is possible to explore what scenarios are more likely to occur as a result of a certain policy intervention. We can do this by simulating the same policy intervention many times.
The added value of an ABM when used as a tool to test solutions to improve service delivery is not only the possibility to test and simulate many intervention scenarios. It is also provides a computational power. People often cannot predict how multiple stakeholders act and interact over time and how this alters the intended effects of a policy intervention. A computer cannot do this either. But what it can do is make hundreds of thousands of calculations based on the simplest decision rules within a system. We can visualise and interpret the outcomes of the calculations to draw conclusions from policy interventions.
3. ABM is a tool for advocacy
The tool offers us the opportunity to engage key stakeholders in a discussion about the necessities to improve service delivery. An ABM often includes a simple, dynamic visual representation of how the system of interest works. Furthermore, in the system analysis phase (diagnosis), assumptions about relations and decision rules are documented. This information provides the means to discuss how the system works, how service delivery can be improved and how this differs from the view of other discussants. This is valuable since the basis for effective collaboration is a shared understanding of each other’s perspective.
ABM in the WASH sector
We have to take into account that we simply cannot model the whole world. Therefore, there will always be some potentially relevant information excluded from our demarcation of the system. Another limitation of the tool is the possibility that the outcome of the ABM is simply the result of some crucial assumptions we make in the process. However, the field of ABM is developing and we have found ways to mitigate these limitations.
During my Master of Science Systems Engineering, Policy Analysis and Management thesis research, I explored how ABM can contribute to our understanding of learning between local governments in rural water service delivery. The premise of ABM has been tested in the WASH sector before and I believe that as our understanding of WASH systems grows, we are more and more certain that complexity theory offers us a handle to improve service delivery. Therefore I argue that we need to develop our tools accordingly so that they keep supporting our analyses. Agent Based Modelling could be one of the tools that can help improve WASH services.