Causal Loop Diagrams: Method in a Nutshell


Like most complex analysis, causal loop diagramming is iterative. The steps outlined below are meant to serve as a high-level guide to the process rather than a strict sequencing.

Key steps in conducting a causal loop diagramming include:

  1. Define your learning question
  2. Define CLD parameters
  3. Identify stakeholders
  4. Collect & model data
  5. Analyze CLD
  6. Share results


The first step in developing a CLD is to establish what you are trying to better understand. Your goal may be to understand a system (e.g., healthcare in general) in its entirety or a sub-part of the system (e.g., prenatal healthcare). You may be trying to understand how a system/organization operates or characterize a context (i.e., problem space) to see how a phenomena or problem emerged and is sustained by related processes, stakeholder behaviors, and perceptions. Or, maybe the goal is to capture and convey a theory of change that underlies a new program or initiative. Whatever the goal may be, it should be established through consultations with key stakeholders before modeling efforts commence.


Since CLDs are often used to understand complex issues, modelling efforts can get overwhelming quickly unless proper parameters have been identified and agreed upon by key stakeholders. Some of the key parameters to consider include:

  • CLD Scope: Establishing boundaries to determine the scope of the modeling effort is critical to avoid developing an unnecessarily complex model. There are different ways to establish scope for a CLD. For example, developing definitions for key concepts associated with the theme of investigation makes some things part of the CLD while leaving others outside the scope of assessment. Similarly, geographical or temporal boundaries can help establish scope for a CLD assessment.
  • Level of Abstraction: The goal of a CLD is never to model everything – in fact a key point to remember about systems analysis is that it helps manage complexity by taking a step back and seeing the “big picture.” As such, determining the level of abstraction for the modeling process ensures keeping everyone on target while ensuring consistency in treatment of various issues across the CLD. A CLD should not depict one part of the system in significant detail while only providing high level coverage of another part. The desired level of abstraction can be established through an initial discussion with key stakeholders and is often linked to the research question. While providing a high level overview may be appropriate for a CLD that provides a general understanding of a local environment or problem context, a detailed depiction may be more appropriate for a CLD that zooms into a problem issue. Those involved in development of a CLD should be mindful of the inevitable tradeoff between depth and breadth and make decisions based on the desired analytic product and purpose of the analysis.
  • Number of CLDs: Although most learning efforts will require development of an all-encompassing, single CLD, there may be cases in which multiple CLDs are warranted for a complete assessment. For example, if you would like to contrast the “before and after” states of an organization or a community following an intervention to characterize differences in processes, perceptions, and behaviors, developing two CLDs may be a desirable goal. If you are exploring issues or problems that operate as a system of systems, nested or linked causal loop diagrams can be developed to allow analysis of cross-cutting relationships and dynamics. Similarly, if the goal is to visually depict alternative initiatives or programs and their associated theories of change, you will need more than one CLD to facilitative a comparative assessment. How many CLDs will need to be developed is closely tied to the learning question and should be part of the initial stakeholder consultations. However, CLD development process is iterative and requires a flexible approach as the necessity for additional CLDs may be determined during the modeling effort in light of a growing understanding of the system.
  • Data sources: The information that feeds into a CLD can come from various sources. It could rely on literature review and previous studies, interviews with experts and stakeholders, focus group discussions, or working group sessions that physically bring key stakeholders together. Usually it is best to use multiple data sources for information triangulation and representation of both objective and subjective realities. Moreover, CLDs have shown to be most effective when developed through participatory modeling process in which different stakeholders argue for and reconcile their perspectives. As such, depending on the learning question and the phenomena being studied, a CLD that only reflects a literature review may fail to capture key issues just as a CLD that conveys only stakeholder-driven information may neglect dynamics that are unknown to the specific group involved.


If the CLD incorporates information elicited from stakeholders, who will be included in the modeling effort needs to be determined. Once the high level categories of stakeholders (e.g., academics, practitioners, local farmers) associated with the issue of interest are determined, specific individuals who will represent each category need to be identified and contacted. One key consideration in this stage is to ensure that diverse stakeholders as well as diverse perspectives within each stakeholder category are represented during the data collection efforts.


During this step, a literature review is conducted to identify key variables and relationships relevant to what is being modelled. Literature reviews can include previous studies, program evaluations, government documents, statistics, newspaper articles, and any other documentation that relate to the identified learning question. A critical component of this literature review is to investigate behavior over time associated with the key variables identified. If data are to be collected from stakeholders or key informants, semi-structured interview questions and focus group questions should be prepared. Alternatively, stakeholders can be brought together in person for a real-time, facilitated discussion and participatory group modeling. If the CLD will be developed through participatory group modeling process, several sessions will be conducted to capture all relevant perspectives as well as emergent ideas and thoughts. During the first session, stakeholders can be presented with a simple, core feedback loop to kick-start discussions.

During the data collection process, some of the key questions considered include: What are the key variables, issues, forces, dynamics and outcomes essential to explain this system or problem? How do they relate to one another? What are some of the key cause-effect relationships, interactions and interdependencies? How can these relationships be reflected in terms of reinforcing (a series of relationships that appear to cause exponential growth or decline in a phenomena) and balancing loops (a series of relationships that appear to prevent change with a push in the opposite direction)? Which one of the effects are immediate and which ones are delayed? How do stakeholders perceive each other, their place in the system and key dynamics associated with the problem? What are some of the economic, social, political, and cultural norms and structures in place and how do they influence the operation of the system and key outcomes? Are there any real or perceived delays in cause-effect relationships identified?

Often data collection and CLD modelling happen simultaneously. As data accumulates and our understanding of key dynamics and forces evolve, mapping key variables and relationships begin.

Once a CLD is formed, its variables or relationships can be color coded to convey another layer of information. For example, variables associated with different domains (e.g., economic or social) can be colored differently; similarly, arrows (that represent relationships) can be color-coded to reflect different types of relationships (dependency, information flow, compliance, etc.).

CLD development is an iterative process. Before the analytic team and key stakeholders feel comfortable with the resulting CLD, the model will almost always go through several revisions and adjustments in light of new information and group learning. Similarly, once a baseline CLD is developed, it can periodically be updated to reflect the ways the system, local context, or problem may be changing as a result of program interventions or natural evolution.


Once a CLD is developed, it is time to take a step back and examine the model for original insights. Synthesis of different perspectives and information often reveals information that you cannot see by examining individual parts. A CLD lends itself to different types of qualitative assessment, including:

  • Trend analysis: In this type of assessment, empirical trends about key STEEP (social, technological, economic, environmental and political) variables associated with important outcomes are researched, analyzed and overlaid to the model. Trend analysis is often conducted to anticipate the direction of the system or problem in the near future. Trend analysis is a critical input into decisions that are concerned with prioritization of problem areas and response initiatives (particularly where multiple programs and initiatives may be considered) and help make resource allocations decisions.
  • Causal pathway analysis: This is an explicit assessment of inputs and outputs associated with key outcomes. Usually outcomes of a causal pathway serve as inputs into another casual pathway, highlighting the complex connections present in a system or problem context. Along each causal pathway, inputs and outputs are sorted into different groups (technology, economic, social, resources, methods, perceptions, etc.) and ways of measuring them along with related indicators are identified.
  • Leverage analysis: In this type of analysis, a CLD is analyzed to identify systemic levers (actionable points for intervention) for positive change and assess their effectiveness. While high leverage intervention points enable system-wide, lasting change with relatively small resources, low leverage points in a system allow for limited change that requires continuous application of resources to sustain positive results. Leverage analysis often relies on the seminal work (1999) by Donella Meadows, who identified 12 different places to intervene in a system with differing levels of effectiveness such as constants, parameters, numbers; driving positive feedback loops; the rules of the system; and the distribution of power over the rules of the system. Accordingly, a CLD is analyzed to determine how many of the twelve actionable points are present in the depicted system/context. Alternatively, a CLD can be examined to identify leverage points based on other forms of qualitative assessments such as feedback loop intensity and influence/exposure scoring of individual variables. Available intervention points are then assessed for alignment with program objectives and stakeholder desires as both high and low leverage interventions may be needed in a problem context.
  • Cascading effects analysis: Since everything is related in complex systems, a change in a key variable often causes changes in other, sometimes distant parts of the system, as its effects travel through extensive causal pathways. A cascading effect analysis can help us anticipate unintended consequences of programmatic actions and take timely action to offset related dynamics. Cascading effects analysis not only enriches our understanding of a specific problem, but also improves our ability to assess feasibility and desirability of various programmatic actions. Where cascading effects are inevitable, this type of analysis can alert us to the timeline of effects expected, and whether simultaneous interventions in different parts of the system are needed to remedy negative developments.
  • Complex Adaptive Systems Assessment: In complex adaptive systems (CASs), we cannot control or dictate actors’ behaviors – we can only hope to influence them. However, previous research has identified key principles that help with managing human behavior in complex systems. With CAS assessment, local incentive and sanction structures that govern stakeholder behaviors are analyzed to see how they can be modified for alignment with practices and principles that are known to work in CASs. Once a CLD is developed, problem-related outcomes and their connections to various human behaviors are explicitly analyzed to uncover critical incentive and sanction structures that are built into the local system. Understanding how to modify these structures in order to mobilize and motivate people towards desired outcomes is a critical task in many programmatic interventions.


For the best results, CLDs should not be presented as a single page analytic product. Typically, a complex CLD can be presented in two ways:

  • Story-Boarding: One way to present a complex CLD is to story-board it. In story-boarding, a CLD is broken into consumable chunks that represent meaningful, distinct parts of a broader story. These parts then presented through consecutive scenes to tell a story, gradually building towards key findings.
  • Causal Trees: Another way to present a CLD and its findings is through causal trees. In causal trees, selected causal pathways are visualized to depict how a series of variables contribute to a particular outcome of interest or how a selected variable impacts a series of other variables in a system.