A Diagram of Activity Is Not a Theory of Change
Most social justice organizations do not have a theory of change. They have a diagram of activity.
Many have a logic model buried somewhere in a grant application. Some have updated it once. A few have actually used it.
The problem is not that logic models are useless. The problem is that most logic models in social justice settings were designed to satisfy a funder rather than to illuminate how change actually happens. They describe what a program does. They rarely explain why any of it matters, what assumptions are holding the whole thing together, or what the organization would do if the conditions changed.
That gap matters most in the kinds of work that cannot be reduced to a contained intervention.
Consider a community coalition working to reduce housing instability among youth aging out of foster care. A logic model for that initiative might look straightforward: staff time, funding, and community partnerships support housing navigation services and case management, which lead to successful placements, which lead to increased housing stability over time.
There is nothing obviously wrong with that description. But it does something subtle and consequential. It centers the program as the primary driver of change and treats everything else, the housing market, landlord behavior, policy constraints, the economic realities of young people exiting care with few resources, as background conditions rather than as active forces shaping whether the logic holds at all.
The model is not wrong. It is partial. And in complex systems, partial explanations can be more misleading than no explanation at all.
It is also worth asking who benefits when the explanation stays partial. Organizations that fund simplified models of change are often the same ones that fund the constraints shaping whether that change is possible. A theory of change that names those constraints makes certain conversations harder to avoid.
A theory of change begins where that partial explanation breaks down. It asks a different set of questions, ones that most logic models are not designed to hold. What has to be true in the system for this outcome to be possible? Who has the power to make those conditions hold or fail? What are we assuming about how change happens that we have never tested? And what happens if those assumptions are wrong?
This guide walks through how to build that kind of explanation, not by abandoning the logic model, but by treating it as a starting point and then deliberately expanding it until it can carry the weight of the work it is supposed to represent.
The housing coalition appears in each step, so you can see not just what the process requires in the abstract but what it actually does to a model when you take it seriously.
Why a Logic Model Is Not Enough
The appeal of the logic model is easy to understand. It imposes order on complexity. It translates messy, relational, politically contested work into a sequence that can be named, diagrammed, and evaluated. For organizations accountable to funders, boards, and community stakeholders, that translation is genuinely useful.
A well-constructed logic model builds a common understanding of program design, identifies where the causal logic is weak or missing, and points to a balanced set of key measurement areas. It’s a road map that highlights how a program is expected to work and what activities must precede others. For programs with relatively contained, predictable causal chains, that map is sufficient.
Most social justice work does not operate in contained systems.
The difference becomes clearer when you think about the range of problems organizations try to solve. Following a recipe is straightforward: replicate the steps and get the same result. Launching a rocket is complicated but ultimately predictable if you control the variables. Raising a child is neither. The outcomes are emergent. Context shapes everything. The same inputs produce different results across different children, families, and conditions. No formula covers it.
Social justice initiatives are closer to the third category. They involve multiple actors pursuing competing goals, operate across contested political environments, depend on relationships that take years to build, and aim for outcomes, such as shifts in institutional culture or changes in public narrative, that cannot be fully specified in advance. In these conditions, simple logic models risk overstating the causal contribution of any one program while rendering invisible the conditions and feedback loops that actually drive change.
Return to the housing coalition. A logic model for that initiative shows placements producing stability. It does not show what happens when a landlord declines to renew a lease after the initial placement period. It does not show how eligibility criteria for subsidized housing exclude some of the youth most in need. It does not show the ways case management, when structured around compliance and documentation rather than trust, can drive youth away from the very services designed to support them.
Those outcomes are not unpredictable. They are happening. The model just does not show them.
A theory of change does not solve every one of those problems. But it forces a reckoning with them. It requires the organization to name the system it is operating in, surface the assumptions that hold the causal logic together, account for power dynamics shaping who benefits and who does not, and build in the capacity to revise its understanding when conditions change.
Step One: Build the Foundational Logic Model
Before complicating things, build a solid structural foundation. The logic model is that foundation. Even if your theory of change ultimately looks quite different, the model gives you and your stakeholders a shared visual language for examining what you think you are doing and why.
Map five core categories: four that follow a left-to-right causal chain, and one that shapes all of them from below.
Resources and Inputs are everything the initiative requires in order to function: staff time, funding, relationships, community trust, data, organizational capacity, and the lived expertise of people most affected by the problem. Be honest here. A model that lists “strong community partnerships” as a resource when those partnerships are still being built is telling a false story before the work has even started. For the housing coalition, this includes funding from the state agency, two case managers, relationships with six landlords willing to consider referred tenants, and the knowledge base of a youth advisory panel that includes former foster youth.
Activities are what the initiative actually does, both the visible program work, such as housing navigation sessions, landlord recruitment, and case management, and the less visible relational infrastructure that makes the visible work possible. Think carefully here about which activities are genuinely critical to goal attainment and which are redundant or have implausible connections to desired outcomes.
Outputs are the direct, countable products of those activities: the number of youth navigated, the placements facilitated, the landlords who agreed to participate, the sessions delivered. Outputs tell you whether your activities happened. They do not tell you whether they mattered.
Outcomes distinguish between short-term changes, those most directly associated with your outputs; intermediate changes, those that result from applying short-term gains over time; and longer-term outcomes, the broader shifts that follow from sustained intermediate progress. For the housing coalition, short-term outcomes include youth securing housing placements. Intermediate outcomes include sustained tenancy past the three-month mark. Long-term outcomes include housing stability as a foundation for employment, education, and reduced involvement with systems.
External Influences belong in the model explicitly, not in a footnote. These are the contextual factors outside the program’s control that will determine whether the logic holds: the local rental market, zoning and subsidy policy, economic conditions shaping youth employment, and the political environment governing foster care transition support. Naming these prevents the model from implicitly overpromising what any single initiative can produce.
Once those categories are populated, read the model as a series of conditional statements. If these resources are available, and if these activities occur, then these outputs will result. If those outputs reach these participants under these conditions, then these short-term changes will follow. The Kellogg Foundation’s logic model development guide calls this the “if, then” structure of the logic model, and it is the right frame, provided those conditional statements are treated as hypotheses rather than guarantees.
Where the chain feels thin, where the “then” does not convincingly follow from the “if,” you have found the edge of what the logic model can explain. That edge is where the theory of change begins.
Step Two: Build the Theory of Change That Sits Behind the Model
A theory of change does not replace the logic model. It is the explanatory framework that makes the logic model honest.
A logic model shows what a program does. A theory of change starts from the long-term change you want to see in the world and works backward through the preconditions and assumptions that would have to hold for that change to be achievable.
Start with the long-term outcome.
For the housing coalition, the anchor is not placements. It is sustained housing stability over time. Stable housing, not temporary placement, is what shifts life trajectories for youth aging out of care. That distinction matters. It changes what has to be true for the initiative to succeed.
Working backward from sustained stability, you quickly encounter dependencies the logic model does not show. Stability requires affordability sustained past the initial placement period. It requires income, which depends on employment access, transportation, and whether employers will hire young people with foster care histories. It requires landlord relationships that hold when a tenant misses a payment or needs support rather than eviction. It requires that youth trust the system enough to engage with case management rather than disappearing from services when things become difficult.
None of those conditions are produced by housing navigation alone. All of them are causal strands in the theory of change. Rogers (2008) calls these “simultaneous causal strands,” parallel pathways that must all be in place for the intervention to produce the intended result. She argues that for complex community initiatives, a single causal chain is almost always an oversimplification.
This is also where assumptions surface. The coalition may be assuming that landlords who participate in the program will remain engaged over time. It may be assuming that youth who secure initial placements will maintain contact with case managers. It may be assuming that the state agency will sustain funding past the first year. Those assumptions are often reasonable. They are also often untested, and they operate as invisible premises in the logic model.
Writing them explicitly changes their status. They are no longer given. They are claims about how the world works that can be examined and revised.
For advocacy and organizing work, the backward-mapping process surfaces a different set of dependencies. Klugman (2011) argues that in social justice advocacy, policy change alone is an insufficient long-term outcome because implementation can fail and gains can be reversed unless organizational capacity, movement infrastructure, and normative shifts are sustained alongside the policy victories. The Advocacy and Policy Change Composite Logic Model developed by Coffman et al. (2007) operationalizes this by identifying the interim outcomes that advocacy strategies must produce before policy change becomes possible: coalition power, narrative reframing, the emergence of new champions in decision-making roles, shifts in public will. A theory of change for an advocacy initiative needs to include those interim outcomes explicitly, because without them, the pathway from activities to policy change has no visible mechanism.
At this stage, the model stops being a diagram of activities. It becomes an argument about change, one that can be examined, tested, and revised.
That argument does not operate only through programs and policies. It is also shaped by how problems are understood in the first place. We have explored this in depth in Narrative as Infrastructure, where storytelling is treated not as communication, but as a structural force that shapes what solutions feel possible and legitimate. A theory of change that ignores narrative is leaving one of its core causal mechanisms unexamined.
Step Three: Model the Negative Logic
This is the step most organizations skip. It is also the one that matters most in social justice settings.
Onyura et al. (2021) introduce the concept of dark logic modeling, drawn from public health evaluation. For every pathway you have mapped toward a positive outcome, a parallel pathway exists along which the intervention could fail to produce change or actively produce harm. Dark logic modeling asks you to map that pathway before the program runs, so that mitigation can be built into the design rather than discovered in the aftermath.
The examples from the literature are instructive. Cultural competency trainings intended to reduce bias have been shown in some contexts to surface and even legitimize implicit views rather than shifting them, producing worse outcomes than no training at all. Data systems designed to improve service coordination have exposed undocumented participants to risk when privacy protections were inadequate. Leadership development programs have tokenized participants when the structural supports for genuine decision-making power were absent.
In social justice work, this problem is acute because many initiatives operate with communities that have long histories of being harmed by well-intentioned programs. This means the adverse outcomes are often not unpredictable at all. They are predicted, by community members, in advance. The question is whether those predictions are treated as credible data that should shape program design.
For the housing coalition, the dark logic pathways are visible if you look directly at them. A landlord recruitment strategy that prioritizes ease of engagement may result in a pool of participating landlords who exclude the youth at greatest risk. Data collected to improve coordination may, without explicit protections, create records that follow youth into encounters with law enforcement or future housing applications. Case management structures built around compliance, attendance requirements, documentation, and regular check-ins, may drive away the youth most in need of flexible support, the ones for whom rigid structure represents the conditions that have already failed them.
These are not hypothetical. They are recurring patterns in programs serving youth exiting foster care.
Mapping them changes who is centered in the analysis. Instead of asking only whether the program works, the dark logic model asks for whom it works, under what conditions, and at whose expense. Programs do not simply fail. They fail in patterned ways, and the patterns are usually visible before the program runs if you know where to look. That question is not an add-on. It is part of what makes the theory of change honest.
Step Four: Stress-Test the Causal Logic
The theory of change now contains a set of causal claims. Some are well-supported by evidence. Some are grounded in practice wisdom. Some are assumptions that have never been directly tested. Treating them as equally certain is one of the fastest ways to undermine the usefulness of the model.
Onyura et al. (2021) recommend two forms of analysis. A direct logic analysis asks whether the program design aligns with available evidence. For the housing coalition, that means examining what the research shows about the relationship between short-term navigation and long-term stability, and whether the case management model being used reflects what has actually produced durable outcomes in comparable populations.
A reverse logic analysis asks whether other pathways to the same outcome exist that this initiative has not considered. If the evidence suggests that housing vouchers without attached services produce better long-term stability than case management models, the coalition does not have to abandon its approach. But it has to grapple honestly with that finding rather than writing over it with confident claims.
The question is how strong the evidence is for each hypothesis in the chain. Where that evidence is weak or absent, the model needs to either be revised, explicitly marked with uncertainty, or targeted for more rigorous evaluation.
What it should not do is present a chain of confident causal statements that have never been seriously interrogated. In a funder-facing document, that kind of overconfidence is common. In a theory of change intended to guide real decisions about real people, it is irresponsible.
A causal claim that has never been interrogated is not a plan. It is a hope wearing the clothes of one.
Step Five: Design for Participation
Up to this point, the work described in this guide can be done entirely within an organization. That is also where it is most likely to go wrong.
The theory of change is only as accurate as the knowledge that informs it. In social justice work, a significant portion of the knowledge that matters most, knowledge about where systems actually fail, where trust breaks down, what support looks like from the inside, sits with people who are rarely treated as co-authors of program design.
For the housing coalition, youth who have exited foster care hold knowledge that no literature review or staff meeting can replicate. They know which landlords treat tenants differently once the caseworker stops checking in. They know the specific moments when young people disengage from services and why. They know which success indicators reflect what they actually need and which reflect what is convenient to count. Community partners, frontline staff, and landlords hold different pieces of that same system.
Braithwaite et al. (2012) document what genuine participation looks like in practice through their community-based participatory evaluation model, developed with the Healthy Start project of the Augusta Partnership for Children. Their model moves through nine stages, beginning with recruiting both community members and evaluation specialists to the same committee from the start of the process, not after the design is complete. Community members are oriented to the evaluation process, win-win dynamics are actively cultivated, and program aims are bilaterally articulated. Assessment instruments are designed, selected, and pilot-tested with community input before any data collection begins.
The diagram depicting this process is a spiral rather than a linear sequence, with what the authors call “community intelligence” and “cultural appropriateness” running through every stage. That shape is an argument. Evaluation is shaped by whose perspectives are treated as credible. A process designed to extract validation from community members produces a different model than one designed to incorporate their knowledge into the explanatory framework itself.
Scarinci et al. (2009) document what that difference looks like when it actually occurs. In their multi-state participatory evaluation initiative, community partners did not affirm the logic model that academics had designed. They reshaped it. They restructured working groups by intervention level, rather than by the cancer site categories that made sense to researchers, because that structure better reflected how they understood the problem. They pushed back on assessment instruments they experienced as burdensome academic exercises. They defined success on their own terms, and that definition produced a different model than the one the grant had funded.
Three lessons emerge from that process that apply directly to theory of change development: constant and open dialogue among partners, flexibility to revise the theory as community input accumulates, and evaluators who act as facilitators between community knowledge and technical expertise rather than as top-down designers.
This is not about inclusion as a procedural value. A model built without the knowledge of those most affected by the problem will systematically miss key parts of how change happens. The moral case is compelling. The methodological claim is undeniable.
Easterling et al. (2023) confirm this in a different context. In a multi-site participatory logic modeling process across seven National Cancer Institute centers, engaging funded groups as genuine partners produced a more accurate and more complete model than the funder had initially developed. Grantees identified contextual factors that inhibited success, operationalized assumptions that had been left vague, and added health equity dimensions the original model had not captured. The process took longer. The resulting theory of change better reflected how change was actually expected to happen and was more likely to be owned and used across the initiative.
For the housing coalition, bringing youth advisors into the theory of change development process changes the model. It surfaces the compliance-driven case management problem before it is built into the design. It shifts outcome indicators from placement counts to something closer to what stability actually means in a young person’s life. It identifies the landlord relationship dynamics that the staff model assumes away. The theory of change that results is not just more equitable. It is more accurate.
Step Six: Treat the Model as a Living Document
The final mistake most organizations make is finalizing the theory of change and filing it away.
Onyura et al. (2021) are direct: logic models and theories of change should be treated as dynamic rather than static, with an expectation that they will evolve as contexts shift and as evaluation data accumulates. For initiatives with emergent outcomes, a series of evolving models developed alongside the work is more appropriate than a single fixed diagram produced in advance.
For the housing coalition, implementation will test the model in real time. If placements increase but tenancy past the three-month mark does not, the assumptions linking short-term and intermediate outcomes need to be revisited. If landlord participation fluctuates, the recruitment and retention strategy is not producing the conditions the model assumed it would. If youth disengage from case management, the model’s assumptions about trust and service design are incomplete in ways that matter.
Each of those moments is not a failure of the model. It is the model doing its job, revealing where the current explanation of change does not hold.
Easterling et al. (2023) describe this ongoing revision as essential rather than optional. In their case study, the initiative’s Health Equity Task Force used the logic model as a diagnostic tool, asking at each stage where equity was explicitly represented and where it was absent, then incorporating those findings into updated versions of the model. The result was not a different model than the one they started with. It was a more honest one.
The conceptual shift that makes this possible is moving from attribution to contribution. Rather than asking whether the organization can prove it caused an outcome, ask how it is contributing to change alongside other actors in a system it does not control. That question makes revision less threatening and more generative. When the model changes, it is not evidence that the work has failed. It is evidence that the organization is learning.
What You Are Actually Building
By the time this process is complete, the housing coalition, or the advocacy campaign, or the community organizing initiative, has more than a logic model.
It has a structured description of what it does, an explicit explanation of why those activities are expected to matter, a mapped account of how they could fail or cause harm, a set of assumptions tested against available evidence, a design that reflects the knowledge of those most affected by the problem, and a process for revising that understanding as the work unfolds.
The model it started with showed staff time and partnerships producing placements producing stability.
The theory of change it now has shows the housing market conditions that make stability possible or impossible, the income pathways that must run alongside housing navigation, the landlord dynamics that determine whether placements hold, the trust conditions that shape whether youth remain engaged with services, the policy environment that either expands or forecloses what the initiative can accomplish, and the assumptions about all of it that are currently being treated as facts.
Most organizations stop at the first model.
The ones producing durable change build the second one, not because the process is elegant, but because the systems they are trying to change are not simple enough to respond to activity alone.
The logic model tells the performance story. The theory of change tells the truth behind it.
