Chapter 1 DAGs in War

This chapter examines a deceptively simple but deeply revealing conflict scenario: Country Z launches an attack on Country I, and its ally, Country U, is drawn toward joining the assault. What begins as a single strike quickly activates alliance pressures, strategic fears, and political incentives that push U toward escalation. In this causal environment, Z’s attack triggers coordination mechanisms, U’s own military response, and ultimately a joint offensive against I. By mapping these relationships as a Directed Acyclic Graph (DAG), we expose the hidden pathways—alliances, norms, domestic politics, and relative power—that transform a localized crisis into a wider war. Understanding this structure is the first step toward redesigning it, so that the same initial shock can lead not to joint attack, but to ceasefire and peace.

War is not chaos. It is a causal chain shaped by:

Mediators that transmit escalation
Confounders that push multiple actors toward conflict
Effect modifiers that intensify reactions
Proxies that mislead decision‑makers
Co‑incident nodes that create spurious coordination

1.1 The Core Escalation Pathway

The war pathway is:

\[ B \rightarrow K \rightarrow C \rightarrow Y \]

B (Z attacks I) is the exposure.
K (alliance pressure) is the mediator transmitting escalation.
C (U attacks I) is the second mediator.
Y (joint attack) is the outcome.

This is the causal architecture of escalation. It shows how a crisis can move, step by step, from a single attack to a joint war. Understanding this structure is the first step toward redesigning it—toward blocking harmful paths, cutting mediators, neutralizing effect modifiers, and ultimately redirecting the flow toward restraint and peace.

1.2 Confounders in War

Several background conditions cause both Z’s attack and U’s response. These are confounder nodes:

A — Alliance structure
S — Strategic environment
P_u — U’s domestic politics
P_z — Z’s domestic politics
R — Relative power

These confounders open backdoor paths that make escalation more likely even before the first shot is fired.

1.3 Proxies and Misinterpretation

U’s mobilization is a proxy for escalation — it predicts war but does not cause Z’s attack. Misreading proxies as causes accelerates conflict.

1.4 Effect Modifiers

Relative power modifies the effect of B on C:

When U feels vulnerable, B → C is stronger.
When U feels secure, B → C is weaker.

1.5 Co‑Incident Nodes (Colliders)

“Joint response” is a co‑incident node: Z’s and U’s independent threat perceptions meet here. Conditioning on it creates coincidental (spurious) correlations, making actors believe their motives are aligned when they are not.

1.6 Complete War‑DAG

(with Confounders, Proxies, Effect Modifiers, and Colliders)

          A      S      P_u     P_z      R
         ↘      ↓       ↓       ↓       ↙
                 B  (Z attacks I)
                       ↓
                       K  (Alliance pressure)
                       ↓
                       C  (U attacks I)
                       ↓
                       Y  (Joint attack)

Additional causal elements:

Proxy:
U mobilizes  →  (predicts C but does not cause B)

Effect Modifier:
R modifies the strength of B → C

Collider:
Y is a co‑incident node where Z’s and U’s independent threat perceptions meet

Teaching‑Friendly Version

Confounders:
A, S, P_u, P_z, R
   ↘   ↓    ↓    ↓   ↙
                B  (Exposure)
                ↓
                K  (Mediator 1)
                ↓
                C  (Mediator 2)
                ↓
                Y  (Outcome / Collider)

Proxy:
U mobilizes → predicts C but does not cause B

Effect Modifier:
R modifies B → C

Collider:
Y is co‑incident (Z and U threat perceptions meet)

Description of the Complete War‑DAG

The war‑DAG reveals how a crisis that begins with a single attack can cascade into a wider conflict through a predictable sequence of causal pressures. What appears, from the outside, as a rapid and chaotic escalation is, in fact, a structured flow of causes and effects. Once the arrows are drawn, the logic becomes visible.

At the center of the war‑DAG is B, the initiating shock: Z attacks I. This exposure does not operate in isolation. It immediately activates K, the alliance‑pressure mediator. K represents the diplomatic expectations, reputational costs, and strategic obligations that push U toward involvement. In many conflicts, this step feels automatic: alliances are designed to respond, and once the machinery is engaged, it is difficult to slow.

From K, the causal chain moves to C, U’s military response. C is the second mediator in the escalation spine. It is the point where U shifts from deliberation to action, transforming the crisis from a bilateral confrontation into a broader conflict. Once U attacks, the system is no longer simply reacting to Z’s initial strike; it is generating its own momentum.

The chain culminates in Y, the joint attack. Y is a co‑produced outcome: Z and U now act together, whether by coordination, mutual reinforcement, or parallel escalation. What began as a single strike has become a joint offensive.

Surrounding this spine are the confounders—structural and political conditions that push both Z and U toward escalation even before the first shot is fired. Alliance structure (A), the strategic environment (S), domestic politics in U (P_u) and Z (P_z), and relative power (R) all shape the likelihood that B will lead to K, and that K will lead to C. These confounders open backdoor paths that amplify escalation pressures, making the war‑pathway more likely and more forceful.

Other nodes distort perception rather than behavior. U’s mobilization, for example, is a proxy: it predicts escalation but does not cause Z’s attack. Misreading proxies as causes accelerates conflict by creating false narratives about intent.

Relative power (R) also acts as an effect modifier. When U feels vulnerable, the effect of B on C is stronger; when U feels secure, the same attack may not trigger a military response. The strength of the causal arrow changes depending on the strategic balance.

Finally, the war‑DAG contains a crucial co‑incident node. The “joint response” or “joint attack” node is where Z’s and U’s independent threat perceptions meet. Conditioning on this node—looking only at cases where both countries responded—creates spurious associations between their motives and assessments. It can make independent decisions appear coordinated, even when they are not.

Interpretation of the Complete War‑DAG

The war‑DAG shows how a crisis that begins with a single attack can escalate through a structured sequence of causal pressures. At its center is the escalation spine:

\[ B \rightarrow K \rightarrow C \rightarrow Y \]

B — Z attacks I (exposure)
K — alliance pressure (mediator)
C — U attacks I (mediator)
Y — joint attack (outcome)

But this spine does not operate in isolation. It is surrounded and shaped by additional causal forces that make escalation more likely, more rapid, and more difficult to reverse.

1.6.1 ✔ Confounders (A, S, P_u, P_z, R)

These structural and political conditions cause both the exposure and the outcome, opening backdoor paths that amplify escalation:

A — alliance structure
S — strategic environment
P_u — U’s domestic politics
P_z — Z’s domestic politics
R — relative power

They increase the probability that:

B will trigger K
K will trigger C
C will culminate in Y

Confounders make escalation feel “automatic” even before the first shot is fired.

1.6.2 ✔ Proxy (U mobilizes)

U’s mobilization is downstream of the true cause and predicts escalation without causing it. Misinterpreting this proxy as a cause accelerates conflict by creating false narratives about intent.

1.6.3 ✔ Effect Modifier (R)

Relative power changes the strength of the causal arrow from B to C:

When U feels vulnerable → B strongly increases the likelihood of C
When U feels secure → B may not lead to C at all

The same exposure can have different effects depending on the strategic balance.

1.6.4 ✔ Collider (Y)

The joint attack Y is a co‑incident node where Z’s and U’s independent threat perceptions meet. Conditioning on Y — looking only at cases where both countries responded — creates spurious associations between their motives and assessments.

It can make independent decisions appear coordinated even when they are not.

Compact Notation

\[ (A, S, P_u, P_z, R) \rightarrow B \rightarrow K \rightarrow C \rightarrow Y \]

with:

Proxy: Mobilization → predicts C
Effect Modifier: R modifies B → C
Collider: Y is co‑incident

By the end of Chapter 1, the structure of escalation is no longer mysterious. The war‑DAG has revealed how a single attack can activate alliance pressure, trigger reciprocal strikes, and culminate in a joint offensive. We have also seen the deeper architecture that surrounds this spine: the confounders that predispose actors toward conflict, the proxies that distort interpretation, the effect modifiers that intensify reactions, and the co‑incident nodes that create the illusion of coordination. Once these elements are visible, escalation stops looking like a chain of accidents and begins to appear as a causal system—predictable, structured, and, crucially, open to intervention.

This realization changes the question. If escalation follows a recognizable causal pathway, then the pathway itself can be altered. The same arrows that once carried the crisis toward joint attack might be weakened, blocked, or redirected. Confounders can be neutralized, mediators transformed, effect modifiers softened, and co‑incident nodes avoided. In other words, the war‑DAG is not only a diagnostic tool; it is a blueprint for intervention.

Chapter 2 takes up this challenge. Using the causal grammar established in Chapter 1, we examine how the escalation spine can be interrupted at each link. We explore how diplomatic, institutional, political, and structural interventions can weaken the pull toward conflict and open channels for restraint. This is the beginning of the shift from mapping war to designing de‑escalation—from understanding how crises escalate to learning how they can be steered toward ceasefire and negotiation.