causal-loop-analysis

Causal Loop Analysis

Extract the implicit causal claims in a policy argument and make them fully explicit — as a named-loop Mermaid diagram with cross-connections, a layer model where applicable, structural actor positions, and a ranked leverage point menu. This is the Takshashila publication standard for causal analysis.

Why This Matters

Every policy argument contains a theory of change: "if we do X, then Y will follow." These causal chains are often:

• Assumed rather than argued
• Linear when they are actually circular (feedback loops)
• Missing second-order effects and unintended consequences
• Overstated in their certainty ("X will cause Y" vs. "X is associated with Y under conditions A and B")

Making causal structure explicit forces the author to:

• Defend every causal link
• Acknowledge feedback loops that may dampen or amplify the intended effect
• See second-order consequences before a reviewer does

---

Vocabulary

Symbol	Meaning
`-->	+
`-->	-
`-->	~+
`-->	~-
R loop	Reinforcing loop: A → B → A (amplifying — virtuous or vicious cycle)
B loop	Balancing loop: A → B → -A (stabilizing or constraining)

---

Process

Step 0: State the Core Causal Chain

Before drawing loops, state the argument's central causal logic as a single sentence chain:

"deny X → deny Y → deny Z" "subsidise A → attract B → generate C → reduce D"

Then ask: which link is the weakest? Which link, if broken, would collapse the entire argument?

This step forces clarity before complexity. Do not skip it.

Step 1: Extract Causal Claims

Read the piece and list every causal claim, explicit or implicit.

• Explicit: "PLI subsidies will attract foreign investment"
• Implicit: (if the argument is that PLI → jobs, there is an implicit link: foreign investment → jobs)

For each claim, note:

• Source node → Target node
• Link direction (positive or negative)
• Is there a delay?
• Is this link cited/evidenced, or assumed?

Step 2: Build the Mermaid Diagram

Render the causal structure as a Mermaid flowchart. Use flowchart LR (left to right) for linear chains; flowchart TD (top down) for hierarchical structures.

flowchart LR
  A[Node A] -->|+| B[Node B]
  B -->|+| C[Node C]
  C -->|-| A

Keep node labels short and concrete. Avoid abstractions as node names.

Step 3: Identify and Name Loops

Walk through the diagram and identify all feedback loops. For each:

• Type: Reinforcing (R) or Balancing (B)
• Label: R1, R2, B1, B2 etc.
• Descriptive name: Give each loop a plain-language name that captures its character. Not just "R1" — name it (e.g. "indigenous acceleration loop", "ecosystem flywheel", "debt-trap spiral", "talent exodus cycle"). The name should be memorable and explain what the loop does.
• One-sentence description: What does this loop do, and what would break it?

Example:

• R1 "Indigenous Acceleration Loop": China's substitution efforts increase AI output on own hardware, which validates further substitution investment, which increases output again. Breaking this loop would require showing that domestic hardware cannot scale.
• R2 "Ecosystem Flywheel": US chip exports to China feed industry revenue, which funds R&D, which produces better chips, which attract more exports. Cross-connection: China's R1 output directly undermines this by eroding US market share.

Step 3b: Map Cross-Connections Between Loops

The most analytically important insight is often where one loop undermines or amplifies another.

For each pair of loops, ask:

• Does output from Loop A feed into Loop B (amplification)?
• Does output from Loop A undercut a node in Loop B (erosion)?

State cross-connections explicitly:

"China's domestic AI output (R1) directly erodes US chip industry revenue — a key node in R2's ecosystem flywheel. This cross-connection means that R1 and R2 are structurally antagonistic: the stronger R1 gets, the weaker R2 becomes."

Cross-connections are the strategic insight. Surface them prominently.

Step 3c: Layer Model (if applicable)

If the system has stacked interdependencies — where one layer must exist before the next can function — name the layers and show which layer the intervention targets.

Example (AI compute):

Layer 1: Energy → Layer 2: Chips → Layer 3: Infrastructure → Layer 4: Models → Layer 5: Applications

Show: Which layer does this intervention target? What adaptation pressure does it create on other layers? Can actors substitute within a layer or must they build the whole stack?

Step 4: Flag Unsupported Links

For each causal link in the diagram:

• Is this link cited in the piece, or assumed?
• If assumed: is it reasonable (well-established in the literature) or contested?
• Mark contested/uncited links explicitly.

Step 5: Identify Missing Links (Second-Order Effects)

What happens after the intended causal chain ends? What effects does the argument not trace?

• Does the intervention have unintended consequences in adjacent domains?
• Are there time-horizon effects (short-run positive, long-run negative, or vice versa)?
• What happens if a key assumption fails?

Step 6: Identify Leverage Points

Where in the causal map could intervention be most effective? (Meadows' hierarchy of leverage points — simplified):

• Breaking a vicious reinforcing loop
• Strengthening a virtuous reinforcing loop
• Reducing a balancing loop's resistance
• Changing a key parameter in the chain

Also identify external leverage points — constraints or actors outside the causal diagram that could shift the system (e.g., US export controls in a semiconductor analysis, an IMF condition in a fiscal analysis, a judicial ruling in a regulatory analysis).

Step 6b: Rank Leverage Points

After identifying leverage points, rank each on two dimensions:

Magnitude — how much does intervening here change system behaviour?

• High: Breaking or strengthening this would fundamentally shift the outcome
• Medium: Meaningful improvement but other constraining loops remain
• Low: Marginal effect; other parts of the system absorb the change

Accessibility — how feasible is intervention here given political, institutional, and technical constraints?

• High: Within the mandate of a single actor; no major political barriers
• Medium: Requires coordination across actors; politically contested but possible
• Low: Requires overcoming deeply entrenched interests or changing parameters outside domestic control

Plot each leverage point on a 2×2 (Magnitude × Accessibility). Top-right quadrant (High × High) = priority interventions.

Note: High-magnitude but low-accessibility leverage points are worth naming even if not actionable — they clarify why easy interventions fail.

Step 7: Policy Intervention Menu

Translate the ranked leverage points into concrete, actor-specific recommendations:

For each priority leverage point:

• "To break loop [R1]: [specific intervention] — who must act, what they must do"
• "To strengthen loop [B1]: [specific intervention]"
• Distinguish: necessary conditions (without which the outcome is impossible) from sufficient conditions (alone enough to achieve the outcome)
• Note: interventions targeting contested causal links are risky leverage — flag if the link itself is unsupported

---

Output Format

## Causal Loop Analysis

### Core Causal Chain

[State the argument's central logic as a single sentence chain]
> "X → Y → Z"

**Weakest link:** [Which link, if broken, collapses the argument?]

### Causal Map

```mermaid
flowchart LR
  [nodes and links here]

Loop Inventory

ID	Name	Type	Nodes	What sustains it	What breaks it
R1	"[Descriptive name]"	Reinforcing	A → B → A	[driver]	[breaking condition]
R2	"[Descriptive name]"	Reinforcing	B → C → B	[driver]	[breaking condition]
B1	"[Descriptive name]"	Balancing	A → C → -A	[what it constrains]	[what overwhelms it]

Cross-Connections Between Loops

[For each pair of interacting loops:]

"Loop R1's output [node X] directly [undermines/amplifies] a key node in Loop R2. This means R1 and R2 are structurally [antagonistic/synergistic]: the stronger R1 gets, the [weaker/stronger] R2 becomes."

Layer Model (if applicable)

Layer 1: [foundation] → Layer 2: [next] → Layer 3: [next] → Layer 4: [output layer]

• Which layer does this intervention target?
• What adaptation pressure does it create on adjacent layers?
• Can actors substitute within a layer, or must they build the whole stack?

Structural Actor Positions

[Where does each key actor sit in the causal map?]

• Actor A: [position, e.g. "consumer-integrator — benefits from Layer 4 outputs but does not control Layers 1–2"]
• Actor B: [position]

Unsupported Links

Link	Issue	Recommendation
A → B	Assumed; no citation	Cite [source type] or qualify with "may"

Missing Second-Order Effects

[What the argument doesn't trace that a reviewer might raise]

Leverage Points

Leverage Point	Loop Affected	Mechanism	Magnitude	Accessibility	Priority
[Intervention]	R1	[How it affects the loop]	High/Med/Low	High/Med/Low	1/2/3

Priority matrix:

              LOW ACCESSIBILITY    HIGH ACCESSIBILITY
HIGH MAGNITUDE   Important but       Priority
                 hard — name it      interventions
                 anyway

LOW MAGNITUDE    Deprioritise        Easy wins

Policy Intervention Menu

Priority interventions (High magnitude × High accessibility):

• To [break/strengthen] loop [ID] ("[loop name]"): [Specific intervention] — Actor: [who] / Action: [what]
  • Necessary condition? Yes / No
  • Risk: [if this causal link is contested or unsupported]

Important but difficult (High magnitude × Low accessibility):

• [Intervention] — blocked by [constraint]. Worth naming to explain why easier interventions fail.

External leverage points:

• [Constraint or actor outside the causal diagram that shapes the system]

Key Causal Vulnerabilities

[1–3 links in the chain that, if broken, would undermine the central argument — flag which of these are also contested/unsupported links]

---

## Notes on Mermaid in Obsidian and GitHub

- Mermaid diagrams render natively in Obsidian (no plugin needed in newer versions).
- GitHub Markdown renders Mermaid in `.md` files natively.
- For Google Docs: export the Mermaid code and render via mermaid.live, then paste as image.
- Keep node count under 15 for readability. For complex arguments, break into sub-diagrams.

---

## Example (Semiconductor PLI argument)

Argument: "India's PLI scheme for semiconductors will attract fab investment, which will create skilled jobs, develop supplier ecosystems, and ultimately reduce import dependence."

### Core Causal Chain

> "PLI subsidies → fab investment → domestic capacity → reduced import dependence"

**Weakest link:** PLI subsidies → fab investment. Subsidy alone does not overcome India's infrastructure and ecosystem gaps; this is the most contested step.

### Causal Map

```mermaid
flowchart LR
  A[PLI Subsidies] -->|+| B[Foreign/Domestic Fab Investment]
  B -->|+| C[Fab Capacity]
  C -->|+| D[Skilled Job Creation]
  C -->|+| E[Supplier Ecosystem Development]
  E -->|+| F[Input Cost Reduction]
  F -->|+| B
  C -->|-| G[Semiconductor Imports]
  G -->|-| H[Import Dependence]
  D -->|+| I[Talent Pool]
  I -->|+| B

Loop Inventory

ID	Name	Type	Nodes	What sustains it	What breaks it
R1	"Ecosystem Flywheel"	Reinforcing	Investment → Ecosystem → Cost Reduction → Investment	Domestic supplier density	Persistent infrastructure gaps that keep input costs high despite ecosystem growth
R2	"Talent Accumulation Loop"	Reinforcing	Investment → Talent Pool → Investment	Education pipeline depth	Brain drain; failure to retain talent domestically

Cross-Connections Between Loops

R1 and R2 share the Investment node — both loops strengthen when investment rises. However, R2 runs on a slower clock than R1: the education pipeline takes 8–12 years. This timing mismatch means R1 can stall before R2 provides enough talent to reinforce it, creating a window of fragility where neither loop is self-sustaining.

Layer Model

Layer 1: Land/Power → Layer 2: Equipment Supply → Layer 3: Fab Capacity → Layer 4: Chip Output → Layer 5: Design Applications

PLI targets Layer 3 directly. Adaptation pressure falls on Layers 1–2 (India's infrastructure deficit) and Layer 5 (existing chip design firms may not benefit from fab-side import substitution). Actors cannot easily substitute within Layer 2 — equipment is controlled by a small number of US/Dutch/Japanese suppliers.

Structural Actor Positions

• India: Positioned at Layer 5 (design/applications) with aspiration to build Layer 3. Does not yet control Layers 1–2.
• Foreign fabs (TSMC, Micron): Gatekeepers of Layer 3; their investment decisions determine whether the R1 loop fires at all.
• Equipment suppliers (ASML, Applied Materials): Structural chokepoint at Layer 2; outside India's direct policy reach.

Unsupported Links

Link	Issue	Recommendation
PLI subsidies → Fab investment	Assumes subsidy overcomes infrastructure/ecosystem gaps	Cite comparative PLI evidence from Vietnam/Malaysia fabs or qualify with "contingent on infrastructure co-investment"
Fab capacity → Supplier ecosystem	Assumes local supplier density will emerge; may not if global supply chains remain cheaper	Cite component localisation data from existing Indian electronics PLI

Missing Second-Order Effects

• Chip design firms may not benefit from fab-side import substitution — design-side import dependence persists.
• If fabs remain uncompetitive on cost after PLI, the R1 loop stalls but government subsidy commitment continues (fiscal drain without strategic gain).
• Talent trained for semiconductor fabs may emigrate, weakening R2.

Key Causal Vulnerabilities

1. PLI subsidies → Fab investment (weakest link; contested): If foreign fabs judge India's infrastructure inadequate, the entire chain fails at step one.
2. Fab capacity → Supplier ecosystem (assumed; slow): Supplier ecosystems take 15+ years in comparable cases; the PLI timeline may be too short for this link to fire.
3. Talent Pool → Investment (delayed): R2 depends on an education pipeline that does not yet exist at scale for advanced semiconductor manufacturing.