brycewang-stanford Argument Development & Logic Check (amr-data-analysis)
AMR publishes NO empirical data. There is nothing to estimate, plot, or test. The "analysis" in an AMR paper is the analysis of the argument itself: does each proposition follow logically from the constructs and mechanisms? At AMR, logical soundness plays the role that statistical rigor plays at empirical journals.
The empirical-analog reframe (keep the folder, change the content)
This skill replaces an empirical "identification + robustness" stage. The mapping:
| Empirical sibling (AMJ/ASQ/SMJ) | AMR theory analog |
|---|---|
| Identification strategy (IV, DiD, RD, matching) | Generative mechanism — the why (Whetten 1989, DOI 10.5465/amr.1989.4308371) |
| Robustness checks / alternative specifications | Internal consistency + counterfactual probes on premises |
| Ruling out confounders | Engaging and bettering the strongest rival theory |
| Replication package (data + code) | Transparent reasoning — premises and derivations a reader can re-derive |
| "Estimates are significant and robust" | Propositions are falsifiable in principle (AMR's "testable knowledge-based claims") |
There is no instrument, no parallel-trends test, no placebo here; their presence signals a misfiled empirical paper.
When to trigger
- Propositions are written but you are not sure they actually follow from the argument
- The theory "feels right" but has not been adversarially tested
- A reviewer would raise an alternative explanation you have not addressed
- The argument chain has hidden leaps between premises
The four logic tests
Run every proposition through these before drafting.
1. Premise-to-conclusion check (per proposition)
For each Pn, write the chain explicitly: premise → premise → mechanism → conclusion. If any step is missing, the proposition is asserted, not derived. Use a Toulmin frame: claim / grounds / warrant / backing / rebuttal. The warrant (the mechanism that licenses the inference) is where most theory papers are thin.
2. Thought experiment / counterfactual
Manipulate the focal construct in your head and trace the consequence: "If construct X
rose sharply while everything else held, what does the theory predict for Y, and is that
prediction sensible?" Then run the counterfactual: "Under what condition would X move and
Y not follow?" If the counterfactual is plausible and unexplained, you are missing a
boundary condition (route back to amr-theory-development).
3. Alternative-explanation audit
For each proposition, name the strongest rival theoretical account of the same relationship. Then either (a) show why your mechanism is more complete/parsimonious, or (b) integrate the rival as a boundary condition. Ignoring rivals is the fastest path to a reject — reviewers are the rival theorists.
4. Disconfirming-case search
Actively look for a case where the proposition should fail. A theory that "explains everything" explains nothing. Either the disconfirming case is covered by a stated boundary condition, or the proposition needs to be narrowed.
Internal-coherence checks across the whole theory
- Consistency: no two propositions contradict each other (unless the tension is the point and is theorized). Constructs mean the same thing throughout — no concept drift (a core Suddaby construct-clarity criterion, AMR 2010, DOI 10.5465/amr.2010.0419).
- Non-circularity: a construct is not defined by its effects, then used to explain those effects.
- Sufficiency: the constructs and mechanisms are enough to generate the propositions — nothing is smuggled in mid-argument.
- Parsimony: every construct earns its place; drop any that does no logical work.
Exemplar: Oliver (AMR 1991, DOI 10.5465/amr.1991.4279002) "analyzes" by argument — deriving a typology and propositions from antecedent conditions and addressing why organizations might resist rather than conform (the rival expectation) — all logic, no data.
Checklist
- Each proposition has an explicit premise → mechanism → conclusion chain
- The warrant (mechanism) for each inference is stated, not assumed
- A thought experiment has been run on each focal relationship
- Counterfactuals are addressed by boundary conditions, not ignored
- The strongest alternative explanation for each proposition is named and handled
- A disconfirming case has been sought for each proposition
- The theory is internally consistent, non-circular, sufficient, and parsimonious
- No empirical evidence is invoked as proof (AMR has none)
Anti-patterns
- Propositions presented as self-evident, with the argument left to the reader
- Hand-waving the mechanism ("it stands to reason that...")
- Defending the theory by asserting it would be "supported by data" — there are no data
- Ignoring the obvious rival theory the reviewers hold
- A theory that cannot be wrong: no boundary, no disconfirming case, no rebuttal addressed
- Circular reasoning: defining a construct by the outcome it is meant to explain
Output format
【Per-proposition logic】P1: chain ok? / gap at warrant? ... Pn
【Thought experiments run】[focal construct → predicted consequence]
【Counterfactuals → boundary conditions】[...]
【Alternative explanations handled】[rival → resolution]
【Disconfirming cases】[case → covered by boundary / narrow proposition]
【Coherence】consistent / non-circular / sufficient / parsimonious : pass/fix
【Next step】amr-contribution-framing