Agent Skills
› brycewang-stanford/Awesome-Journal-Skills
› amj-theory-development
amj-theory-development
GitHub针对AMJ论文，构建理论机制并推导可检验假设。解决无机制、HARKing或理论薄弱问题，确保假设源于明确逻辑链（前因-机制-结果-边界），强调先验推导与理论严谨性，区分理论与数据分析任务。
Trigger Scenarios
假设仅为预测缺乏机制解释
存在事后诸葛亮(HARKing)风险
审稿人指出理论单薄或逻辑不足
数据含中介/调节变量但未理论化
Install
npx skills add brycewang-stanford/Awesome-Journal-Skills --skill amj-theory-development -g -y
SKILL.md
Frontmatter
{
"name": "amj-theory-development",
"description": "Use when the theoretical argument and hypotheses are the bottleneck for an Academy of Management Journal (AMJ) manuscript — building a mechanism and deriving testable hypotheses a priori. Constructs the theory; it does not run the analysis (amj-data-analysis) or write the final contribution paragraph (amj-contribution-framing)."
}
Theory Development & Hypotheses (amj-theory-development)
When to trigger
- Hypotheses read as bald predictions ("A is positively related to B") with no mechanism
- You have results and are tempted to write hypotheses around them (HARKing risk)
- The argument leans on one borrowed citation rather than a developed logic
- Mediators/moderators are present in the data but not theorized
- A reviewer says "the theory is thin," "the logic is underdeveloped," or "why would this be true?"
The AMJ theory bar
AMJ does not publish atheoretical work — its mission is "to publish empirical research that tests, extends, or builds management theory." Every hypothesis must be derived from an articulated theoretical mechanism, written before the results are known. The argument should make a reader feel they understand why the effect occurs and under what conditions it strengthens or reverses.
For theory-building (typically qualitative) papers, AMJ does not expect a priori hypotheses; it expects a grounded model with emergent propositions. Eisenhardt and Graebner's AMJ guidance on building theory from cases is the canonical exemplar for how rich data become crisp constructs and propositions — do not force such a study into a hypothetico-deductive template.
Building a hypothesis (the mechanism chain)
For each hypothesis, write the explicit chain — do not skip steps:
- Antecedent — the predictor and why it matters in this context.
- Mechanism — the theoretical process linking antecedent to outcome (e.g., social exchange, signaling, attention allocation, identity threat, resource dependence). Name the theory and the process verb (motivates, signals, depletes, legitimates).
- Outcome — the dependent construct, defined at a stated level of analysis.
- Direction & form — sign, and whether linear, curvilinear, or threshold.
- Boundary — the moderator(s) and the theoretical reason the mechanism is stronger/weaker.
State the level of analysis explicitly (individual, team, firm) and keep the theory at the level where the mechanism operates; flag cross-level mediation as such.
Mediation and moderation done right
- Mediation: theorize the process construct before testing it; do not infer the mechanism from a significant indirect effect alone. An indirect effect is evidence for a theorized mechanism, not a substitute for the theory.
- Moderation: give a substantive reason the slope changes — "it depends on" is not a theory until you say why it depends.
- Curvilinear: justify the turning point theoretically (two opposing forces), not just by a significant squared term.
Hypothesis hygiene
- Number hypotheses and keep them to a defensible set (typically a focal effect plus a mechanism and one or two boundaries). Avoid a "kitchen sink" of ten weakly motivated predictions.
- Each hypothesis must be falsifiable and tied to a specific test you will run.
- Competing/alternative theoretical explanations should be named and addressed, not ignored.
- A figure of the theoretical model (boxes and arrows) should map one-to-one to the hypotheses.
Checklist
- Every hypothesis has an explicit mechanism, not just a predicted sign
- The driving theory is named and its process clearly invoked (not a one-line citation)
- Level of analysis is stated and consistent across argument and design
- Mediators/moderators are theorized a priori with substantive reasons
- Hypotheses were derived before analyzing the data (no HARKing)
- A theoretical-model figure matches the hypotheses exactly
- At least one rival explanation is named and engaged
Anti-patterns
- HARKing: hypothesizing after results are known; reviewers detect it when every hypothesis is supported and none is surprising.
- Bald predictions: "We hypothesize A is positively related to B" with no why.
- Borrowed theory: invoking a famous theory by name without using its actual logic.
- Mechanism by mediation: claiming a process exists only because the indirect effect is significant.
- Kitchen-sink hypotheses: many predictions, each thinly argued, that no single theory ties together.
- Boundary without reason: a moderator with no theoretical account of why the effect should differ.
Output format
【Focal theory】... (process verb: ...)
【H1 (focal effect)】antecedent → mechanism → outcome; direction/form; level
【H2 (mechanism/mediation)】...
【H3+ (boundary/moderation)】... reason the slope changes
【Rival explanation addressed】...
【HARKing check】hypotheses fixed before analysis? yes/no
【Model figure】matches hypotheses one-to-one? yes/no
【Next step】amj-literature-positioning, then amj-methods
Version History
- 1839142 Current 2026-07-05 12:14
