Trigger timing
- The introduction is already written, but the "Theoretical Analysis and Research Hypotheses" section is still just a continuation of the literature review
- There is an empirical design and there are results, yet there is no ex ante logical deduction explaining "why this effect should exist"
- The hypothesis is written as "X affects Y", with no direction, no theoretical grounding, and no way to falsify it
- A reviewer asks, "How did you derive your hypothesis from theory?"
- H2 and H3 are proposed but there is no corresponding test anywhere later in the paper
Economic Research Journal places great emphasis on theoretical contribution. This section usually comes after the introduction and before the research design, and bears the full responsibility for "ex ante argumentation".
The two mainstream ways of writing the theoretical analysis section
Approach A: Mathematical model / microfoundations
Build a simplified model and, starting from microfoundations such as utility/profit maximization, derive testable propositions.
When formal modeling is needed: When the paper's core contribution lies in the mechanism itself, when there are multiple competing forces whose net effect must be characterized, or when counterfactual/welfare analysis is required.
When it is not needed: When the direction of the effect is clearly given by existing theory and modeling merely repackages intuition into algebra (a "decorative model")—reviewers will see through it at a glance.
The model must derivepropositions that correspond to the empirical specification, rather than being an isolated mathematical exercise.
命题 1：在假设 [A1]、[A2] 下，∂Y/∂D > 0，即处理 D 提高结果 Y。
（对应假设 H1，由式（3）对 D 求偏导得到。）
命题 2：该效应随 [调节变量 Z] 增大而增强，即 ∂²Y/∂D∂Z > 0。
（对应假设 H3。）
Approach B: Mechanism-based logical deduction (more common)
Based on existing theory, argue in layers how "treatment D affects outcome Y through mechanism M", and derive the hypotheses step by step. This is the most mainstream way of writing in Economic Research Journal.
Three-layer structure:
- Theoretical starting point: what D affects (grounded in a specific theory, such as principal-agent, financing constraints, or information asymmetry).
- Transmission mechanism: how D is transmitted to Y through M (this layer must fully explain the logic of M→Y; see the "methodological reminder" below).
- Expected result: therefore D has a directionally clear effect on Y → this becomes H1.
Norms for moving from theory to hypotheses
Each hypothesis (H1, H2, ...) must simultaneously satisfy:
- (i) Grounded in theory: state which literature/theoretical proposition it comes from;
- (ii) Clear direction: use "increase/decrease" or "strengthen/weaken", not "affect" or "correlate with";
- (iii) Falsifiable: one can write "if the hypothesis holds, X is observed; if not, non-X is observed".
H1：[处理 D] 显著[提高/降低][结果 Y]。
H2（机制）：[处理 D] 通过[机制 M]影响[结果 Y]。
H3（调节）：[调节变量 Z]越高，[D 对 Y 的效应]越[强/弱]。
Good hypotheses vs. bad hypotheses
| Dimension | Bad hypothesis | Good hypothesis |
|---|---|---|
| Direction | Digital transformationaffectsfirm performance | Digital transformationimprovesfirm performance |
| Theoretical grounding | Relies on intuition that "there should be a relationship" | Derived from [financing constraints theory] |
| Falsifiability | The policy "is of great significance" | The policy reduces Y, and the decline in subsample A is larger than in B |
| Connection to mechanism | H2 repeats H1 with different wording | H2 specifies that the H1 effect is realized through M |
Error:H：环境规制与企业创新存在关系。(no direction, not grounded in theory, not falsifiable)
Correct:H1：环境规制通过提高合规成本倒逼企业增加绿色研发投入（波特假说）。
Correspondence between hypotheses and later sections
Theoretical hypotheses must correspond one-to-one with the empirical structure; otherwise reviewers will question "proposing without testing."
| Hypothesis | Type | Corresponding empirical test |
|---|---|---|
| H1 | Baseline effect | Baseline regression (main table, column 1) |
| H2 | Mechanism | Mechanism test (→ er-mechanism) |
| H3 | Moderation / heterogeneity | Heterogeneity / interaction analysis (→ er-heterogeneity) |
| H4 | Competing hypothesis | Discriminating test of opposing mechanisms |
After writing the hypotheses, go back and check: can each H be matched to a corresponding subsection in the table of contents? Does each empirical subsection correspond to some H?
Division of labor between theoretical analysis and mechanism analysis
- Theoretical analysis (this section, ex ante): purely logical deduction, arguing "why this effect should exist and through which path it should operate." Regression results do not appear.
- Mechanism test (later, ex post): empirical evidence verifying whether the path deduced in the theory holds.
Error: pasting regression tables in the theoretical analysis section. Correct: theoretical analysis only discusses logic and derives testable propositions; leave the data and regressions to the research design and mechanism chapters.
Competing hypotheses / opposing mechanisms (bonus point)
A good theoretical analysis proactively proposes opposing expectations and explains how to distinguish them empirically. This is a clear plus for reviewers.
机制 A（融资约束缓解）预期：处理后高约束企业反应更大。
机制 B（监管套利）预期：处理后高约束企业反应更小。
本文通过[按融资约束分组的异质性检验]区分二者：若系数在高约束组更大，支持 A、否定 B。
Important methodological reminder: the burden of argument for M→Y lies in the theoretical stage
Under a causal framework, the mechanism variable M is usuallynot random(M is jointly determined by D and many unobserved factors). Therefore, empirically it is not appropriate to directly run a regression of Y on M to "prove the mechanism"—that coefficient has no causal meaning (see Jiang Ting 2022, "Mediation and Moderation Effects in Empirical Causal Inference Research," China Industrial Economics, 2022, Issue 5).
Therefore,the plausibility of M→Y is mainly borne by the theoretical analysis: in this section, explain thoroughly using existing theory "why M affects Y"; the empirical part only needs to solidly estimate the credible segment (usually D→M). If this theory section does not explain M→Y clearly, the later mechanism test will not hold up.
Must-check list
- Method A (mathematical model) or B (logical deduction) has been determined, and it matches the manuscript's positioning
- Each hypothesis is grounded in specific theory / literature (parenthetical citation: author, year)
- Each hypothesis has a clear direction (increase/decrease), with no vague phrasing such as "influence" or "correlation"
- Each hypothesis is falsifiable (one can write out the observable differences between holding and not holding)
- Hypotheses do not overlap (H2 is not a rephrasing of H1)
- Each hypothesis has a corresponding empirical test later (fill in the "hypothesis-test" correspondence table)
- The logic of M→Y is explained thoroughly in this section using theory (not relying on an endogenous regression of Y on M)
- At least one set of competing hypotheses / opposing mechanisms is considered, with an explanation of how to distinguish them
- If a mathematical model is used, propositions correspond one-to-one with the empirical setup (no decorative model)
Anti-patterns
- Theoretical analysis is written as a continuation of the literature review (piling up "who found what," without deriving this paper's hypotheses)
- Assume no direction: "X affects Y" rather than "X increases/decreases Y"
- Hypotheses overlap / are mere rewording of each other, padding the count
- Propose H2, H3 but do not test them later
- "Reverse-engineer" a theory for a known empirical result (HARKing: look at the results first, then add hypotheses)
- Mathematical model disconnected from the empirical setup (model variables are never measured later)
- Paste regression tables directly into the theoretical analysis section, turning "ex ante deduction" into "ex post explanation"
- Use "the regression of Y on M is significant" as evidence of a mechanism, yet never theoretically argue M→Y
Output format
【写法】A 数理模型 / B 逻辑推演
【假设数】H1…Hn
【方向性】全部明确 / 待修：[...]
【挂理论】每个 H 对应文献（作者，年份）
【竞争性假设】有 / 无（如何区分：...）
【M→Y 理论论证】充分 / 待补
【假设-检验对应】完整 / 缺口：[...]
【下一步】er-identification