Scientific Argumentation

---
name: scientific-argumentation
description: Building rigorous scientific arguments — claim hierarchies, evidence standards, handling objections, and logical fallacies to avoid.
category: scientific
---

## Overview

Science advances by argument: claims supported by evidence, tested
against alternatives, and revised under criticism. This skill covers
structuring arguments (claim hierarchies, warrants, qualifiers),
matching evidence strength to claim strength, steelmanning objections,
and the logical fallacies that most often corrupt scientific reasoning.

## When to use

- Structuring a paper's argument: main claim, sub-claims, evidence chain
- Preparing for peer review: anticipating and preempting objections
- Evaluating competing hypotheses or interpretations fairly
- Writing discussion sections that persuade rather than assert
- Spotting weak reasoning in your own drafts (or others' papers)

## Core concepts

- **Claim hierarchy:** main claim → supporting sub-claims → evidence for each. If a sub-claim falls, know whether the main claim falls with it (load-bearing) or survives (redundant support) — design arguments with redundancy where possible.
- **Toulmin structure:** claim, grounds (evidence), warrant (why the evidence supports the claim), backing (why the warrant holds), qualifier (how strongly), rebuttal (when it wouldn't hold). Making warrants explicit exposes hidden assumptions.
- **Evidence standards scale with claim strength:** extraordinary claims (new phenomena, overturned paradigms) need extraordinary evidence — multiple independent lines, not one clever analysis.
- **Alternative hypotheses:** the strongest argument explicitly considers and rules out the best alternatives — "we considered X; the data disfavor it because..." beats ignoring X and hoping reviewers don't notice.
- **Qualifiers as honesty:** "suggests", "is consistent with", "demonstrates" encode different evidence strengths — match the verb to the evidence; hedging everything and claiming everything are both failures.
- **Falsifiability:** state what observation would change your conclusion — an argument that can't specify its own defeat conditions isn't scientific reasoning, it's advocacy.

- **Inference to the best explanation:** when deduction isn't available, argue comparatively — your hypothesis explains more of the evidence with fewer assumptions than alternatives (but "best" needs explicit criteria, not gut feeling).
- **Robustness reasoning:** the same conclusion from independent methods/data is stronger than any single line — design arguments to converge; a conclusion resting on one analysis is fragile.
- **Dialectical structure:** anticipate the strongest objection, state it fairly, answer it — the ancient structure (thesis, antithesis, synthesis) still organizes the most persuasive discussions.

## Practical workflow

### 1. Map the argument

1. Write the main claim in one sentence. List the 2–4 sub-claims it depends on. For each, list the evidence and the warrant connecting them.
2. Mark load-bearing sub-claims (main claim fails without them) vs supporting ones — strengthen or add redundancy to the load-bearing set.
3. For each warrant, ask: what assumption am I making? Is it stated? Is it justified?

### 2. Stress-test against alternatives

1. List the 2–3 strongest alternative interpretations of your evidence — the ones a smart skeptic would propose.
2. For each: what does it predict that differs from your account? Do your data discriminate? If not, say so and propose the discriminating test.
3. Steelman, don't strawman: present alternatives in their strongest form — defeating a weak version persuades no one who matters.

### 3. Calibrate claim strength

1. Match verbs to evidence: "demonstrates/proves" (decisive, replicated), "shows/establishes" (strong, single-study), "suggests/indicates" (preliminary, correlational), "is consistent with" (compatible, not discriminating).
2. Distinguish what the data show from what you infer — label the boundary explicitly ("These results show X; we infer Y because...").
3. State limitations as boundary conditions on the claim, not as ritual apologies — "our conclusion holds for [conditions]; beyond them, [caveat]".

### 4. Write it to persuade

1. Lead with the strongest evidence; order supporting arguments from strongest to weakest — primacy effects are real.
2. Address the obvious objection before the reader forms it — preemption defuses; omission invites.
3. End discussions with the refined claim: what, precisely, should the reader now believe, and with what confidence?

### 5. Write a Discussion that argues, not summarizes

1. Don't re-list results — interpret them: what mechanism do they support, what alternatives do they rule out, what remains ambiguous?
2. Confront the literature directly: where you agree, say why the convergence matters; where you disagree, explain the discrepancy (methods? populations? era?) rather than ignoring it.
3. End with the precise, qualified claim the reader should take away — hedged where evidence is thin, firm where it converges.

### 6. Quick-reference checklist

- [ ] Central claim stated precisely and qualified appropriately
- [ ] Evidence mapped to claims (every claim has support)
- [ ] Strongest alternative stated fairly and addressed (steelmanned)
- [ ] Independent lines of evidence converge where possible
- [ ] Limitations presented as boundary conditions, not buried
- [ ] Hedging matches the evidence (firm where convergent, cautious where thin)
- [ ] Jargon and undefined terms eliminated
- [ ] Takeaway sentence written — what should the reader carry away?

## Common pitfalls

- **Affirming the consequent:** "if theory T, expect X; we see X; therefore T" — X may follow from alternatives too; rule them out or weaken the claim.
- **Texas sharpshooter:** drawing the hypothesis around the data pattern — preregistration and holdout tests are the defense.
- **Motivated stopping:** ending the analysis when the result looks right — predefine stopping rules and robustness checks.
- **Absence of evidence as evidence of absence:** "no significant effect" with n=12 is not "no effect" — report power/precision, not just p > 0.05.
- **Causal language from correlational designs:** "X drives Y" from observational data without identification strategy — use causal verbs only with causal designs.
- **Single-study syndrome:** treating one experiment as decisive on a contested question — converge multiple lines of evidence before strong claims.
- **Strawmanning alternatives:** defeating a weak version of the rival hypothesis persuades no one who holds the strong version — steelman or don't bother.
- **Adverbial hedging as a substitute for evidence:** "possibly", "might", "could suggest" stacked three deep — either the evidence supports the claim or it doesn't; say which.