PHIL252-Unit7

PHIL 252 — Unit 7: Science & Causal Reasoning

Unit 7 asks: how do we think critically about scientific claims? It covers how causal claims are made, how they can fail, and how study design can corrupt conclusions before reasoning even begins.

graph TD
    U7[Unit 7: Science & Causal Reasoning]
    U7 --> A[Analogy in Science]
    U7 --> B[Causation vs. Correlation]
    U7 --> C[The False Cause Family]
    U7 --> D[Selection Bias Variants]
    A --> A1[False Analogy]
    B --> B1[Probabilistic / Sufficient / Necessary]
    C --> C1[Post Hoc · Mere Correlation · Spurious]
    C --> C2[Data Dredging · Slippery Slope · Irrelevant Thesis]
    D --> D1[WEIRD · Extrapolation · OSE]
    D --> D2[Berkson's · Data Censoring · Right-Censoring]

Analogy in Science

Science uses analogy to reason from familiar → unfamiliar. A good analogy is relevant, insightful, partial, provisional, and fruitful. A false analogy compares things that are only superficially similar, or similar but not in the relevant respect for the argument being made.

See Analogy.

Causation vs. Correlation

Correlation = two variables move together. Causation = one variable produces the other. Correlation is easy to measure; causation is hard to prove. Three types of cause:

• Probabilistic: A raises the chance of B
• Sufficient: If A, then always B (but B can happen without A)
• Necessary: Without A, B cannot happen (but A alone doesn’t guarantee B)

See Causation.

The False Cause Family

A family of fallacies that claim causation without adequate evidence:

1. Post Hoc Ergo Propter Hoc — B followed A, so A caused B
2. Mere Correlation — A and B move together, so A causes B
3. Spurious Correlation — A and C correlate because a hidden factor B drives both
4. Data Dredging — mining datasets until a chance pattern appears
5. Slippery Slope — A will inevitably lead to B, C, D… (asserted, not argued)
6. Irrelevant Thesis — divert with an unrelated issue, claim the original is settled

See FalseCause.

Selection Bias as a Threat to Causal Claims

Even sound causal reasoning fails if the sample is corrupted. Six variants:

1. WEIRD Populations — Western, Educated, Industrialized, Rich, Democratic samples
2. Extrapolation — applying results from one group to a different group
3. Observation Selection Effects — observer’s position is linked to the variable measured
4. Berkson’s Paradox — selection filter creates a negative correlation that doesn’t exist in the general population
5. Data Censoring — non-random dropout corrupts a sample that started clean
6. Right-Censoring — study ends before the event being measured has fully occurred

See SelectionBiasVariants.

Key Points for Exam/Study

• Analogy is legitimate but must be relevant to the conclusion — not just superficially similar
• Correlation ≠ causation; always ask for a mechanism
• Know all three cause types cold: probabilistic / sufficient / necessary
• Keyword test: “always” → Sufficient; “unless/required” → Necessary; “raises odds” → Probabilistic
• Mere correlation vs. spurious correlation: the difference is whether a hidden third factor is identified
• Berkson’s Paradox requires: filtered sample + at least one of two factors required for entry + fake negative correlation
• Observation Selection Effect: you can only see the survivors/returners/volunteers — the missing data is the important data
• Right-Censoring is a structural feature of studies that end before all events occur, not deliberate fraud

Cross-Unit Connections

InformalFallacies — False Cause, Slippery Slope, and Irrelevant Thesis are all informal fallacies
Bias — Selection bias is the data-context version of cognitive bias
FallaciesOfAmbiguity — Unit 6 fallacies; Unit 7 extends the fallacy taxonomy into causal reasoning
DataVisualization — misleading graphs often exploit selection bias and spurious correlation