Correlation and Causation

Foundations of Inductive and Causal Logic

• Epistemological Nature of Inductive Arguments:

  • Inductive arguments are probabilistic rather than deductive; they formulate factual claims about empirical reality.
  • Factual claims possess contingent truth value, meaning they are contingently true—circumstances in the physical world could have unfolded differently than they actually did.
  • Contingent truths contrast fundamentally with necessary truths, which are propositions that must be true under all possible circumstances and whose truth can be established with absolute certainty (e.g., "If someone is a bachelor, then they are an unmarried man").
  • The evaluative metric applied to inductive claims is a continuum of higher versus lower degrees of probability rather than deductive validity or sound certainty.

• Nature and Scope of Causal Arguments:

  • Causal arguments are arguments structured to substantiate a causal claim or provide a causal explanation, asserting that an identified effect is produced by a specific cause.
  • Causal assertions represent empirical claims about the contingent state of the world; they are only probabilistically true and remain inherently vulnerable to falsification.
  • Because causal arguments are fundamentally inductive, alternative explanations always remain logically and empirically possible.

• Hypothetical Scenarios of Causal Underdetermination:

  • The Alarm Scenario:
    • Primary causal claim: "I awoke because my alarm went off."
    • Alternative competing hypotheses: Waking up due to a loud commercial truck passing the residence; waking up because a pet cat, Zara, began meowing for food; waking up as a consequence of a nightmare; or waking up due to an extraterrestrial abduction.
  • The Food Contamination Scenario:
    • Primary causal claim: "I contracted food poisoning because I consumed bagged lettuce contaminated with Escherichia coli (E. coli) bacteria."
    • Extended causal claim: "The purchased bag of lettuce was contaminated with E. coli because it was processed and packaged at an industrial facility that simultaneously handles beef products."
    • Alternative competing hypotheses: The pathogen responsible was Salmonella rather than E. coli; the vehicle of transmission was undercooked poultry rather than bagged salad; the contamination originated from agricultural irrigation run-off at the farm site; or the illness resulted from inadequate hand hygiene by an individual preparing the meal.

Philosophical Foundations: David Hume on Causation and Regularity

• Hume's Empirical Critique of Causal Necessity:

  • Philosopher David Hume posited that causation is fundamentally an epistemic habit of the mind developed through sensory experience rather than an observable feature of mind-independent reality.
  • Under Hume's analysis, when human experience repeatedly observes phenomenon $A$ sequentially followed by phenomenon $B$, the mind develops a conditioned expectation that future observations of $A$ will inevitably precede $B$.
  • While Hume's reduction of causality to psychological association has been widely challenged by modern philosophy of science, his critique remains fundamental to understanding the gap between correlation and causation.

• Fallacy of Regular Succession (The Rooster Paradox):

  • Human experience routinely observes strong empirical associations that lack genuine causal relations.
  • For example, the vocalization of a rooster's crow is strongly and consistently associated with the rising of the sun over the horizon.
  • Despite this temporal regularity, the rooster's crow does not cause the sun to rise; treating the association as causal conflates regular temporal succession with an active causal mechanism.

Principles and Criteria for Distinguishing Correlation from Causation

• Definition and Function of Correlation:

  • A correlation is formally defined as an association, link, or connection between two distinct kinds of phenomena or variables.
  • Statistically, a dataset exhibits correlation when the frequency or magnitude of occurrences between variables moves synchronously (increasing or decreasing together).
  • A correlation in data points to three distinct possibilities:
    • It may reflect an underlying causal link.
    • It may signify a non-causal relationship that illuminates how two phenomena interact.
    • It may represent a statistical coincidence (such as the rooster's crow coinciding with dawn).

• Analytical Utility of Non-Causal Correlations:

  • Correlations that lack direct causal mechanisms still yield diagnostic insight into systemic conditions.
  • Breakfast and Academic Performance:
    • Consistently eating breakfast is strongly correlated with higher academic achievement among school-aged children.
    • This statistical relationship does not imply that breakfast consumption directly causes increased intelligence or academic success.
    • Instead, the correlation highlights that nutritional stability and food security are critical baseline indicators of student well-being.
    • Such non-causal correlations provide empirical justification for concrete socioeconomic policy interventions, including local, state, or federal programs that allocate tax revenues to reduce food insecurity in schools.

• Evidentiary Thresholds for Establishing Causation:

  • Establishing causation demands an evidentiary standard substantially higher than observing statistical association.
  • According to Ravi Dhawan and Denys Shay (Principles of Causation, National Library of Medicine: National Center for Biotechnology Information, July 2024), establishing causality requires satisfying multiple rigorous criteria:
    • Temporal Consistency: The hypothesized cause must reliably precede the observed effect in time, while recognizing that complex phenomena often involve multicausality.
    • Statistical Strength and Predictability: The correlation within the data must be robust, consistent, and predictable across measurements.
    • Explanatory Mechanism: A plausible mechanism (e.g., biological pathways, physiological processes, or structural social dynamics) must be identified to explain how the cause produces the effect.
    • Replication Across Contexts: Findings must be verified and reproducible across diverse experimental studies, populations, and environmental settings.

Typology of Statistical Correlations

• General Classification:

  • Within empirical datasets, relationships between variables are classified into three distinct categories: positive correlation, negative correlation, or no correlation (zero correlation).

• Positive Correlation:

  • Definition: A positive correlation exists between variables $A$ and $B$ if the values of $A$ tend to increase and decrease in synchrony (parallel) with the values of $B$.
  • Operational Test: In the dataset, variable $A$ reliably increases when variable $B$ increases, and variable $A$ reliably decreases when variable $B$ decreases.
  • Scatterplot Characteristics: Data points trend upward from bottom-left to top-right. In a condition of some positive correlation, data points cluster loosely along an upward trajectory; in a perfect positive correlation, data points align along a straight diagonal line.
  • Exemplar: Plotting the number of hours spent reading against the total number of books read demonstrates a positive correlation: as reading hours accumulate, the total count of books completed increases concurrently.

• Negative Correlation:
  • Definition: A negative correlation exists between variables $A$ and $B$ if $A$ tends to increase when $B$ decreases, and conversely, $A$ tends to decrease when $B$ increases.
  • Operational Test: Variables move in opposite directions within the dataset; when one variable increases, the other systematically declines.
  • Scatterplot Characteristics: Data points trend downward from top-left to bottom-right. Points cluster along a downward slope in some negative correlation, while aligning along a straight descending path in a perfect negative correlation.
  • Exemplar: Plotting the rate of alcohol consumption per hour against cognitive coordination and memory demonstrates an inverse relationship: as hourly alcohol intake increases, measured coordination and memory faculties systematically decrease.
  • The Canadian Banana Tree Problem:
    • In general discourse, the term "correlation" is often mistakenly assumed to refer exclusively to positive correlation, ignoring inverse associations.
    • Consider the correlation between being Canadian and cultivating a backyard banana tree: within a random sampling of people, the higher the percentage of Canadians in a group, the lower the incidence of backyard banana trees.
    • This represents an authentic, measurable negative correlation dictated by geographic and climatic conditions.

• Zero Correlation (No Correlation):
  • Definition: No correlation exists between variables $A$ and $B$ if variable $A$ changes haphazardly as variable $B$ changes.
  • Operational Test: Variables operate entirely independent of one another, displaying no linear or statistical predictability within the data.
  • Scatterplot Characteristics: Data points are distributed randomly and evenly across the coordinate space without an identifiable linear slope or clustering vector.
  • Exemplar: Comparing the number of documented Bigfoot sightings per month against the monthly volume of frisbees purchased yields no correlation; shifts in one variable show zero statistical relationship to shifts in the other.

Methodological Fallacies and Applied Case Studies

• Case Study: Methodological Flaws in "Rapid Onset Gender Dysphoria" (ROGD):

  • The Causal Hypothesis: Proponents of the ROGD hypothesis claimed that adolescent gender questioning was driven by a "social contagion." They argued that young people identifying as transgender or nonbinary (NB) frequently consumed digital media produced by and about trans/NB individuals, asserting that media consumption caused youth to become trans/NB.
  • The Correlational Reality: The observed relationship represents a non-causal correlation. Digital consumption of queer media among trans/NB youth can be explained without social contagion framing, evidenced by the fact that many individuals who consume identical media do not identify as trans or nonbinary.
  • Flawed Operationalization: In studies reporting ROGD, the diagnostic variable was operationalized through subjective parental questionnaires. Parents surveyed frequently held unsupportive or transphobic perspectives and perceived their child's gender disclosure as sudden.
  • Direct Youth Testimonies: Qualitative interviews conducted directly with transgender and nonbinary youth revealed sustained, long-term personal processing of their gender identity that preceded any disclosure. Disclosures appeared sudden to parents primarily because unsupportive home environments forced youth to conceal their identities until they felt safe.

• Case Study: Youth Mental Health Trends and Confounding Variables:

  • Jonathan Haidt's Thesis in The Anxious Generation:
    • Haidt identifies historical trends showing that through the $1990\text{s}$ and $2000\text{s}$, Millennial mental health markers remained stable, tracking slightly better than Generation X.
    • Beginning around $2012\text{--}2013$, adverse mental health indicators began rising sharply.
    • While the rise was gradual for boys, data for adolescent girls displayed a sharp upward inflection ("a very sharp elbow").
    • The increase was pronounced among younger girls aged $10$ to $14$, who exhibited spikes in emergency hospitalizations for self-harm—an outcome that had historically been exceptionally rare.
    • Haidt's Causal Conclusion: The abrupt rise is causally driven by the expansion of smartphones and algorithmic social media platforms targeting adolescents.
  • Confounding Factors and Alternative Explanations:
    • Cultural Destigmatization: The post-$2012$ period coincided with broader public destigmatization of mental healthcare, facilitated in part by social media, which may have led to higher reporting rates rather than an exclusive increase in raw pathology.
    • Healthcare Access via the Affordable Care Act (ACA): Implemented beginning in $2010$, "Obamacare" expanded health insurance coverage mandates for pediatric and adolescent populations, increasing clinical visits and hospitalizations.
    • Clinical Screening Guidelines: In $2011$, the US Preventive Services Task Force revised clinical guidelines to recommend universal depression screenings for adolescent girls, mandating that commercial insurance plans cover these diagnostic and therapeutic interventions.

• Case Study: Ectopic Pregnancy Mortality and Abortion Legislation:

  • Empirical Dataset: Analysis by Lucas Waldron and ProPublica utilizing CDC WONDER multiple cause of death and natality data across three-year cohorts ($2014\text{--}2016$, $2017\text{--}2019$, $2020\text{--}2022$, and $2023\text{--}2025$) evaluated mortality rates involving ectopic pregnancies per million live births.
  • Categorization Criteria: States implementing abortion bans at $6\text{ weeks}$ of gestation or earlier were classified as having a "strict abortion ban."
  • Reported Statistical Trajectories:
    • States with strict abortion bans: Maternal mortality involving ectopic pregnancies rose from approximately $5.0\text{ per million live births}$ ($2014\text{--}2016$) to $5.5$ ($2017\text{--}2019$), climbed to $9.8$ ($2020\text{--}2022$), and reached $13.1$ ($2023\text{--}2025$).
    • National Average: Traveled from approximately $4.2\text{ per million}$ ($2014\text{--}2016$) to $4.5$ ($2017\text{--}2019$), rose to $8.0$ ($2020\text{--}2022$), and settled at $9.8$ ($2023\text{--}2025$).
    • States without abortion bans: Shifted from approximately $3.8\text{ per million}$ ($2014\text{--}2016$) to $3.8$ ($2017\text{--}2019$), increased to $5.8$ ($2020\text{--}2022$), and concluded at $7.1$ ($2023\text{--}2025$).
  • Analytical Evaluation (Correlation versus Causation):
    • The data confirms a strong positive correlation between the implementation of strict abortion bans and accelerated rates of ectopic pregnancy deaths relative to non-ban states.
    • Determining whether this relationship is causal requires isolating specific institutional mechanisms: evaluating whether criminal penalties and statutory ambiguity cause emergency clinicians to delay standard-of-care surgical interventions (e.g., salpingostomy or methotrexate administration) until life-threatening hemorrhage occurs.