Causality in Scientific Method: Theories, Models, and Applications

Foundations of Causal Reasoning and Human Development

• Innate Human Drive for Causal Discovery:

  • Young infants demonstrate a spontaneous interest in learning how they can influence their environment.
  • This exploratory behavior requires no formal instruction; it is a natural developmental process.
  • Through cumulative interactions with their surroundings, infants acquire structured knowledge about how the world functions.
  • By the age of $6$ to $7$ months, this basic comprehension of causal relationships is distinctly present, facilitated by the experiences accumulated during the preceding months.

• Parallels Between Children and Scientists:

  • Humans are predisposed from early childhood to identify environmental patterns, manipulate variables, and trigger events to expand their understanding of the world.
  • In this respect, children function like scientists, and scientists function like children: both are fundamentally driven to uncover patterns.

• Causality as 'The Cement of the Universe':

  • Causal concepts provide the primary framework through which both developing children and scientists explain observed patterns.
  • Establishing whether causality exists—and tracing the precise mechanisms through which causal influences occur—allows for a deeper understanding of worldly events.
  • Causality is frequently characterized as "the cement of the universe" because of its unique capacity to link otherwise disjointed events in human experience.
  • Because of this explanatory power, scientific inquiry is largely oriented toward constructing and testing causal explanations by identifying the specific event or chain of events responsible for an observed phenomenon.

Characteristics and Practical Significance of Causal Reasoning

• Three Central Characteristics of Causal Reasoning:

  • Learning through Timing, Location, and Frequency: Causal relationships are typically derived from spatiotemporal patterns and occurrence rates. A correlation in time or space frequently signals an underlying causal connection (e.g., feeling fatigued shortly after eating lunch suggests lunch may be the cause). Likewise, observing shifts in the frequency of an event directs attention to other concurrent changes, constituting correlation in the frequency of two outcomes.
  • Testing Hypotheses through Active Intervention: Determining causation usually requires actively intervening in the world. Systematically varying a suspected causal factor while holding all other factors constant provides substantially deeper insight than passive observation of correlational frequencies. For instance, to test whether wastewater injection from hydraulic fracturing operations causes earthquakes, operators can perform fracturing without injecting wastewater into disposal wells; a significant drop in regional seismic activity indicates wastewater injection is primarily responsible, whereas no change directs suspicion toward other factors, such as the fracturing process itself.
  • High Practical Significance for Control and Prediction: Understanding causes allows humans to bring about desirable events and prevent catastrophic ones.

• Applied Domains of Causal Reasoning:

  • Assessing the real-world outcomes of macroeconomic policies.
  • Diagnosing underlying medical conditions and deducing physiological responses.
  • Determining legal liability and corporate or individual accountability.

• Mechanistic Insight versus Intervention Handles:

  • Process Insight: Identifying a causal mechanism reveals the underlying dynamics of a phenomenon (e.g., understanding how negative emotional states prompt alcohol consumption).
  • Targets for Intervention: Causal mechanisms provide clear leverage points for actionable intervention (e.g., teaching individuals prone to heavy drinking alternative psychological coping mechanisms for managing negative affect).

• The Evolutionary Flip Side of Pattern Recognition:

  • Evolutionary pressures have optimized human cognitive faculties to detect patterns with high sensitivity.
  • This intense pattern-seeking drive can cause people to perceive causal connections where none exist.
  • In scientific communication and popular media, this cognitive bias frequently leads to premature or invalid causal inferences drawn from purely correlational findings.

Skepticism, Perception, and Classification of Causal Claims

• Humean Skepticism (David Hume, 1711–1776):

  • The Scottish philosopher David Hume investigated what sensory experience genuinely reveals regarding the nature of causal relations.
  • Hume concluded that experience reveals far less than is commonly assumed: while humans regularly perceive a constant conjunction between two events and their correlation across space and time, sensory perception cannot detect an intrinsic "causal power" or "necessary connection."
  • Hume maintained that experience demonstrates only that events follow one another in a regular sequence, casting doubt on whether any independent causal property exists in the physical world beyond continuous association.

• Perception of Causality (Albert Michotte, 1881–1965):

  • In the 1940s, Belgian psychologist Albert Michotte experimentally demonstrated that human perception of causal relations is robust, automatic, and largely reliable.
  • Michotte showed that when visual stimuli move in specific spatiotemporal configurations, observers cannot help but perceive direct causation rather than mere independent movements.
  • Spatiotemporal Contiguity: When two events occur at the exact same location and immediately consecutive moments, human observers perceive them as causally connected without requiring repeated prior exposure.
  • Spatiotemporal Gaps: The introduction of a spatial or temporal gap between events drastically diminishes the perceptual intuition that one caused the other.

• The Post Hoc, Ergo Propter Hoc Fallacy:

  • Translates from Latin as "after this, therefore because of this."
  • Refers to the logical error of inferring direct causality solely from spatiotemporal sequence.
  • Although spatiotemporal cues are critical components of human causal perception, they are not infallible and frequently lead to erroneous inferences.

• Proximate versus Distal Causes:

  • Proximate Causes (Nabije Oorzaken): Causal factors positioned close in time and physical location to the resulting event.
  • Distal Causes (Distale Oorzaken): Causal factors situated further back in time or at a greater distance from the eventual outcome.

• Taxonomy of Causal Statements:

  • Event-to-Event in a Specific Situation: Yesterday's news reports concerning an impending lockdown ($A$) caused panic buying in the Van Woerkom brothers' supermarket ($B$).
  • Event-to-Event at a Generic Level: Having smoked ($C$) causes lung cancer ($D$).
  • Variable-to-Variable in a Specific Situation: The amount of time I spend studying ($E$) affects how well I perform on my exam ($F$).
  • Variable-to-Variable at a Generic Level: The degree to which an individual experiences negative moods throughout the day ($G$) affects the volume of alcohol consumed that evening ($H$).
  • Psychological research focuses predominantly on variable-to-variable relationships evaluated at the generic level.

• Causality and Causal Chains:

  • Causality denotes an interrelated sequence (causal chain) of events or variables where the presence, absence, or magnitude of one factor (partially) determines the occurrence or value of a subsequent factor.
  • Causal chains can extend indefinitely, expanding from simple dyads ($A \rightarrow B$) to complex multi-step progressions ($A \rightarrow B \rightarrow C \rightarrow D$).
  • Individual events regularly possess multiple contributing causes across proximate and distal tiers.

• Causal Inference at Individual versus Population Levels:

  • Causal inference focuses on identifying causal relationships that hold true for individual entities.
  • Population-level causal statements represent the aggregate average of individual-specific causal relationships operating within that population.

Causality versus Correlation and the Counterfactual Framework

• The Counterfactual Framework:

  • Determining whether an empirical effect is causal requires comparing the factual outcome with a counterfactual outcome.
  • A counterfactual statement evaluates an unobserved scenario that is contrary to actual fact, typically signaled by conditional verbs such as "would."
  • In evaluating smoking and lung cancer, one must compare the actual probability of smokers developing lung cancer, $P(\text{lung cancer} \mid \text{smoking})$, against the counterfactual probability that these exact same individuals would have developed lung cancer had they not smoked, $P(\text{lung cancer} \mid \text{not smoking})$.
  • The core challenge of causal inference is that counterfactual states are fundamentally unobservable in the real world.

• Definition and Properties of Correlation:

  • A correlation (or statistical association) exists when event $P$ occurs with a different frequency in the presence of event $Q$ than in its absence ($\neg Q$).

• Correlation indicates statistical dependency, which by itself does not establish causality because it evaluates only factual observations without accounting for counterfactual outcomes.

• Symmetry versus Asymmetry: Correlation is symmetric, meaning $\text{Corr}(X, Y) = \text{Corr}(Y, X)$. Causality is strictly asymmetric ($X \rightarrow Y \not\equiv Y \rightarrow X$). For example, while serious illness and death correlate symmetrically, severe illness causes death, whereas death does not cause severe illness.

• Common Causes (Gemeenschappelijke Oorzaken):

  • A correlation between two variables can be generated entirely by an unmeasured third variable acting as a common cause.
  • Ice cream sales and homicide rates are positively correlated, yet neither causes the other; higher ambient summer temperatures independently increase both ice cream consumption and outdoor interpersonal conflicts.

• Spurious Correlations (Valse Correlaties):

  • Statistically significant correlations can arise by coincidence without any meaningful causal, mechanistic, or logical link.
  • US Dairy Association data on per capita cheese consumption correlates strongly with Centers for Disease Control data on accidental strangulation deaths caused by bedsheet entanglement, despite zero causal connection.

• Causation in the Absence of Statistical Correlation:

  • Two variables may be causally connected even if their net statistical correlation is zero, which occurs when opposing causal mechanisms cancel one another out.
  • Pregnancy is an established cause of thrombosis.
  • Oral contraceptive pills prevent pregnancy, thereby indirectly reducing thrombosis risk.
  • Simultaneously, oral contraceptives independently elevate thrombosis risk via direct physiological pathways.
  • If the preventative indirect effect exactly balances the direct risk-elevating effect, the net statistical correlation between oral contraceptive usage and thrombosis becomes zero despite active causal links.

Theoretical Conceptions of Causality: Difference-Making and Physical Processes

• Dual Conceptions in the Philosophy of Science:

  • In scientific philosophy, causality lacks a single universal definition; Potochnik et al. distinguish two primary frameworks that move beyond simple correlation:

• The Difference-Making Account:

  • Defines causal relations as scenarios where the occurrence or value of one event makes a measurable difference to the occurrence or value of another.
  • Billiard Analogy: If a cue ball strikes the 8-ball, the 8-ball moves; had the cue ball missed, the 8-ball would have remained stationary. Striking the 8-ball at a different angle or velocity alters its trajectory and speed accordingly. Thus, the cue ball makes a difference to the 8-ball's movement.
  • Methodological Role in Psychology: Difference-making underlies experimental design: researchers manipulate an independent variable to observe whether it produces a differential change in a dependent variable.
  • Manipulability: An authentic causal factor can be manipulated to produce a corresponding shift in an outcome measure. Manipulating a variable that is merely correlated with an outcome produces no change in that outcome.
  • Scope: Difference-making identifies that an antecedent factor alters an outcome, without having to specify the physical mechanism through which that change is realized.

• The Physical Process Account:

  • Asserts that genuine causality requires a continuous, uninterrupted physical process—such as a transfer of energy or momentum—connecting cause and effect.
  • Billiard Analogy: When the cue ball hits the 8-ball, a portion of its kinetic energy is physically transferred across the point of contact, driving the 8-ball forward. This physical energy transmission constitutes the underlying causal mechanism.
  • Applications and Limitations: While straightforward in physical sciences, identifying continuous physical energy transmissions is substantially more challenging in higher-level disciplines such as psychology, economics, or sociology.

• Theoretical Interplay:

  • Difference-making and physical process theories can function as complementary frameworks, competing models, or distinct tools whose suitability varies by scientific context.

Interventions, Experiments, and Causal Modeling

• Empirical Strategies for Isolating Difference-Makers:

  • Controlled Experimentation: Running an experiment—ideally a double-blind randomized controlled trial—serves as the primary standard for uncovering difference-making relations.
  • Causal Modeling: When experimental manipulation is unethical or impractical, scientists build structured causal models and apply statistical controls to observational data to infer causal connections.

• Objectives in Causal Testing:

  • Determining whether a causal link exists between two targets.
  • Characterizing the nature, parameters, and direction of the relationship.
  • Measuring the relative effect size or strength of an established causal link.
  • Pharmacological Examples: Clinical drug testing evaluates safety (verifying the drug does not cause adverse outcomes), efficacy (confirming the drug causes the intended therapeutic improvement), and relative potency (testing whether the compound outperforms existing treatments on the market).

• The Methodological Role of Interventions:

  • External interventions intentionally disrupt the normal functioning of a system to isolate specific causal paths.
  • An intervention assigns the value of the suspected cause independently, preventing confounding influences from the broader causal background from dictating that value—hence the term "independent variable."
  • Standardized control groups minimize the likelihood that changes observed in the dependent variable originate from extraneous background conditions rather than the intervention itself.

Hume's Regularity Theory and Its Limitations

• Regularity Theories of Causality:

  • Both regularity theories and interventionist approaches share an underlying commitment to difference-making principles.
  • The regularity framework asserts that causal connections represent a specific category of empirical regularity, verifiable by checking whether an association satisfies fixed criteria.

• Hume's Three Necessary Conditions for Causation:

  • Spatiotemporal Proximity (Spatiotemporele Nabijheid): Cause $X$ and effect $Y$ must be adjacent in both space and time (e.g., direct skin contact with a stinging nettle followed immediately by a localized rash).
  • Temporal Priority (Temporele Prioriteit): Cause $X$ must strictly precede effect $Y$ in time ($X$ occurs first, followed by $Y$). This condition highlights the intrinsic temporal asymmetry of causality.
  • Constant Conjunction (Constante Associatie): Every occurrence of $X$ must invariably be accompanied by $Y$ (every instance of falling into stinging nettles produces a rash).

• Advantages of Hume's Regularity Approach:

  • Avoids metaphysical assumptions, aligning closely with logical positivism.
  • Provides concrete, testable conditions for identifying cause-and-effect relations.
  • Simplifies causal investigation by reducing it to the identification of observable regularities that satisfy the three criteria.

• Critical Limitations of Hume's Approach:

  • Absence of Invariable Association in Real-World Scenarios: Most accepted causal relationships do not exhibit invariable constant conjunction. For example, smoking causes lung cancer, yet many chronic smokers never develop the disease; under Hume's strict criteria, smoking could not be classified as a cause.
  • Confounding by Common Causes: Spatiotemporally contiguous regularities that show constant conjunction can arise from a shared common cause without any direct causal link between $X$ and $Y$.

Necessary Causes, Sufficient Causes, and the Causal Background

• Sufficient Causes (Voldoende Oorzaken):

  • A factor whose presence guarantees the occurrence of the effect.
  • If $X$ is a sufficient cause of $Y$, then whenever $X$ occurs, $Y$ must follow ($X \implies Y$).
  • Example: Being struck directly by lightning ($X$) is a sufficient cause of death ($Y$).
  • Asymmetry of Sufficiency: The occurrence of $Y$ does not imply that $X$ occurred; an individual can die from numerous causes other than lightning strikes ($Y \not\implies X$).
  • Hume's requirement of constant conjunction treats cause $X$ as an absolute sufficient cause for $Y$.

• Necessary Causes (Noodzakelijke Oorzaken):

  • A condition that must be present for the effect to occur at all.
  • If $X$ is a necessary cause of $Y$, the effect cannot happen without it ($\neg X \implies \neg Y$, or equivalently $Y \implies X$).

• Example: The presence of oxygen ($X$) is a necessary cause for the ignition and continuation of fire ($Y$). Fire cannot exist without oxygen, meaning fire implies oxygen was present.

• Asymmetry of Necessity: The presence of oxygen does not imply the presence of fire; oxygen exists ubiquitously without combustion occurring ($X \not\implies Y$).

  • A factor can be a necessary cause without being sufficient, or a sufficient cause without being necessary.

• The Causal Background (Causal Field):

  • Few causal factors operate as unconditional necessary or sufficient causes; their efficacy depends entirely on a stable causal background.
  • Definition: The causal background comprises all ambient contextual factors that exert a genuine or potential influence on the target events and their interaction.
  • Example: Maintaining excellent personal dental hygiene can be viewed as a sufficient cause of good long-term oral health, but this holds true only given specific background conditions: visiting a properly qualified dentist, avoiding severe accidents that knock out teeth, and lacking rare genetic enamel defects.
  • Causal assertions routinely leave the causal background implicit, yet background stability is essential for causal relations to unfold as predicted.

Probabilistic Causality and Contributing Causes

• The Shift to Probabilistic Models:

  • Strict constant conjunction, absolute sufficiency, and strict necessity are rarely encountered outside ideal physical laws.
  • Probabilistic causality resolves this limitation by evaluating factors that alter the likelihood of an outcome rather than strictly determining it.
  • This approach serves as a modern correction to classical regularity theories.

• Contributing or Partial Causes (Contribuerende of Partiële Oorzaken):

  • Definition: A contributing cause is any factor that increases the probability of an outcome occurring, while being neither strictly necessary nor sufficient on its own.
  • Mathematical Formulation: For a suspected contributing cause $X$ and an outcome $Y$, the probability of $Y$ given $X$ exceeds the probability of $Y$ in the absence of $X$:     $P(Y \mid X) > P(Y \mid \neg X)$
  • Contributing causes are far more common in empirical science than absolute necessary or sufficient causes.
  • As the strength of a probabilistic causal relationship increases, it approaches the ideal limit of deterministic sufficiency.

• Advantages of the Probabilistic Framework:

  • Reflects actual empirical scientific findings across biomedical and behavioral sciences.
  • Accommodates the reality that single variables rarely dictate human outcomes due to extensive variations in genetic composition, biological susceptibility, and environmental exposures.

• Disadvantages of the Probabilistic Framework:

  • Probabilistic dependency relies on statistical correlation, which is inherently symmetric.
  • Because $P(Y \mid X) > P(Y \mid \neg X)$ mathematically mirrors a symmetric correlational structure, probability calculations alone cannot distinguish the cause from the effect without supplemental information regarding temporal sequence or physical intervention.

Questions & Discussion

• Question 7.7.5:
  • Prompt: Provide a definition of a contributing cause in your own words.
  • Response: A contributing cause (also known as a partial cause) is an antecedent factor that increases the probability of an event occurring, despite being neither strictly necessary nor sufficient to produce that effect. In formal probabilistic terms, for an event $Y$ and a contributing factor $X$, the probability of the outcome is greater when the factor is present than when it is absent: $P(Y \mid X) > P(Y \mid \neg X)$.