Correlation, Causation, and Methodological Bias in Research
Etymology and Fundamental Concepts of Correlation and Causation
Etymological Breakdown:
Correlation: Derived from "co-relation" (removing one 'r' reveals "co relation"), defined as two variables that possess a statistical relationship or adapt and change together across time.
Coincidence: Derived from "co-incidence", describing two distinct incidents or events that occur at the exact same time ("co-incident") without an underlying structural or causal link.
Causation: Derived from "cause", representing the underlying reason or direct mechanism that produces a specific consequence or outcome.
Core Distinction Between Correlation and Causation:
Correlation denotes that two variables adapt or change in sync, appearing as if one happens as a result of the other.
Causation establishes that changes in one variable directly create the observed outcome in the second variable.
Establishing a correlation between variables does not prove causation, as co-occurring data points may stem from coincidences, bidirectional influences, or unmeasured third variables.
Types of Statistical Correlation
Positive Correlation:
Defined as a relationship in which two variables move in the same direction; as one variable increases, the other increases, or as one decreases, the other decreases.
Example: The relationship between trait extraversion and the enjoyment of large social situations. As self-reported trait extraversion increases, statistical enjoyment of densely populated social settings simultaneously rises.
Negative Correlation:
Defined as an inverse relationship in which two variables move in opposite directions; as one variable increases, the other decreases.
Example: The relationship between trait introversion and the enjoyment of large social situations. Individuals with high self-reported introversion (who prefer smaller groups or remaining reserved) demonstrate decreasing levels of enjoyment as social group sizes increase.
No Correlation:
Occurs when there is no statistical overlap or synchronization between the variability of two variables. Changes in one variable do not systematically correspond to changes in the other.
Spurious Correlation:
Defined as a mathematical correlation between two variables that is statistically high but functionally non-existent, fake, or erroneous in reality.
Robberies in Alaska vs. US Professor Salaries:
Demonstrates a high positive correlation coefficient of , indicating that data lines up over of the time despite having no logical or causal connection in reality.
Distance Between Neptune and the Sun vs. Popularity of the First Name "Andrea":
Demonstrates a correlation coefficient of . As Neptune orbits the Sun and its distance increases and decreases, it matches the popularity trends of the name "Andrea", illustrating how automated systems or AI can fabricate absurd causal explanations for purely coincidental statistical alignments.
Case Study on Spurious Correlation: Youth Mental Health and Smartphone Use
Context of the Case Study:
Analysis from the research podcast If Books Could Kill (hosted by a journalist and a lawyer evaluating popular pop-science books).
Focuses on the book The Anxious Generation by Jonathan Haidt, which asserts that smartphone adoption and social media access directly caused an epidemic of mental illness among adolescents.
Jonathan Haidt's Core Argument and Claimed Statistics:
Mental health indicators for the Millennial generation (born between and ) remained stable or slightly better than Generation X throughout the and .
A sharp uptick in adolescent depression, anxiety, and suicidality began around –.
Self-Harm Hospitalization Metrics (United States):
Older adolescent girls experienced a to increase in self-harm hospitalizations.
Younger adolescent girls (ages to ) experienced a increase in self-harm hospitalizations.
Self-Harm Hospitalization Metrics (Britain):
Girls aged to exhibited a increase in self-harm incidents.
Haidt concluded that smartphone connectivity led girls to share ideas of self-harm and anxiety, relying on hospitalization data as an absolute metric comparable to homicide rates in crime statistics.
Methodological Flaws and Uncontrolled Third Variables:
Healthcare Expansion via the Affordable Care Act (Obamacare):
Uninsured rates among children dropped from in down to by 2016$.\n * Extended Medicaid coverage and expanded private insurance dramatically increased access to mental health admissions, clinical diagnoses, and psychiatric prescriptions around 2010$.
Clinical Diagnostic Screening Mandates:
In , the US Preventive Services Task Force issued guidance recommending mandatory depression screening for adolescent girls, requiring insurance companies to cover these evaluations.
Medical Record Documentation Guidelines:
New diagnostic protocols instructed physicians to explicitly list suicidal ideation as a primary cause of harm on standardized medical records.
Timeline and Age Inconsistencies:
Historical clinical data demonstrates that adolescent mental health distress was already trending upwards prior to 2010$.\n * Children aged 101420102012 window; smartphone adoption in that age demographic occurred significantly later.\n * **Conclusion**:\n * The rise in reported self-harm statistics was heavily driven by changes in administrative documentation, insurance access, and screening mandates rather than a simple causal effect of smartphone usage, rendering Haidt's core claim a spurious correlation.\n\n# Methodological Control and Individual Differences\n\n* **The Necessity of Variable Control**:\n * Human experience is inherently complex; isolating a true causal relationship requires researchers to identify, record, and statistically control for confounding internal and external factors.\n* **Individual Differences**:\n * **Demographic Variables**: Data including age, gender, race, national origin, and socioeconomic status. Collecting demographic parameters enables researchers to isolate group-specific variance.\n * **Intersectional Identity Groups**: Analyzing overlapping demographic traits to understand how combined social identities influence human behavior and societal interactions.\n* **State vs. Trait Differences**:\n * **State**: Temporary physical or psychological conditions experienced at a specific moment in time (e.g., a participant taking a survey during a stressful day will report heightened pessimism).\n * **Trait**: Stable, enduring individual traits and personality characteristics across time (e.g., trait extraversion or introversion).\n* **External Noise**:\n * Environmental and procedural distractions that impact data validity, such as ambient room noise during testing, ambiguous question phrasing (as modeled in the Bullseye framework), or the artificial nature of the research setting.\n\n# Directionality and Experimental Biases in Research\n\n* **The Temporal and Directionality Requirements**:\n * **Temporal Precedence**: To establish that Variable A causes Variable B, Variable A must occur prior to Variable B in time.\n * **Directionality Determination**: Researchers must verify whether Variable A drives Variable B (A \rightarrow BB \rightarrow A).\n* **Empirical Research Example: Reading and Transportability**:\n * **Initial Hypothesis**: Childhood pleasure reading increases narrative transportability (the cognitive ability to be immersed in narrative worlds across media such as film).\n * **Theoretical Framework**:\n * *Trait Fantasy*: Hypothesized as a mediating factor representing the ability to generate mental imagery ("mental movies").\n * *Perspective-Taking*: Hypothesized as a moderating factor between book reading and trait fantasy, enhanced through reading literature.\n * **Statistical Findings**: Pleasure book reading positively correlated with narrative transportability, yielding a statistically significant result with p < 0.0010.1\%$$ probability of occurring by random chance).
Reverse Directionality Fallacy: High baseline transportability (an inherent capacity for narrative immersion) may cause individuals to read more books, reversing the assumed causal vector.
Cognitive and Research Biases:
Confirmation Bias:
The cognitive propensity to seek out, select, and interpret information in a manner that confirms prior hypotheses or pre-existing beliefs while disregarding contradictory data.
Affects search term selection during information gathering and leads individuals to dismiss opposing evidence as biased or obsolete.
Selection Bias:
Occurs when a research sample is non-representative of the general population.
Landline Telephone Polling: Skews sample populations toward older demographics because younger populations primarily rely on mobile devices.
Opt-In Polling: Attracts individuals who are disproportionately vocal, assertive, or extroverted.
Academic Sampling Constraints: Heavy reliance on college freshmen and sophomores creates sample populations skewed heavily toward younger adults.
Hawthorne Effect:
A phenomenon where research participants alter their natural behavior due to the awareness of being observed or measured.
Example: Measuring physiological parameters (such as heart rate and eye tracking via attached electrodes) while participants watch television alters their stress responses and natural viewing experiences.