18:43
PSYC1040 – Week 2A: From Claims to Causal Inference
SocialNeuro
Overview
This video explains the complexities of establishing causal claims in psychology, moving beyond the simple "correlation does not imply causation." It introduces the counterfactual definition of causality: Y would have been different if X had not occurred. The video details four major threats to causal inference – confounding, selection effects, reverse causation, and history/maturation – illustrating each with examples. It then applies these concepts to the debate surrounding smartphones and adolescent mental health, highlighting how strong causal claims require robust research designs that can approximate the unobservable counterfactual, emphasizing that causality is earned through design, not simply discovered.
How was this?
Save this permanently with flashcards, quizzes, and AI chat
Chapters
- Psychology frequently implies or states causal relationships, which are powerful but difficult to justify.
- The common phrase 'correlation does not imply causation' is true but incomplete; association allows for multiple causal explanations (X causes Y, Y causes X, or a third variable causes both).
- True causal inference requires understanding what would have happened if the cause had not occurred (a counterfactual), which is inherently unobservable.
- Formal research designs are necessary to approximate this unobservable counterfactual and overcome cognitive biases that favor simple, visible causes.
Understanding the difficulty of causal claims helps you critically evaluate psychological research and avoid oversimplifying complex relationships.
Observing that X and Y are related doesn't tell us if X causes Y, Y causes X, or a third variable causes both.
- Confounding occurs when a third variable is related to both the presumed cause (X) and the outcome (Y), making it impossible to isolate the effect of X.
- This threat is embedded in the study's design, not just something identified after data collection.
- Statistical adjustments can help but cannot guarantee all confounders are accounted for.
Recognizing confounding prevents you from mistakenly attributing an effect to a cause when an unmeasured factor is actually responsible.
The apparent link between eating ice cream and sunburn is confounded by temperature and sun exposure, which influence both ice cream consumption and sunburn risk.
- Selection effects arise when participants are not randomly assigned to conditions, leading to pre-existing differences between groups.
- These differences, related to the outcome, bias comparisons, making it unclear if observed effects are due to the treatment or who joined the group.
- The core issue is that the groups being compared are not equivalent at the outset, violating the counterfactual requirement.
Understanding selection effects helps you question findings where groups naturally differ, such as in studies of voluntary programs or self-selected samples.
Students attending optional exam workshops performing worse might be due to anxious students self-selecting into the workshops, not the workshops themselves being detrimental.
- Reverse causation occurs when the presumed effect (Y) actually causes the presumed cause (X).
- This is common in observational research and violates the intuitive forward-in-time causal narrative.
- Without clear temporal precedence, the direction of causality remains ambiguous.
Being aware of reverse causation prevents you from assuming the most obvious temporal order is the correct causal one.
Higher loneliness reported by people with AI companions could mean AI use causes loneliness, or that lonely people seek out AI companions.
- History effects refer to external events occurring concurrently with a study that could influence the outcome.
- Maturation effects refer to natural changes over time within participants (e.g., growing older, adapting).
- When these changes coincide with a treatment, it's difficult to determine if the treatment or these other factors caused the observed outcome.
This highlights that changes observed over time might be due to broader societal shifts or natural development, not just the intervention being studied.
A decline in student stress after a university program might be due to the program, natural adjustment to university life (maturation), or external campus policy changes (history).
- Jonathan Haidt's claim that smartphones caused increased adolescent mental health issues is a strong causal assertion.
- Evidence often cited includes parallel time-series data of rising smartphone use and rising mental health problems.
- The counterfactual is missing because we cannot observe the same adolescents with and without smartphones at the same time; history is changing concurrently.
- Plausible alternative explanations (academic pressure, economic uncertainty, changing diagnostic practices) weaken the unique support for the smartphone-causation claim.
- Intervention claims (reducing smartphone use will improve mental health) require even stronger evidence, like experimental data, which often shows small or variable effects.
This case study demonstrates how strong causal claims, even when supported by seemingly compelling data, can be undermined by weak research designs and plausible alternative explanations.
The observed correlation between increased smartphone adoption and increased adolescent anxiety does not definitively prove smartphones caused the anxiety, as other societal changes occurred simultaneously.
Key takeaways
- Causal claims are the strongest type of assertion in science and require rigorous justification beyond mere association.
- The fundamental challenge in causal inference is the unobservability of the counterfactual – what would have happened without the cause.
- Confounding, selection effects, reverse causation, and history/maturation are critical threats that must be addressed by research design.
- Research design is paramount; even large datasets cannot compensate for fundamental flaws in how a study approximates the counterfactual.
- Correlation is compatible with multiple causal stories, and ruling out alternatives requires careful, deliberate design.
- Disagreements in science often stem from differing thresholds for accepting causal evidence, not necessarily bad faith or ignorance.
- Causality is not discovered; it is earned through robust research designs that systematically address threats to inference.
Key terms
Causal inferenceCorrelation vs. CausationCounterfactualConfoundingSelection effectsReverse causationHistory effectsMaturation effectsResearch designAssociation
Test your understanding
- What is the core difficulty in establishing causality, according to the counterfactual definition?
- How does confounding threaten causal inference, and what is an example of a confounding variable?
- Explain the difference between confounding and selection effects in the context of causal claims.
- Why is reverse causation a significant threat, and what does it imply about the temporal order of variables?
- What kind of evidence would be needed to more strongly support the claim that smartphones cause adolescent mental health problems?