Causality. Why you shouldn't use Bradford Hill criteria!

Global Health with Greg Martin
27 Oct 202107:08
EducationalLearning
32 Likes 10 Comments

TLDRThe video discusses the concept of causation and the limitations of Bradford Hill criteria in establishing causal relationships. It argues that while these criteria seem logical, they essentially restate the observed correlation without proving causation. The speaker emphasizes the importance of excluding chance, bias, confounding, reverse causation, and fraud to confidently assert a causal relationship. The video also highlights the University of Limerick's public health master's program, praising its practical approach to preparing graduates for the workforce.

Takeaways
  • πŸ” The discussion revolves around the concept of causation and the famous phrase 'correlation doesn't mean causation'.
  • πŸ“ˆ The Bradford Hill criteria are mentioned as a traditional method in public health to discuss causation, but the speaker expresses dissatisfaction with them.
  • πŸ‘Ž The speaker criticizes the Bradford Hill criteria for not being useful in determining the nature of the relationship between exposure and outcome.
  • πŸ”„ The criteria include strength of association, consistency, temporal sequence, dose-response, reversibility, specificity, and plausibility.
  • πŸ’‘ The speaker argues that most of the criteria restate the correlation without providing evidence of causation, except for plausibility.
  • 🚫 Plausibility only rules out relationships that are completely implausible, such as a rooster's crow causing the sun to rise.
  • πŸ€” A proper approach to understanding causation should exclude chance, bias, confounding, reverse causation, and fraud.
  • 🎯 By excluding these factors, one can strengthen the argument that a relationship is causal, provided it is plausible.
  • 🏫 The University of Limerick's Master's degree in Public Health is endorsed for its quality and practical preparation for the workplace.
  • 🌴 The example of ice cream consumption and shark attacks is used to illustrate confounding, where a third variable (hot summer days) is associated with both.
  • ⚠️ The video warns of the dangers of fraud in scientific research, where fabricated relationships can be published and accepted as true.
Q & A
  • What is the main topic of the transcript?

    -The main topic of the transcript is causation and the discussion of causality, specifically the critique of Bradford Hill criteria in public health research.

  • What does the phrase 'correlation doesn't mean causation' imply?

    -The phrase 'correlation doesn't mean causation' implies that just because two events or variables are related or occur together, it does not necessarily mean that one causes the other.

  • What are the Bradford Hill criteria mentioned in the transcript?

    -The Bradford Hill criteria are a set of guidelines used to help determine whether an observed correlation implies causation. They include strength of association, consistency, temporal sequence, dose-response, reversibility, specificity, and plausibility.

  • Why does the speaker express dislike for the Bradford Hill criteria?

    -The speaker dislikes the Bradford Hill criteria because they believe that most of the criteria simply restate the observed correlation without providing evidence about the nature of the relationship, and plausibility only indicates that a relationship is possible, not necessarily causal.

  • What are the alternatives to causation that the speaker mentions?

    -The alternatives to causation mentioned by the speaker are chance, bias, confounding, reverse causation, and fraud.

  • How can we strengthen the argument for causation between an exposure and an outcome?

    -We can strengthen the argument for causation by excluding chance, bias, confounding, reverse causation, and fraud, and if the relationship is plausible, we can conclude that the relationship is likely due to causation.

  • What is an example of confounding provided in the transcript?

    -An example of confounding provided is the correlation between ice cream consumption and shark attacks, which both increase during hot summer days. However, it is not the ice cream consumption that causes shark attacks; rather, they are both associated with the third variable, hot weather.

  • What is the role of plausibility in determining causation?

    -Plausibility is useful in determining causation when the counterfactual seems to be the case. It indicates that the relationship is possible, but it is not a strong indicator of causation on its own.

  • What is the importance of excluding bias when assessing causation?

    -Excluding bias is crucial because bias can systematically distort the results of a study, leading to incorrect conclusions about the relationship between exposures and outcomes.

  • How does reverse causation differ from the expected direction of causation?

    -Reverse causation occurs when the assumed direction of causation is incorrect; instead of exposure A causing outcome B, it is actually outcome B that causes exposure A.

  • What is the significance of the University of Limerick's Master's degree in Public Health mentioned in the transcript?

    -The University of Limerick's Master's degree in Public Health is highlighted as outstanding and designed to prepare graduates for the workplace, ensuring they are ready to contribute effectively in the public health sector.

Outlines
00:00
πŸ” Understanding Causality and the Critique of Bradford Hill Criteria

This paragraph introduces the concept of causality and the common saying that 'correlation does not imply causation.' It discusses the Bradford Hill criteria, which are often used in public health to assess causation, but the speaker expresses dissatisfaction with these criteria. The speaker argues that the criteria merely restate the observed correlation without providing evidence about the nature of the relationship. The paragraph emphasizes the need to exclude chance, bias, confounding, reverse causation, and fraud to establish a causal relationship, and suggests that plausibility alone is not sufficient to prove causation.

05:00
🧐 Exploring Alternatives to Bradford Hill Criteria for Causality Assessment

The second paragraph delves into the limitations of the Bradford Hill criteria by discussing alternative explanations for observed relationships, such as chance, bias, confounding, reverse causation, and fraud. It provides examples to illustrate these concepts, like the correlation between ice cream consumption and shark attacks being due to a confounding factor (hot summer days). The speaker suggests that by excluding these alternatives, one can strengthen the argument for a causal relationship, provided it is plausible. The paragraph concludes with a brief mention of the importance of identifying a mechanism for the proposed causal relationship.

Mindmap
Keywords
πŸ’‘Causation
Causation refers to the relationship between cause and effect, where one event (the cause) directly results in the occurrence of another event (the effect). In the video, causation is central to the discussion, as the speaker aims to explore how certain exposures can lead to specific outcomes in public health. The speaker critiques the Bradford Hill criteria for not adequately distinguishing causal relationships from mere correlations, emphasizing the importance of understanding the true nature of these relationships for public health research.
πŸ’‘Correlation
Correlation is a statistical measure that describes the extent to which two variables change together, but it does not necessarily imply that one causes the other. The phrase 'correlation does not mean causation' is highlighted in the video to caution against confusing these two concepts. The speaker argues that many of the Bradford Hill criteria merely restate correlations rather than providing evidence of causation.
πŸ’‘Bradford Hill Criteria
The Bradford Hill Criteria are a group of principles that can be useful for establishing a causal relationship between a specific exposure and an outcome. The video briefly discusses these criteria, which include strength of association, consistency, temporal sequence, dose-response, reversibility, specificity, and plausibility. However, the speaker expresses dissatisfaction with these criteria, arguing that they mostly reiterate observed correlations without proving causality.
πŸ’‘Plausibility
Plausibility refers to whether a proposed relationship between two variables is logically conceivable or believable within the current understanding of the subject. In the video, plausibility is discussed as one of the Bradford Hill criteria, but it is critiqued for only indicating that a relationship is possible, not necessarily causal. The speaker points out that while plausibility is necessary, it is not a strong indicator of causation.
πŸ’‘Chance
Chance is mentioned in the video as one of the alternative explanations for observed relationships between variables. It refers to the possibility that the observed correlation is a random occurrence, not indicative of a genuine link between exposure and outcome. The speaker emphasizes the importance of excluding chance to strengthen the argument for causation.
πŸ’‘Bias
Bias in the context of the video refers to systematic errors in study design or data collection that can lead to incorrect conclusions about relationships between variables. The speaker distinguishes bias from chance by noting that bias involves consistent errors, while chance involves random errors. Excluding bias is crucial for establishing a causal relationship.
πŸ’‘Confounding
Confounding is explained in the video as a situation where the observed relationship between two variables (A and B) is actually influenced by a third variable (C) that affects both A and B. This third variable is not on the causal pathway between A and B but may create the illusion of a causal relationship. Identifying and adjusting for confounders is vital for accurate causality assessment.
πŸ’‘Reverse Causation
Reverse causation is a scenario where the presumed cause and effect are actually reversed, meaning that what is thought to be the effect is actually the cause. The video highlights the importance of considering this possibility when examining relationships between variables to ensure that conclusions about causality are accurate.
πŸ’‘Fraud
Fraud is mentioned in the video as a potential explanation for observed correlations between variables, where the data or results have been intentionally fabricated or manipulated. The speaker notes that excluding fraud is a step in affirming the integrity of the data and the validity of the causal relationship.
πŸ’‘Public Health
Public Health is the overarching theme of the video, with a focus on understanding how exposures lead to health outcomes. The discussion revolves around applying principles of causation and methodology in public health research to accurately identify and mitigate health risks. The speaker also promotes a public health program at the University of Limerick, emphasizing its role in preparing students for impactful work in this field.
Highlights

Today's discussion revolves around causation, causality, and the distinction between correlation and causation.

The Bradford Hill criteria are introduced as a traditional method in public health to discuss causation.

The speaker expresses a dislike for the Bradford Hill criteria and promises to explain why.

Strength of association is described as a strong correlation between the exposure and outcome of interest.

Consistency of finding implies that the same correlation is observed across multiple studies.

Temporal sequence is crucial, where the cause must precede the effect.

Dose-response and reversibility are two sides of the same coin, indicating a relationship's intensity and potential reversal.

Specificity suggests that the exposure is uniquely related to the outcome, rather than being connected to many outcomes.

Plausibility refers to the possibility of the relationship being causal, but it is not a strong indicator of causation.

The speaker argues that the Bradford Hill criteria essentially restate the correlation without providing evidence of causation.

The proper approach to understanding causality involves excluding chance, bias, confounding, reverse causation, and fraud.

By excluding alternative explanations, one can strengthen the argument for a causal relationship.

Chance can lead to observing relationships that do not exist due to sampling error.

Bias can result from flawed study design or incorrect measurement of data.

Confounding occurs when a third variable is associated with both the exposure and outcome, creating a false impression of causation.

Reverse causation is the incorrect assumption of the direction of causality.

Fraud in scientific research can lead to false causal relationships.

The University of Limerick's Master's degree in Public Health is recommended for its practical focus and world-class faculty.

Transcripts
Rate This

5.0 / 5 (0 votes)

Thanks for rating: