How can we make science more trustworthy? (with Stuart Ritchie)

Jan 19, 2023 Episode Page ↗
Overview

Stuart Ritchie, a Lecturer at King's College London, discusses making science trustworthy, examining controversies like COVID treatments (ivermectin, vitamin D) and the reliability of IQ research. He emphasizes critical thinking and understanding research limitations.

At a Glance
14 Insights
1h 26m Duration
15 Topics
8 Concepts

Deep Dive Analysis

Trust in Science and COVID-19 Pandemic Lessons

Failures in Top Medical Journals: Surgisphere Scandal

Ivermectin and Vitamin D: Case Studies in Scientific Controversy

Challenges of Controlling Variables in Observational Studies

Power of Randomized Controlled Trials and Natural Experiments

The Decline Effect in Scientific Research Findings

Revisiting Ivermectin: Advocacy, Fraud, and Nuance

Hype Cycle in Science: Power Posing and Growth Mindset Examples

Demystifying Intelligence (IQ) Research Controversies

Misconceptions About IQ Tests and Their Predictive Power

IQ, Skill Acquisition, and Compensation by Other Traits

Philosophical Interpretations of General Intelligence (G-factor)

Limitations of Current Brain-Level Understanding of Intelligence

Beyond General Intelligence: Specific Domains and Trainability

Improving Information Ecosystems: Flagging Controversial Claims

Replication Crisis

A phenomenon in science, particularly psychology, where initial strong findings often fail to be reproduced in subsequent studies, leading to questions about the reliability of published research. It highlights issues like small sample sizes and publication bias.

Confounding Factors

Variables in observational studies that are correlated with both the independent variable (e.g., low vitamin D) and the dependent variable (e.g., bad COVID cases), making it difficult to determine a direct causal link. Researchers attempt to 'control' for these statistically.

Randomized Controlled Trial (RCT)

The 'gold standard' of scientific research, where participants are randomly assigned to either a treatment group or a control group. This randomization ensures that, on average, there are no systematic differences between groups, allowing for strong causal inferences.

Natural Experiment

A research design that leverages naturally occurring events or policy changes that randomly or quasi-randomly assign individuals to different conditions, allowing researchers to study causal effects without direct experimental manipulation.

Decline Effect

The observation that the effect sizes reported in initial studies on a topic tend to be larger and more dramatic than those found in later, often more rigorous, replication studies. This can be due to publication bias, small sample sizes, and methodological improvements over time.

Indifference of the Indicator

A concept proposed by Charles Spearman, suggesting that a good measure of general intelligence (G-factor) can be extracted from a wide variety of cognitive tests, as long as a sufficient number and variety of tests are used. The specific tests in the battery don't dramatically alter the resulting G-factor.

Bonds Theory of Intelligence

Proposed by Godfrey Thompson, this theory suggests that what appears as a general factor of intelligence (G) in test data might actually arise from multiple, uncorrelated specific mental abilities. Tasks might tap into various combinations of these abilities, creating an illusion of correlation between tests.

Near vs. Far Transfer

In the context of cognitive training, 'near transfer' refers to improvements in tasks closely related to the trained task (e.g., working memory training improving other working memory tasks). 'Far transfer' refers to improvements in broader, untrained cognitive abilities (e.g., working memory training improving general intelligence), which is rarely observed.

?
How can listeners discern trustworthy science from untrustworthy science?

It's not as simple as trusting all science or relying solely on journal reputation; listeners must consider the transparency of study plans, the quality of methods, sample sizes, and the provenance of data, as exemplified by the COVID-19 pandemic.

?
Why are observational studies often contradicted by randomized controlled trials?

Observational studies struggle with confounding factors and measurement error in controls, making it difficult to isolate true causal effects, whereas randomized controlled trials inherently balance unmeasured confounders.

?
What is the current scientific consensus on the effectiveness of ivermectin and vitamin D against COVID-19?

For both ivermectin and vitamin D, larger, higher-quality randomized controlled trials generally found no significant effect against COVID-19, despite initial enthusiasm and some early studies suggesting benefits.

?
Why is IQ research often controversial, and is the common criticism that 'IQ tests only measure how well people do on IQ tests' accurate?

IQ research is controversial due to historical abuses, political biases, and cultural ideas about equality, but the claim that IQ tests only measure test-taking ability is inaccurate; IQ scores correlate with and predict many important life outcomes to a useful extent.

?
How does intelligence relate to skill acquisition and peak performance?

Higher intelligence generally means people learn skills faster and can potentially reach higher levels of peak performance in cognitively demanding tasks like chess, but other factors like conscientiousness can compensate for lower cognitive abilities.

?
What is the 'general factor' (G) of intelligence, and how is it measured?

The general factor is the common underlying ability shared across a wide variety of cognitive tests; it is measured by administering a diverse battery of tests and extracting the shared variance, which remains consistent regardless of the specific tests used.

?
Can specific cognitive abilities be trained, and does this improve overall intelligence?

Specific cognitive abilities, like working memory, can be improved through targeted training (near transfer), but there is generally a lack of evidence that such training leads to improvements in broader general intelligence (far transfer).

1. Be Skeptical of All Claims

Remember that “people make stuff up, even people you like.” Apply critical thinking and skepticism to all information, as deliberate falsehoods or self-deception can occur.

2. Evaluate Scientific Studies Critically

Don’t blindly trust scientific claims, even from top journals. Look for transparency, justified methods, and adequate sample sizes, and be wary of studies lacking these qualities.

3. Prioritize Randomized Controlled Trials

When evaluating treatments or interventions, prioritize evidence from randomized controlled trials (RCTs) as they are the gold standard for establishing causality and are less prone to confounding factors than observational studies.

4. Expect Decline in Effect Sizes

Be aware of the “decline effect,” where initial studies often show larger effects that diminish in subsequent, more rigorous research. This means early findings may overstate the true impact.

5. Avoid Overstating Claims

When advocating for a finding or treatment, avoid making outrageously strong claims, even if you believe in it. Overstating benefits can erode trust and make your position seem less credible.

6. Trace Information to Source

When encountering claims, especially controversial ones, trace them back to their original source to verify their provenance and avoid spreading made-up stories.

7. Question “Controlled For” Claims

When researchers claim to have “controlled for” variables in observational studies, ask for specifics on how they measured and controlled those factors, as measurement error and over-controlling can invalidate conclusions.

8. Map Causal Assumptions

Before drawing conclusions from data, map out the assumed causal relationships between variables using tools like directed acyclic graphs. This practice helps reveal hidden assumptions and encourages circumspection about the data’s true causal structure.

9. Identify Controversial Claims

Seek to identify when a claim is controversial, meaning many people disagree with it, rather than just accepting it as objective fact. This awareness should prompt greater skepticism and a second thought before immediately adopting the claim as true.

10. Utilize Learning Strategies

Employ effective learning strategies from cognitive psychology to learn more efficiently, regardless of your innate intelligence level. Teachers should also equip students with these tools.

11. Cultivate Conscientiousness

Develop high conscientiousness, as it can compensate for lower cognitive abilities by enabling highly organized and productive work habits, leading to success in demanding fields.

12. Focus on Specific Skill Development

Instead of solely focusing on general intelligence, identify specific domains or skills you want to improve and actively practice them. You can get better at any particular domain through effort and dedicated practice.

13. Understand IQ as Learning Potential

View IQ as a measure of potential to pick up and learn new skills faster, rather than an absolute limit. Higher IQ can accelerate skill acquisition, but lower IQ does not prevent learning, it just may take longer.

14. Broaden Social Exposure

Actively seek experiences and interactions with people across the full range of intelligence and educational backgrounds. This can help align your subjective understanding of intelligence with the broader evidence.

It does create this real issue, which is, if people say trust the science, well, which science, right? There's good science and there's bad science. And it's not like so clearly delineated, right? It's not like all bad science is published in bad science journals and all good science is published in good science journals. Like it really actually takes some nuance to tell what is what.

Stuart Ritchie

People make stuff up, even people you like. That's my, that's like my, like, people always are constantly making stuff up, even people you like. Like, that's like the bottom line for critical thinking for me.

Stuart Ritchie

The predictive ability of IQ tests exists regardless of what the interpretation of the general factor is, right? There's the predictive validity when you have these, you just have these tests, and then you just see what predictions you can make. And then trying to understand why that is the case is a different question.

Stuart Ritchie
50%
Effect size reduction in replication studies Effect sizes in replication studies are often 50% smaller than initial studies, even when replication is successful.
800-1000 people
Participants in a large power posing replication study A large replication study of power posing conducted by Spencer Greenberg's team.
40-50%
Variance explained by the general factor of intelligence The general factor of intelligence explains somewhere between 40% and 50% of the overall variation among different cognitive tasks.