What do you think?
Rate this book
Causal Inference and the People's Health
320 pages, Hardcover
Published January 24, 2025
Ratings & Reviews
Friends & Following
Create a free account to discover what your friends think of this book!
Community Reviews
Displaying 1 of 1 review
November 30, 2025
Warning: this is a long and stats-heavy review so feel free to keep scrolling if that is not of interest!
As the title promises, this is a small book with big ideas about causal inference in epidemiology. Schwartz and Prins’ core thesis is that the potential outcomes framework in causality limits the types of causal questions we can answer, and they suggest an alternative approach.
What is the potential outcomes framework and why does this limit our causal questions? If you’re a normal person who is unfamiliar, potential outcomes has roots in Jerzy Neyman’s research but was formalized by Donald Rubin in the 1970s. The key idea is that we understand causation as the difference between what happened and what would have happened under identical conditions but with a different treatment. For example, Y(treatment = 1) is an individual’s LDL cholesterol having taken a medication, Y(treatment = 0) is their cholesterol in a parallel world having not taken the medication, and the treatment effect is Y(treatment = 1) - Y(treatment = 0). While this cannot be observed for an individual, an RCT allows us to estimate the average treatment effect because randomization ensures the groups are similar on average but for the treatment. As such, any post-treatment difference in outcomes is the causal effect of that treatment.
So far so good. However, this limits the types of causal effects we can identify to those that are based on a hypothetical intervention. In other words, we can estimate the effect of medications, lifestyle changes, etc., if we work backwards from the gold standard of imagining this being tested in an RCT. But what about everything else? Using this framework, we cannot study the causal effect that immutable but equally important characteristics like race have on health outcomes, because we cannot imagine manipulating them in an RCT. The potential outcomes framework thereby leads us to overlook social determinants of health, despite their critical importance. Schwartz and Prins include a brilliant quote from Thomas Kuhn to capture this challenge:
“One of the things a scientific community acquires with a paradigm is a criterion for choosing problems that, while the paradigm is taken for granted, can be assumed to have a solution. To a great extent these are the only problems that the community will admit as scientific or encourage its members to undertake. Other problems, including many that had previously been standard, are rejected as metaphysical, as the concern of another discipline, or sometimes just as too problematic to be worth the time. A paradigm can, for that matter, even insulate the community from those socially important problems that are not reducible to the puzzle form, because they cannot be stated in terms of the conceptual and instrumental tools the paradigm provides”.
TK was cooking here!
So, why does this matter and what do we do about it?
This matters because understanding etiologic processes — the actual mechanisms by which structures impact health — helps us design more effective interventions. If we're running unsuccessful smoking cessation trials when the underlying cause is chronic economic anxiety that makes cigarettes an essential stress management tool, any intervention that ignores economic conditions is fighting an uphill battle. In a paradoxical way, thinking less rigidly about interventions — being willing to study causal processes even when we can't immediately manipulate them — can actually help us conceive of more effective interventions. When you understand the mechanisms, you know where the points with the most leverage are.
What do we do about it? They recommend a “quasi-popperian, informal Bayesian approach”, beginning with theory. For example, start with the idea that structural racism impacts health outcomes in four ways:
1. Historical redlining concentrated Black families in neighborhoods with environmental hazards (lead, pollution)
2. Ongoing discrimination creates chronic stress
3. Segregation limits access to quality healthcare
4. Wealth inequality from historical exclusion affects nutrition, housing quality, healthcare access
From there, you make predictions based on theory, test them, update your theories, and use this better understanding of mechanisms to design interventions that target them. If, for example, wealthy black Americans have worse outcomes than equally wealthy white Americans, we know number four does not tell the full story.
Anyone who has tried mediation analysis knows it’s a complex, messy process, but we are scientists — we can and should do hard things!
Loved the book and appreciate the authors’ willingness to explore such a difficult yet important topic.
As the title promises, this is a small book with big ideas about causal inference in epidemiology. Schwartz and Prins’ core thesis is that the potential outcomes framework in causality limits the types of causal questions we can answer, and they suggest an alternative approach.
What is the potential outcomes framework and why does this limit our causal questions? If you’re a normal person who is unfamiliar, potential outcomes has roots in Jerzy Neyman’s research but was formalized by Donald Rubin in the 1970s. The key idea is that we understand causation as the difference between what happened and what would have happened under identical conditions but with a different treatment. For example, Y(treatment = 1) is an individual’s LDL cholesterol having taken a medication, Y(treatment = 0) is their cholesterol in a parallel world having not taken the medication, and the treatment effect is Y(treatment = 1) - Y(treatment = 0). While this cannot be observed for an individual, an RCT allows us to estimate the average treatment effect because randomization ensures the groups are similar on average but for the treatment. As such, any post-treatment difference in outcomes is the causal effect of that treatment.
So far so good. However, this limits the types of causal effects we can identify to those that are based on a hypothetical intervention. In other words, we can estimate the effect of medications, lifestyle changes, etc., if we work backwards from the gold standard of imagining this being tested in an RCT. But what about everything else? Using this framework, we cannot study the causal effect that immutable but equally important characteristics like race have on health outcomes, because we cannot imagine manipulating them in an RCT. The potential outcomes framework thereby leads us to overlook social determinants of health, despite their critical importance. Schwartz and Prins include a brilliant quote from Thomas Kuhn to capture this challenge:
“One of the things a scientific community acquires with a paradigm is a criterion for choosing problems that, while the paradigm is taken for granted, can be assumed to have a solution. To a great extent these are the only problems that the community will admit as scientific or encourage its members to undertake. Other problems, including many that had previously been standard, are rejected as metaphysical, as the concern of another discipline, or sometimes just as too problematic to be worth the time. A paradigm can, for that matter, even insulate the community from those socially important problems that are not reducible to the puzzle form, because they cannot be stated in terms of the conceptual and instrumental tools the paradigm provides”.
TK was cooking here!
So, why does this matter and what do we do about it?
This matters because understanding etiologic processes — the actual mechanisms by which structures impact health — helps us design more effective interventions. If we're running unsuccessful smoking cessation trials when the underlying cause is chronic economic anxiety that makes cigarettes an essential stress management tool, any intervention that ignores economic conditions is fighting an uphill battle. In a paradoxical way, thinking less rigidly about interventions — being willing to study causal processes even when we can't immediately manipulate them — can actually help us conceive of more effective interventions. When you understand the mechanisms, you know where the points with the most leverage are.
What do we do about it? They recommend a “quasi-popperian, informal Bayesian approach”, beginning with theory. For example, start with the idea that structural racism impacts health outcomes in four ways:
1. Historical redlining concentrated Black families in neighborhoods with environmental hazards (lead, pollution)
2. Ongoing discrimination creates chronic stress
3. Segregation limits access to quality healthcare
4. Wealth inequality from historical exclusion affects nutrition, housing quality, healthcare access
From there, you make predictions based on theory, test them, update your theories, and use this better understanding of mechanisms to design interventions that target them. If, for example, wealthy black Americans have worse outcomes than equally wealthy white Americans, we know number four does not tell the full story.
Anyone who has tried mediation analysis knows it’s a complex, messy process, but we are scientists — we can and should do hard things!
Loved the book and appreciate the authors’ willingness to explore such a difficult yet important topic.
Displaying 1 of 1 review