Subscribed to comment. I spent a large part of the 2015-2020 era working on the opioid epidemic with state governments. I met with probably 30+ state health directors about modernizing their systems for opioid response.
The truth of the matter is that public health depts in Dem run states were absolutely concerned with cracking down on pill mill drs and clawing funds from bad actors in pharma (good). But then the epidemic moved to hard drugs, synthetics, and fentanyl and with that move the states lost visibility into the problem (bad) and have never fixed it.
They could have turned to law enforcement as a better proxy but they did not. My experience with these states was that they did not take it nearly as seriously as they should have. Public health officials were happy to spend money on a wide range of programs but despite cataclysmic deaths there was a surprising lack of urgency. In New Hampshire, for example, the health agency CIO actually killed the project because of his discomfort in combining healthcare and public health media data in a single analytics system.
The people are not wrong for reacting like this. Cracking down on the criminal supply chain is still an unsolved problem esp at the state level. And once the issue moved out of the neat ICD codes, Medicare SORs, and PDMPs then States lost visibility and really lost interest in solving from the public health agency level.
“The paper finds that higher rates of cancer deaths in 1996 made locations the target of Purdue Pharma’s initial wave of marketing, which led to significantly higher deaths, reliance on public assistance and social insurance programs, and a rising Republican share of the vote””
Aren’t they just finding older, poorer areas? Cancer death rates correlate with age and I’m going to guess inversely correlate with socioeconomic status.
The pain control push was a nationwide affair and it was pushed strongly into the medical system. Only some places fell into the throes of full blown opioid crisis, and it seems like those were the places that were already ground down by other problems and would likely undergo a shift anyway. The Rust Belt was going Republican / grievance politics before the opioid epidemic and likely didn’t require the epidemic to continue.
It’s been well documented that Republicanism has been having much more success with lower SES individuals, particularly since Trump.
Yeah, that’s what I was going to say. This isn’t that impressive as an argument that the opioids are causal unless you can show that the thing causing the opioids is independent of other relevant factors. How much variance is there in cancer rates that isn’t explained by age and poverty? And even if one county has higher cancer rates caused by a cosmic ray burst or whatever other deus ex machina you want, why think it’s the opioids causing the political change rather than the cancer itself?
This seems like a major flaw in the hiring of political professionals if their skill is supposed to be winning elections. "Time and again, political professionals assume that their frame for understanding an issue is the primary one, stubbornly sticking to an analysis of root causes, only to find out voters are speaking a completely different language."
I find that people with economics backgrounds tend to be really sharp and alert to flawed causal arguments... except in economics. The basic techniques in this paper for converting correlation into causation are, basically, a bullshit generating engine. The fact that they're standard practice in economics is why economics writ large is a bullshit generating engine. TBC, I can totally believe that opioids made people more Republican, because I buy the larger argument that public disorder drives a conservative authoritarian backlash. I just don't believe that the incredibly contrived pseudorandomization in this white paper - cancer rates in the 90s! - does anything at all to resolve the causal structure and I desperately wish people would stop crediting this approach.
This is just a standard instrumental variables study. The idea is that if B causes C and you know that A causes B and has no obvious alternative pathways by which it could cause C, then A should have a correlation with C that can be explained by its effect on B. In this case, A is cancer mortality, B is opioid use, and C is political shift.
Of course, this doesn't give you undeniable proof of causation (you can't be sure that A doesn't correlate with C through other means), but it is evidence.
The problem here is that A seems to have *many* obvious pathways by which it could affect C - cancer tends to be correlated with a lot of political-economic variables like age and poverty and types of factories, and cancer itself is a significant life experience that can affect people’s politics (and that of those in their life).
That's true, but a priori, cancer could have a negative, positive, or null relationship to a particular political ideology. If it has the relationship predicted by your theory, is that not evidence (albeit weak evidence) of your theory?
Yeah, I think there’s interesting work to be done to figure out all the pathways by which cancer rates in the mid 90s affect changes in partisanship a few decades later. I just don’t think one can confidently claim that it’s all via opioids. (Though I haven’t read the paper to see what sorts of statistical methods they use to try to rule out alternatives.)
Yes. Instrumental variables studies are exactly what I am describing as bullshit.
- Correlation is not transitive
- There are many unobserved variables with unknown causal relationships to A, B, and C
- Specifically, A (cancer incidence) is not a random variable. It is a function of many variables - average age of the population, primary occupation, smoking, pollution, race - which all have independent relationships with the outcome of interest (political lean). In fact, if it weren't for these variables, cancer incidence would be geographically uniform and there would be nothing to correlate political outcomes with in this paper.
- Even in the event that you had unlimited independent and identically distributed measurements of the full graph, and both the graph and errors happen to satisfy some incredibly strict constraints that don't apply in reality, learning the true Bayesian graph is NP hard: https://jmlr.csail.mit.edu/papers/volume5/chickering04a/chickering04a.pdf
- But we don't have unlimited i.i.d. measurements of the graph.
- The ability to pick your "instrumental" variable gives almost unlimited latitude to the authors to engineer whatever outcome they want. Here, they not only get to choose to use diagnoses of a disease, but which disease (cancer is not actually a natural choice, many more people prescribed opiates for cancer in the 90s would die of their disease than O.D.) and time frame (the 90s???)
All this is without bothering to read the white paper, which I guarantee would reveal additional methodological problems.
Sure, but a priori there is no reason to expect the instrument to have any particular relationship with the outcome. So if it has the relationship predicted by your theory, that is evidence for your theory.
Now, you're right that the ability of researchers to pick the instrument presents the opportunity for bias to seep in, which is why you shouldn't update too much on any given IV finding and always stay open to the possibility of future evidence casting doubt on it. Still, I think you can update on it somewhat, especially if you have mitigations in place like preregistration.
But as I and Kenny pointed out, there are many cancer correlates that are a priori expected to have a relationship with political lean. Age and blue-collar lifestyle are not minor factors in politics. They're like the biggest stories of the political realignment!
And even if we had no a priori expectation about them, the correlation would still only be evidence that some cofactor influenced partisan lean. It still wouldn't tell you anything about the specific cofactor the authors want you to pay attention to.
I know this isn't really the purview of this post, but how do we make sense of the polling finding that the issue with immigration is crime when we also know that illegal immigrants commit fewer crimes than native citizens?
Like, if we had a magic button that would safely teleport all illegal immigrants currently in the US to their home countries the crime rate in the US would go up...
I realize that the average American is not a perfect machine that feeds knowing all facts directly into their political worldview.
But to your point about having to go through the steps of identifying issues, agreeing that they are the fundamental issue, and then moving on to solutions, what do you do if the problem voters are seeing are...fake or, more politely, incorrect.
First, people perceive publicly visible disorder and ugliness as a crime-like threat, and a threat that makes them believe there's more crime around, even when there isn't. And the poor logistical handling of the Biden-era migration surge did produce some of that disorder and ugliness.
Second, racism. People are more inclined to be suspicious of others' possible criminal intentions when those others don't look like them and/or don't speak their language.
As depressing as both of those forms of irrationality are, quoting the statistics about low rates of immigrant crime at people is unlikely to change them. We need people who are good at understanding social-psychological intuitions to devise more effective forms of what one might call "bias interruption".
I am very aware of the latter idea. The former makes sense. But it still doesn't totally answer my question (not saying you specifically have to).
In a world where voters demand solutions, this still represents a made up problem. Per my example, removing all illegal immigrants will not fix their stated concern.
Further, while I take the point that disorder roughly equals "crime" in the average person's mind, there is something else going on here. You can see that from how high the anti-immigration sentiment is/was in places with incredibly low levels of both immigration and crime!
NYC unfairly had immigrants literally bused to them. It did increase burden and definitely cause a backlash. But it didn't, like, turn NYC red or anything of the sort.
On the other hand you have people in 99% racially homogenous places with basically zero homelessness or vagrancy (by any race) saying immigration is a huge issue for them...
Great column, thank you. Yglesias makes a related point today about Democratic partisans:
“The way you become mayor of San Francisco and then rise to statewide office in California is by calibrating your appeal to hardcore partisan Democrats, so of course you feel that his vibes are impeccable. A politician whose vibes are attuned to maximizing crossover appeal and winning swing votes probably will not vibe that hard with you, unless you personally are a swing voter.”
I can't really support it, but my interpretation of the data is that the folks who are trying to point to this to say that drug-testing of recipients is unfair / unnecessary are being dumb -- that it's the common expectation that drug-testing will be required for welfare that is _causing_ the decline in drug use among recipients, even in states that actually aren't testing. A lot of people who are on drugs, and could use the help, _believe_ they won't be able to get it, and thus don't bother applying.
On the one hand, there is a valid argument that what we really want is for the people on drugs to still seek help, including treatment to break their addictions. On the other hand, the testing seems to be doing exactly what it was intended to do. On the gripping hand, the testing regime as it's currently administered almost certainly costs the state more money than it saves in reduced welfare payments, both directly, and through the indirect cost of causing many people who _aren't_ on drugs to end up losing their welfare, and falling deeper into poverty, where they're more likely to _become_ criminals and/or addicts.
I think it’s quite plausible that medical treatment aimed at getting people off opioids is actually more effective at preventing deaths than locking people in prison. It’s quite well known that there’s a lot of drug use in prison. Of course we would need actual studies to figure out whether the medical treatment is in fact effective, but I think that’s the claim of the article - that studies have in fact found that Medicaid is more effective than prison at preventing overdose deaths.
I really enjoyed listening to the podcast this morning. Honestly, at first I wasn’t sure about the correlation between the opioid epidemic and parts of America switching to the Republican Party. However, the explanation of secondary and tertiary effects i’m not only the individuals using the opioids, but also the families and neighbors in the community really hit the point home. I also enjoyed this article explaining other major political issues that need to be thought of in a new way when it comes to understanding why Republicans might be trusted more on their policies to address them.
Subscribed to comment. I spent a large part of the 2015-2020 era working on the opioid epidemic with state governments. I met with probably 30+ state health directors about modernizing their systems for opioid response.
I've written about my experience drawing a parallel to the MN Feed Our Families fraud (https://open.substack.com/pub/socialdawn/p/the-banal-fraud-of-minnesota) because I think it's a good comparison in some respects.
The truth of the matter is that public health depts in Dem run states were absolutely concerned with cracking down on pill mill drs and clawing funds from bad actors in pharma (good). But then the epidemic moved to hard drugs, synthetics, and fentanyl and with that move the states lost visibility into the problem (bad) and have never fixed it.
They could have turned to law enforcement as a better proxy but they did not. My experience with these states was that they did not take it nearly as seriously as they should have. Public health officials were happy to spend money on a wide range of programs but despite cataclysmic deaths there was a surprising lack of urgency. In New Hampshire, for example, the health agency CIO actually killed the project because of his discomfort in combining healthcare and public health media data in a single analytics system.
The people are not wrong for reacting like this. Cracking down on the criminal supply chain is still an unsolved problem esp at the state level. And once the issue moved out of the neat ICD codes, Medicare SORs, and PDMPs then States lost visibility and really lost interest in solving from the public health agency level.
“The paper finds that higher rates of cancer deaths in 1996 made locations the target of Purdue Pharma’s initial wave of marketing, which led to significantly higher deaths, reliance on public assistance and social insurance programs, and a rising Republican share of the vote””
Aren’t they just finding older, poorer areas? Cancer death rates correlate with age and I’m going to guess inversely correlate with socioeconomic status.
The pain control push was a nationwide affair and it was pushed strongly into the medical system. Only some places fell into the throes of full blown opioid crisis, and it seems like those were the places that were already ground down by other problems and would likely undergo a shift anyway. The Rust Belt was going Republican / grievance politics before the opioid epidemic and likely didn’t require the epidemic to continue.
It’s been well documented that Republicanism has been having much more success with lower SES individuals, particularly since Trump.
Yeah, that’s what I was going to say. This isn’t that impressive as an argument that the opioids are causal unless you can show that the thing causing the opioids is independent of other relevant factors. How much variance is there in cancer rates that isn’t explained by age and poverty? And even if one county has higher cancer rates caused by a cosmic ray burst or whatever other deus ex machina you want, why think it’s the opioids causing the political change rather than the cancer itself?
This seems like a major flaw in the hiring of political professionals if their skill is supposed to be winning elections. "Time and again, political professionals assume that their frame for understanding an issue is the primary one, stubbornly sticking to an analysis of root causes, only to find out voters are speaking a completely different language."
I find that people with economics backgrounds tend to be really sharp and alert to flawed causal arguments... except in economics. The basic techniques in this paper for converting correlation into causation are, basically, a bullshit generating engine. The fact that they're standard practice in economics is why economics writ large is a bullshit generating engine. TBC, I can totally believe that opioids made people more Republican, because I buy the larger argument that public disorder drives a conservative authoritarian backlash. I just don't believe that the incredibly contrived pseudorandomization in this white paper - cancer rates in the 90s! - does anything at all to resolve the causal structure and I desperately wish people would stop crediting this approach.
This is just a standard instrumental variables study. The idea is that if B causes C and you know that A causes B and has no obvious alternative pathways by which it could cause C, then A should have a correlation with C that can be explained by its effect on B. In this case, A is cancer mortality, B is opioid use, and C is political shift.
Of course, this doesn't give you undeniable proof of causation (you can't be sure that A doesn't correlate with C through other means), but it is evidence.
The problem here is that A seems to have *many* obvious pathways by which it could affect C - cancer tends to be correlated with a lot of political-economic variables like age and poverty and types of factories, and cancer itself is a significant life experience that can affect people’s politics (and that of those in their life).
That's true, but a priori, cancer could have a negative, positive, or null relationship to a particular political ideology. If it has the relationship predicted by your theory, is that not evidence (albeit weak evidence) of your theory?
Yeah, I think there’s interesting work to be done to figure out all the pathways by which cancer rates in the mid 90s affect changes in partisanship a few decades later. I just don’t think one can confidently claim that it’s all via opioids. (Though I haven’t read the paper to see what sorts of statistical methods they use to try to rule out alternatives.)
Yes. Instrumental variables studies are exactly what I am describing as bullshit.
- Correlation is not transitive
- There are many unobserved variables with unknown causal relationships to A, B, and C
- Specifically, A (cancer incidence) is not a random variable. It is a function of many variables - average age of the population, primary occupation, smoking, pollution, race - which all have independent relationships with the outcome of interest (political lean). In fact, if it weren't for these variables, cancer incidence would be geographically uniform and there would be nothing to correlate political outcomes with in this paper.
- Even in the event that you had unlimited independent and identically distributed measurements of the full graph, and both the graph and errors happen to satisfy some incredibly strict constraints that don't apply in reality, learning the true Bayesian graph is NP hard: https://jmlr.csail.mit.edu/papers/volume5/chickering04a/chickering04a.pdf
- But we don't have unlimited i.i.d. measurements of the graph.
- The ability to pick your "instrumental" variable gives almost unlimited latitude to the authors to engineer whatever outcome they want. Here, they not only get to choose to use diagnoses of a disease, but which disease (cancer is not actually a natural choice, many more people prescribed opiates for cancer in the 90s would die of their disease than O.D.) and time frame (the 90s???)
All this is without bothering to read the white paper, which I guarantee would reveal additional methodological problems.
Sure, but a priori there is no reason to expect the instrument to have any particular relationship with the outcome. So if it has the relationship predicted by your theory, that is evidence for your theory.
Now, you're right that the ability of researchers to pick the instrument presents the opportunity for bias to seep in, which is why you shouldn't update too much on any given IV finding and always stay open to the possibility of future evidence casting doubt on it. Still, I think you can update on it somewhat, especially if you have mitigations in place like preregistration.
But as I and Kenny pointed out, there are many cancer correlates that are a priori expected to have a relationship with political lean. Age and blue-collar lifestyle are not minor factors in politics. They're like the biggest stories of the political realignment!
And even if we had no a priori expectation about them, the correlation would still only be evidence that some cofactor influenced partisan lean. It still wouldn't tell you anything about the specific cofactor the authors want you to pay attention to.
I know this isn't really the purview of this post, but how do we make sense of the polling finding that the issue with immigration is crime when we also know that illegal immigrants commit fewer crimes than native citizens?
Like, if we had a magic button that would safely teleport all illegal immigrants currently in the US to their home countries the crime rate in the US would go up...
I realize that the average American is not a perfect machine that feeds knowing all facts directly into their political worldview.
But to your point about having to go through the steps of identifying issues, agreeing that they are the fundamental issue, and then moving on to solutions, what do you do if the problem voters are seeing are...fake or, more politely, incorrect.
First, people perceive publicly visible disorder and ugliness as a crime-like threat, and a threat that makes them believe there's more crime around, even when there isn't. And the poor logistical handling of the Biden-era migration surge did produce some of that disorder and ugliness.
Second, racism. People are more inclined to be suspicious of others' possible criminal intentions when those others don't look like them and/or don't speak their language.
As depressing as both of those forms of irrationality are, quoting the statistics about low rates of immigrant crime at people is unlikely to change them. We need people who are good at understanding social-psychological intuitions to devise more effective forms of what one might call "bias interruption".
I am very aware of the latter idea. The former makes sense. But it still doesn't totally answer my question (not saying you specifically have to).
In a world where voters demand solutions, this still represents a made up problem. Per my example, removing all illegal immigrants will not fix their stated concern.
Further, while I take the point that disorder roughly equals "crime" in the average person's mind, there is something else going on here. You can see that from how high the anti-immigration sentiment is/was in places with incredibly low levels of both immigration and crime!
NYC unfairly had immigrants literally bused to them. It did increase burden and definitely cause a backlash. But it didn't, like, turn NYC red or anything of the sort.
On the other hand you have people in 99% racially homogenous places with basically zero homelessness or vagrancy (by any race) saying immigration is a huge issue for them...
Great column, thank you. Yglesias makes a related point today about Democratic partisans:
“The way you become mayor of San Francisco and then rise to statewide office in California is by calibrating your appeal to hardcore partisan Democrats, so of course you feel that his vibes are impeccable. A politician whose vibes are attuned to maximizing crossover appeal and winning swing votes probably will not vibe that hard with you, unless you personally are a swing voter.”
https://www.slowboring.com/p/gavin-newsom-is-very-similar-to-kamala
People on welfare use drugs. People who have to work don't.
That was true in the '90s, but has not been true for the past ~twenty years.
https://www.clasp.org/press-room/news-clips/welfare-recipients-are-less-likely-be-drug-users-average-american-results-show/
https://www.ebsco.com/research-starters/politics-and-government/drug-testing-welfare-recipients
https://journals.healio.com/doi/10.3928/0279-3695-19970201-13
I can't really support it, but my interpretation of the data is that the folks who are trying to point to this to say that drug-testing of recipients is unfair / unnecessary are being dumb -- that it's the common expectation that drug-testing will be required for welfare that is _causing_ the decline in drug use among recipients, even in states that actually aren't testing. A lot of people who are on drugs, and could use the help, _believe_ they won't be able to get it, and thus don't bother applying.
On the one hand, there is a valid argument that what we really want is for the people on drugs to still seek help, including treatment to break their addictions. On the other hand, the testing seems to be doing exactly what it was intended to do. On the gripping hand, the testing regime as it's currently administered almost certainly costs the state more money than it saves in reduced welfare payments, both directly, and through the indirect cost of causing many people who _aren't_ on drugs to end up losing their welfare, and falling deeper into poverty, where they're more likely to _become_ criminals and/or addicts.
I don't see how Medicaid should save people from opioid deaths. Locking them in prison is much more effective if you care about “life.”
I think it’s quite plausible that medical treatment aimed at getting people off opioids is actually more effective at preventing deaths than locking people in prison. It’s quite well known that there’s a lot of drug use in prison. Of course we would need actual studies to figure out whether the medical treatment is in fact effective, but I think that’s the claim of the article - that studies have in fact found that Medicaid is more effective than prison at preventing overdose deaths.
I really enjoyed listening to the podcast this morning. Honestly, at first I wasn’t sure about the correlation between the opioid epidemic and parts of America switching to the Republican Party. However, the explanation of secondary and tertiary effects i’m not only the individuals using the opioids, but also the families and neighbors in the community really hit the point home. I also enjoyed this article explaining other major political issues that need to be thought of in a new way when it comes to understanding why Republicans might be trusted more on their policies to address them.