For all those commenting on 'aren't more polluted areas just more criminal due to other factors", as far as I can tell the researchers examined crime rates for the same place and find correlation with whether there was more Ozone and PM 2,5 on a particular day or not.
So they are not comparing clean air neighborhoods to highly polluted neighborhoods, but the same neighborhood on different days.
That's crazy though, and makes the headline tamer than it need be doesn't it?
'Exposure to' makes me think it's more prolonged, and over time the effect is[...] but actually it's suggesting that it has a reversible, quickly reacting/short-term impact, which (naively) I find much more surprising.
Relevant excerpt:
> They find that a 10% increase in same-day exposure to PM2.5 (particulate matter less than 2.5 microns in diameter) is associated with a 0.14% increase in violent crimes, such as assault. An equivalent increase in exposure to ozone, an air pollutant, is associated with a 0.3% jump in such crimes. Pollution levels can easily rise by much more than that.
Yes - for once the headline seems to have undersold the results.
Reading the title, I was expecting this to be a long-term relationship, and to happen at extremely high levels of air pollution. Particularly since the study on China last year showing that exposure to PM2.5 has lasting cognitive effects, which found variation at levels far higher than anything found in the US. And ozone at approved levels (but not PM2.5) has been linked to long-term cognitive decline.
Instead, this is claiming day-to-day variation based on that PM2.5 levels that are 25% of the EPA 24-hour exposure threshold, and even below the annual EPA standard. If you told me that air pollution below the EPA safe thresholds had a noticeable impact on cancer or emphysema rates, I wouldn't be surprised at all. But this is wild - a 5% increase in violent crime based on a day of pollution! For reference, one violent crime study I just checked finds that violent crime is about 10% higher in the summer than the winter in a cold city, and late-night crime (past midnight) is about 5% higher on weekends than weekdays. So 15 micrograms of PM2.5 is about as impactful as "people stay up late and get drunk on weekends".
It's an effect so bizarrely strong that my first reaction is to assume the study is bad or confounded. Except... all the obvious patterns don't work, and the controls look solid. Unpleasant weather keeps people inside and reduces (outdoor) crime. Shared causes like high human activity should have raised property crime rates. The effect looks stable across all kinds of temperature and crime subsets. If there is a confounder, it's presumably something strong and interesting.
> Pollution levels can easily rise by much more than that.
Very true, although I note that the effect doesn't seem to continue across the whole range studied. I wonder why effect falls off, and what happens in places that are always this polluted?
It wouldn't be a formal fallacy in itself although certainly impolite. Perhaps like Slippery Slope.
It is related to arguement from ignorance and Dunning-Kreuger effect in that it often assumes their comprehended issues are the end of all problems and that the objections may already be implicitly handled. Listing everything explicitly in exhaustive detail technically addresses it but would be both inefficient and bad communication from how overloaded it would get.
That said assuming uncovered issues are implicitly covered is itself a fallacious inference. It may be dickish pedantry that is wrong based upon incomplete information but not illogical given what is known and communicated.
It'd be a corollary of the Dunning-Kruger effect, I guess. Like, you don't know enough about the problem space to know that your first thought is already well known?
> particular emphasis on aggressive behavior. To identify this relationship, we combine detailed daily data on crime, air pollution, and weather for an eight-year period across the United States. Our primary identification strategy employs extremely high dimensional fixed effects and we perform a series of robustness checks to address confounding variation between temperature and air pollution. We find a robust positive effect of increased air pollution on violent crimes, and specifically assaults, but no relationship between increases in air pollution and property crimes. The effects are present in and out of the home, at levels well below Ambient Air Pollution Standards, and PM2.5 effects are strongest at lower temperatures
Well yeah, but they have to remove all the other factors like heat, wind and rain, which are correlated to air pollution.
There is likely less violent crimes on rainy days, just because more people stay home.
To do this properly, they would have to do a controlled experiment with people in rooms with or without air pollution. And measure how sanguine they get or something of that nature.
"Pollution and crime rates may have common correlations with location and time-varying unobservables. For example, PM2.5 or ozone levels and crime rates may be correlated with county-level covariates such as traffic density, population density, demographics, and industrial activity. Failing to control for such covariates will lead to biased estimates of γPM and γo.
Our identification strategy explicitly addresses omitted variable bias in several ways.
First, we show that endogeneity with respect to violent crimes and pollution can be addressed by including a series of high-dimensional fixed effects. In our primary specification, we include county-by-year-by-month-by-day of the week fixed effects to control for county level unobservables that are either constant over time, such as state and county-level policies, or that are time-variant, such as changes in population density, demographic composition, seasonal variation in pollution or crime, or changes in state and county-level policies that limit pollution or crime enforcement. These fixed effects also control for cyclical within-week, within-county variation in pollution and crime. Thus, our data allows us to compare for example, the effect of changes in pollution within a series of Mondays within a given county within a given month. We argue that changes in pollution across a series of Mondays within a county-month, conditional on weather controls, is random and thus exogenous to crime.
Second, crime has been shown to respond to changes in temperature, and temperature is generally correlated with air pollution (Field 1992; Jacob et al. 2007; Ranson 2014). Thus, failure to adequately control for temperature, and weather more generally, will lead to biased estimates. To address this concern, we include temperature and precipitation splines in our primary specification, we provide robustness checks with alternative functions of temperature, and in Section 5.3 we perform a series of tests to show that our results are not confounded by unaccounted for variation between temperature and air pollution. For
example, we show that effect of PM2.5 on violent crime is larger at lower temperatures, opposite of the effect of temperature alone."
I'm not sure this fully accounts for the correlation though. Anything that could temporarily increase the number of people going out that day would also increase pollution, no? I'm not sure this methodology corrects for acyclic events.
I would like to see the predictive value of this study vis-a-vis greatly increasing electric car usage. Holding all variables constant (keeping temperatures level might be tricky), we should see some crime decreases, particularly in Oslo Norway, where electric cars are now > 33% of new sales.
This sort of thing has happened before. Lead was removed in the late 70's, and, lo, 20ish years later, crime rates dropped -- perhaps, for the reason there are fewer neurological impacts growing up without airborne lead.
Non exhaust emissions account for 85% of PM 2.5 produced by a car. As EV's are heavier than equivalent combustion engine cars, the end tally is a savings of just 1%-3% by going electric.
Don't get me wrong, any savings is an improvement, but EVs are not the 'green' transport revolution the industry wants to picture them as, just an incremental step.
Statistical controls can remove much of the effect from known confounders, but they don't turn an observational study into a controlled experiment.
Sometimes you can use common sense to infer a causal relationship (violent crime linked to emergency room visits), but it seems like a leap here, especially as they're on a bit of a fishing expedition with multiple types of pollution and multiple types of crime.
And as someone living in Chicago or any similar metro area with strong seasonal weather can tell you, there are far higher crime rates in the summer than in the winter. Young people have more down time in the summer and people are outside and get restless in the heat. People stay inside in the cold winter and keep to themselves more.
That was my immediate thought. More people = more pollution, and also more crime.
The authors are aware of this and attempt to control, mainly by assuming that all confounding effects are either constant or cyclical:
"We include county-by-year-by-month-by-day of the week fixed effects to control for county level unobservables that are either constant over time, such as state and county-level policies, or that are time-variant, such as changes in population density, demographic composition, seasonal variation in pollution or crime, or changes in state and county-level policies that limit pollution or crime enforcement. These fixed effects also control for cyclical within-week, within-county variation in pollution and crime. Thus, our data allows us to compare for example, the effect of changes in pollution within a series of Mondays within a given county within a given month. We argue that changes in pollution across a series of Mondays within a county-month, conditional on weather controls, is random and thus exogenous to crime."
I'm not at all convinced this holds true, however - lots of systems with humans in it are chaotic rather than cyclical. I imagine if you did a plot of population over time, it would be very noisy. All it takes is a bit of aperiodic noise in the population, and tight correlations between population/crime and population/pollution, and their effect is rendered void.
People have good reasons to be skeptical about such studies: there's like a million of possible confounders for these kind of things, and it's impossible to control for every one of them (remember, if you control for 2 confounders which are themselves confounded by a common source, you're not controlling for anything and you're just adding bias).
People should just stop doing this kind of studies, we are just not equipped with powerful enough mathematical tools yet to deal with this kind of highly entangled phenomenons (and we often don't have the data either, collecting and curating it would be an interesting and useful research topic by itself).
The way these tools are developed is by doing and then replicating and analyzing these studies. (maybe thats what the intent behind "collecting and curating" was?)
Exposure to air pollution seems to my intuition like it lowers a persons level of consciousness, and the lower a persons level of consciousness (the further down The Spiral Dynamics model they are) the more prone they are to violence.
> The way these tools are developed is by doing and then replicating and analyzing these studies
Nobody in the field of people publishing this kind of studies has the required skill-set to develop the needed tools. This is the job of mathematicians. I'm not meaning it pejoratively, it's just a totally different job. (I wouldn't trust a mathematician to build a house, you shouldn't trust an economist, a biologist, or anyone to build statistics tooling).
> maybe thats what the intent behind "collecting and curating" was?)
Most of the time the data behind this kind of studies isn't easily available. Sometimes you can shoot an email to the author, sometimes it comes from some publicly available data needing a lot of preprocessing before being consumable, and I wish we had a centralized database of well-formed data.
You seem to be overlooking the possibility that such study teams can and do include mathematicians. I'm friends with one who does the statistical legwork for a foundation that studies the spread of infectious disease.
Sorry for the slow reply. My friend sometimes publishes alone and sometimes in concert with researchers from other disciplines; My understanding is that the research team has fairly wide latitude compared to people working in academe, but I don't know their publication history/process in detail.
I see, and agree. It seems though that we've (as a society) gotten into a habit of categorizing people.
Maybe rightfully as specialization has increased, but it seems to me (perhaps wrongfully) that there were more polymaths in Ben Franklins time for example.
Yeah, like looking for perpetual movement or the philosopher stone.
BTW, I should have worded my thoughts differently: I don't mind if individual people waste their time doing that kind of pseudo-science.
I do mind however that these studies get fundings instead of more down-to-earth ones. I do mind when young researchers are driven to this kind of bs by perverse insentives. And I do mind when laypersons are taught by mainstream media that this is science.
> And I do mind when laypersons are taught by mainstream media that this is science.
What is your definition for "science"?
To me, this is very much science. Starting off with a hypothesis, gathering/analyzing data, controlling for what you can, and concluding your findings with a number of caveats detailing the limitations of your research.
Unfortunately, not everything is a hard science, and even those sciences we consider hard sciences break down when you get deep enough.
I think your suggestion is rather dangerous and implies some arbitrary threshold where we become capable of studying something.
A simple example would by psychology. Whenever a psychology study comes out, people are quick to yell that it is useless, that there are flaws, etc. However, without these psychology studies, where would we begin when studying mental health issues?
How many people would be left to suffer untreated if we were to wait for the biology, chemistry, neurology, etc. fields to become advanced enough for us to detect with certainty the presence of a mental illness?
It's easy to belittle an study that turns up nothing or belongs to a soft science, but as the saying goes, "all models are wrong, but some models are useful."
> I think your suggestion is rather dangerous and implies some arbitrary threshold where we become capable of studying something.
The threshold isn't so much about when we become capable of "studying something" but more about when we become capable of claiming something.
In that particular case, we have so many things at stakes, it becomes really tough to find anything useful. How can you know you're not measuring the correlation between weather and crime? Day of the week and crime? Season and crime? Because season and weather are related, you can't just “control for both” (unfortunately, many people will do it no matter what…). What if air pressure had an impact on crime? It's somehow correlated to air pollution so you could mistake it for an influence of pollution itself.
When just asking the question reveals that the answer would need fancy statistics (some not being invented yet, AFAIK, or at least not mainstream). If you realize that, you should just find another problem, or maybe rescope your problem. If you don't realize that, and still go on with your regular methodology, your results are doomed from the start, and you aren't really doing more science than youtubers attempting to find the recipe for perpetual motion. [I'm not talking about this specific paper here, just answering about your remarks about threshold].
When it comes to this paper, I would say that the level of accuracy of their published result (“0.14% increase”) is dubious. In a lab with controlled conditions, such figures aren't that easy to get, in a real world observation of human behavior, it's looks fishy.
Traffic, and especially the noise caused by it would definitely be a suspect. However, the paper did take into account the precipitation, with the same amount of traffic causing less PM 2.5 when it rains.
This doesn't mean it is impossible, as rain might e.g. still be correlated with less traffic. The nature of a hidden variable is exactly that its indirect correlations would be not pinpoint-able in the data.
But IMHO this paper does a decent job of trying to account for lots of dimensions, and trying identify the existence of dominant hidden variables from the data at hand.
This is far from a rush job or a 'spurious correlation' piece of work, and its results most certainly are interesting and a very good basis for further investigation into an important observation. I'd even go further and say that these results warrant societal value systems that subscribe to the precautionary principle to take this on board in further decisions about pollution in populated areas.
From the article you linked: "The economists calculated that a 10 percent reduction in daily PM2.5 could save $1.1 million in crime costs per year, which they called a 'previously overlooked cost associated with pollution.'"
Is that 'million' a typo? I imagine $1.1m amounts to something like one homicide a year.
properly correcting for socioeconomic status in determining causal relationships in epidemiology is still a very hard problem.
What if, in this case, pollution has a delayed effect? How would their methodology deal with that (it wouldn't)?
Personally, I think what happened here is the author started with a prior belief: ""Several years ago, Fort Collins experienced a fairly severe wildfire season," Burkhardt said. "The smoke was so bad that after a few days, I started to get frustrated, and I wondered if frustration and aggression would show up in aggregate crime data.""
and then they conducted a study that supported their belief, possibly making methodological errors along the way , generating a scientific narrative to support their belief.
the article is paywalled, also, since I have plenty of experience in this field, I do not think my remarks were "accounted for", rather they probably wrote a couple hand-wavy sentences about how they corrected for SES, but didn't really.
Edit: found a copy of the paper that wasn't paywalled, read the whole thing, and found their attempts to correct for bias wanting (section 4.2).
" Overall, our heterogeneity results indicate the most important explanatory factor in the relationship between pollution and crime is age, and the
results are not ameliorated by higher incomes."
uh, ok. if you want to defend this paper go ahead but I've already read enough to know these folks started with a biased belief, carelessly analzyed the data in a way that supported their beliefs, and escalated the results into something that sounded far more significant and certain than their data shows.
These studies only report on the evidence of a statistical relationship and not a guaranteed causal link. They also don’t use correlation but usually some sort of regression that includes other factors like temperature to remove some of the effects of these variables.
Linear regression is a form of correlation. The model causality is implied but the only connection is correlation. Unless estimating an average treatment effect or diff-in-diff type framework, causality will be hard to identify in observational frameworks
"The researchers made no claims on the physiological, mechanistic relationship of how exposure to pollution leads someone to become more aggressive; their results only show a strong correlative relationship between such crimes and levels of air pollution.
The researchers were careful to correct for other possible explanations, including weather, heat waves, precipitation, or more general, county-specific confounding factors."
> their results only show a strong correlative relationship
The paper's abstract says "The results suggest that a 10% reduction in daily PM2.5 and ozone could save $1.4 billion in crime costs per year, a previously overlooked cost associated with pollution." In my opinion this is a statement of a causal relationship, not merely a correlation.
I wonder why they didn't reverse that relationship, too, and propose that violent crime increases pollution.
I would love to read the actual paper, but it's only available to people with money.
> I wonder why they didn't reverse that relationship, too, and propose that violent crime increases pollution.
Probably because you can easily rule out that possibility using common sense, whereas causation in the other direction is at least thinkable and worth investigating further.
Related: "much of the decline in crime in the 1990s may have been due to the reduction of childhood lead exposure after the removal of lead from gasoline and house paint."
I absolutely believe that removing it from gasoline is a big factor, but I think it's often over emphasized in the context of today.
Cleveland has twice the prevalence of lead paint over any other big city in the united states, yet has 25% lower violent crime prevalence than say, st louis, baltimore, detroit, and memphis.
Lead paint tends to stay on the wall so as long as you don't eat paint chips it's not as big a risk as particulate lead in fuel exhaust. If you are removing lead paint you need to wear a suitable respirator to avoid breathing in fragments but just being in a room with lead paint or a lump of the stuff isn't especially dangerous.
The abortion theory is pretty much busted by now. You find the same crime-reduction pattern in European countries were abortion has been legal since the 1940s. You don't find subsequent crime increases in countries were abortion has been outlawed or access to it has been severely restricted.
Those are dates in the 60's though, so not sure what European countries you're referring to. Looking at https://en.wikipedia.org/wiki/Timeline_of_reproductive_right... the only country with on-demand abortion before 1950s was the Soviet Union for a brief period.
The hypothesis is that abortion causes lower crime rates. Thus, a few or even a single counter-example is enough to invalidate the hypothesis. Data points that confirm the hypothesis doesn't add much new information since these results too could be caused by the same confounding variables that caused the first result (correlation doesn't imply causation).
Yes, it was the Soviet union and other Eastern European countries that had liberal abortion laws which they then reversed to spur population growth that I referred to. If Donohue and Levitt's theory were correct, the generation born during the years abortion was legal would be less criminal than following generations.
hmm, well I did listen to the above podcast which was recorded in 2019, and they revisited the abortion numbers and still thought it was very relevant.
Donohue and Levitt have, as they say, "dug their heels in." But other researchers have tried and failed to find correlations between abortion and crime in other datasets. See f.e. https://www.jstor.org/stable/40071728?seq=1#metadata_info_ta... and https://pdfs.semanticscholar.org/9401/e8f1a53aa4437beed139ec... Other researchers have focused on Donohue and Levitt's study itself and found methodological flaws. See https://www.nber.org/papers/w15098.pdf and https://papers.ssrn.com/sol3/papers.cfm?abstract_id=866864: "We revisit that paper, showing that the actual implementation of DL's statistical test in their paper differed from what was described. (Specifically, controls for state-year effects were left out of their regression model.) We show that when DL's key test is run as described and augmented with state-level population data, evidence for higher per capita criminal propensities among the youths who would have developed, had they not been aborted as fetuses, vanishes."
The main problem I see with this study is that it could be confounded by the fact that high air pollution areas have other problems too that could be causing the bad behaviour instead, like overcrowding, noise, frustration at slow traffic, etc.
The study solved this problem by looking at changes in the same areas, not between areas.
Slow traffic would increase pollution. Noise could be an interesting one, I'm not sure whether there are statistics that measure noise pollution over a country, and how would noise pollution be coincident with air pollution? Overpopulation wouldn't count as they are measuring the same areas with the same population. Daytime population could be an issue (e.g. weekends vs weekdays) but I can't access the study.
Reminds me of one of the prototypical sociological study examples of correlation vs causation where something akin to icecream sales correlated highly with murder rates.
The fact air pollution, in many cases, directly correlates with population density, make the results seem less surprising. It would be more interesting if less populated areas with high air pollution (perhaps near industrial plants, etc.) had a high correlation to violent crime rates in conjunction with highly dense populated areas.
The article specifically mentions the lead-crime hypothesis as a different example of hypothesized linkages between air pollution and crime. It seems exceptionally likely that it isn’t what they’re looking at.
This paper specifically looks at short term air pollution effect (ie, the kinds that will measurably rise and fall in different parts of a city over an 8-year span). Lead takes so long to affect a single person (since it mostly harms brain development in children, but crimes are mostly not committed by children) that it almost certainly being meaningfully measured here.
And we're currently in this weird situation where older criminals are more violent than younger criminals, in contravention of how things typically go, because older criminals were exposed to lead as children.
Of course I didn't read the article. I hear it's paywall, and I and the economist's paywall really don't get along.
But, it being a short term study doesn't change much. A temporary increase in pollutants will serve to exacerbate underlying conditions. Supposing areas which experience the largest increase in the short term were also susceptible to higher levels historical, and most impacted by previous short term increases and we're back where we started.
First, there's a plausible biological link between lead and violence. We can performed double-blind controlled experiments with mice and see it.
Second, violence decreased in a given state 15 years after that particular state banned leaded gasoline. And it decreased for the particular cohort that had grown up without leaded gasoline but not particularly otherwise.
At a certain point you're left saying something like "Every time I point this gun at someone and pull the trigger they get struck by a bullet but this could be a spurious correlation!"
Some of those are interesting just because I never assumed it happened or it was being tracked. And some of them are a surprising scale. 800 deaths from being tangled in bedsheets!
You'd think it would be fairly easy to control for population density. However the study abstract indicates the only other confounding variable they controlled for was 'temperature'.
Since they were looking at increases in violence and air pollution on a daily interval, temperature was probably there to control for weather-related phenomena.
Very interesting - I remember a study when I was at uni saying that mobile phone masts were linked to cancer rates. They also forgot that mobile phone masts were normally on top of very tall buildings and very tall buildings often house lots and lots of people.
So yes the incidence of cancer was higher in that area, but as soon as you controlled for population density, that effect disappeared.
Stuck with me as the 'most dangerous place' was a group of flats near the house I lived in at that time. I guessed it might be due to population density and was relived to find that was the case some time later.
You can't do it with humans, for ethical reasons, but why not take some other social animal, say mice, put one population in a box with zero pollution, and one hooked up to an old car exhaust with leaded gasoline?
I'd wager the ones with the pollution are going to eat each other's faces at a much higher rate than those without.
For giggles and to test some fall of Rome hypotheses, run a 3rd cage where the water comes through lead pipes.
In any case, I agree you can't do this properly with humans. And lead is definitely a factor in many deceases including mental ones so it could well be one of the causes of increased violence.
It's still widely available (and in use) for small piston aircraft as avgas. And it seems like tetraethyl lead is still available for people to mix their own leaded gas to put in very old cars.
Even within a single area of study "bad days" could involve activities correlated to increased pollution in many different ways.
An example that I can think of offhand: I'm way more stressed on days that neighbors are using powered yard tools (mostly mowers and blowers), which invariably also throw up lots of stuff my body and/or mind react negatively too (never been tested, but I'd be amazed if I didn't have at least an intolerance).
I can't read the article due to the the paywall, but any halfway decent study should be attempting to control for these things. Propensity score matching, or simply including these factors in a regression model should help remove some of the effects. Obviously, there is potentially an unlimited number of possible cofounders, and therefore the assumption that they've all been accounted for should always be met with skepticism, and is why the mantra "correlation is not causation" is so often repeated.
If a modern research paper cited by the Economist has failed to even account for some of the potential confounding here, then I have great concerns for the future of science.
There are many thousands of studies, just like this one, that imply correlation, but not causation and fail to properly account for confounding factors. Occam's razor implies this is more likely to be yet another study that fails to account for X of these factors.
This doesn't mean the future of science is in jeopardy. It just means more study is required and jumping to alter policy or make decisions based upon these studies, well, requires further study! :P
It's such a typical kneejerk "skeptical" reaction at this point to be like "oh well maybe correlation doesn't imply causation" as if there are no causal relationships ever in life itself.
But really if the rationalists just took a few seconds to think about it maybe they'd find a some plausibility in the notion that exposure to harmful pollution could trigger a stress response; stress is intimately wrapped up with aggression.
"Kneejerk" skepticism is a good thing. The real problem is posting kneejerk incredulous comments before bothering to read all there is to read about the matter, since reading further would often clear up such concerns.
This kind of skepticism isn't actually skepticism. Its internalizing good arguments without understanding them, instead using the phrases associated with these legitimate concerns as magic words for social dominance and denialism.
The typical "correlation doesn't imply causation" retort we see denialists level is a classic example of how people tend to use a phrase to avoid a reason-driven conversation about correlations (of which causal relationships are a subset) and instead shut down the entire topic.
The core of skepticism is inquiry, not denial. And modern Reactionary skepticism (because that is what it has evolved into now: a kind of reactionary culture clique) is all about stopping any inquiries it isn't predisposed to prefer.
I think the term for this is "thought terminating cliche." There may be some of that in this thread, but reading those questions/concerns charitably I think most people are curious, but perhaps a bit too busy/lazy to read the article in full before raising their concerns. The trope that being wrong on the internet is the best way to get somebody serve you the truth on a silver platter may also be in play, though I'd expect to a lesser degree. ("I wonder if they controlled for X" -> "I'll accuse them of not controlling for X, so that somebody will tell me whether they did.")
I don't really think it's productive to pathologise people as "denialists". People probably don't think of themselves using that term, so using it allows people to disassociate the critique from their own personal behavior. I think it's more useful to recognize that some tendency towards low-brow responses exists in all of us to some degree (particularly on sites with tempting point/karma/ranking systems), and we should all strive to do better.
Metagaming the purported hidden motives of people you disagree with is not rational. There is no evidence for your claims about how those you disagree with manipulate social discourse in order to shut down topics.
I'm not talking about intent. In fact, I dont know if they feel like they're being good skeptics or not. What I do know is people engaging in this fashion are not being skeptical no matter how they feel or what they intend to do.
I don't think most people are doing the thing you're implying they're doing. Nobody is trying to shut down the conversation. Most studies do an incredibly poor job of controlling for exogenous variables.
It's pretty rare that I read a paper that has not at least given some thought to the types of obvious correlation/causation concerns that come up after just reading the title of the paper. That said I'm always interested in what comes up in discussion threads here, so I wouldn't want to shut off the skepticism.
That itself is an apocryphal claim, but even if I take it at face value it doesn't really apply to curated data sets that we tend to be exposed to by media outlets with modestly competent science departments. The Economist is certainly not the best out there, but it's very far from an uncurated data set.
I would argue that this kneejerk skepticism and incredulousness are actually two sides of the same coin. Neither demonstrates actual curiosity or desire to learn.
I don't think there is any conflict between skepticism and curiosity. I'm sure most of us have been curious and skeptical of something at the same time. I find that skepticism compliments curiosity, since unbridled enthusiasm can easily corrupt my perception of reality while my desire to perceive reality is derived from my innate curiosity. I aim to never stop learning, and to always remain skeptical. Failing the later, I will fail the former.
I don't understand why technical people waste their time complaining about this stuff. We all have the acumen to read, understand, and debate research.
> I don't understand why technical people waste their time complaining about this stuff. We all have the acumen to read, understand, and debate research.
But to be honest, the reason I read the comments (at least on a place like HN) is to get the key insight / failure without reading the article.
If I had access, I'd try to dig into more depth into how they approached decoupling this from temperature, because that seems like a really strong confounder. I'm also a bit curious about the underlying physical cause of the change in pollution levels, (weekends? summertime?) because that seems like it could also be a confounding common cause.
Alas, I'm not interested in paying $35 for the paper.
from my experience, when science stories that go viral turn out to be wrong it's some secondhand reporting by people misunderstanding or misrepresenting the actual firsthand report
The rational approach to all peer review is to assume there's a flaw and look for it. Quite a few commenters seem bothered by this reflexive reaction to published research, when the entire point of publishing research is to have it picked apart by critical thinkers everywhere.
For as long as journals refuse to publish negative results, researchers will scrounge for some type of positive result, even if it's causally questionable, because to do otherwise is to see years of work go for nought. Until these incentives change, the rest of us are right to be skeptical.
Skepticism is about inquiry, not denialism. You're not practicing skepticism if you make a decision that something is false. The point is to defer decision (or threshold your decision probabilisticly) on the evidence as it becomes available.
Saying "no" to everything is no different than saying "yes" to everything from a logical perspective.
> Saying "no" to everything is no different than saying "yes" to everything from a logical perspective.
This is incorrect. The base rate of true findings when talking about causal models is extremely low. If you say "no" to every published finding, you will be right much more often than you're wrong.
Now, that doesn't mean you should say no to every published finding. But the idea that "yes" and "no" should be equally weighted in your priors across the board is an inaccurate representation of the state of research, and the underlying facts of the universe.
I agree with this, but we're not talking about equal weighting here. We are talking about absolute bias is towards no or yes without any inquiry.
The actual prior distribution of effective to ineffective models is extremely hard to infer from everyday life. We don't have access to unbiased data sources. That's why we should focus on inquiry rather than canned responses.
As an example of how this can be misleading common the average person sees only a tiny fraction of the proposed models that are much more likely to be valid because they've passed far enough along the process of publishing to have received some scrutiny and some credibility in the average case.
I think we probably actually agree. I am also opposed to any sort of inquiry-free canned response. I do think that, if you don't have time or desire to investigate, your baseline should still be that results like this are probably false. Even given the bit of investigation i've done into this study, and even though I think it's decently well done and reasonable, if I had to bet money on it, i'd bet that it's false.
However, I would say that if you don't have the time/desire to investigate, your comments in public forums probably aren't worth listening to. So, I do agree that people that simply respond with "correlation != causality" without reading the study probably shouldn't be doing that.
> As an example of how this can be misleading common the average person sees only a tiny fraction of the proposed models that are much more likely to be valid because they've passed far enough along the process of publishing to have received some scrutiny and some credibility in the average case.
Ya, that's definitely true. The base rate of true models in published research is almost tautologically higher than the base rate of true models amongst all possible models. I was going to say that it is tautological...but I suppose it's actually not. All published models could be false. But I certainly agree that the research process is a pretty good filter, and the base rate of true models is almost certainly higher. But i'd be very very surprised if it was higher than 50%, even if you use a fairly high standard for "published research".
Researcher here (in ML), our field is so full of noisy results due to this issue. Everyone talks about it, but you can't get around the fact that you no longer can get away with a low amount of publications.
The math equivalent of a negative result would be more like "I tried for months to prove this theorem, and it didn't work".
The difference is that in math "here is a proof" is all you need, while in sciences people are typically sharing what amounts to collections of observations. If you publish the collections that lead to interesting conclusions, but not the ones that are boring, people looking at what's been published are misled.
A negative result means simply they didn't find compelling evidence for their hypothesis. Nothing is proven or disproven (although it adds weight to the idea that the hypothesis is not true), so people think it is unexciting. For that and other reasons, the system incentives not publishing these, which is a pretty big problem.
Sure it sounds plausible because it fits the world view of most here. The whole reason we have the scientific method and statistics is to remove personal bias that from the equation because if you don't you get nonscientific crap. The process only works if you use it every time. You can't just say "meh, good enough" because you find the results plausible. This study did not attempt to do a very extensive job controlling for other variables and likewise we can't just accept the conclusion at face value because we find it reasonable, more work needs to be done.
It is precisely the process you describe -- that people are very prone to thinking about something for "a few seconds" and "find[ing] some plausibility in the notion" -- that makes the rationalist scientific method important in the first place.
If we could just think about something momentarily to evaluate its probable truth, we would have no need for experiments, hypotheses, or sorting out correlation from causation at all. We need those things because superficial "think about it for a moment" analyses are very often totally wrong.
Also note that pointing out that this correlation does not imply causation is not at all the same thing as denying that there exists a causative link between the factors. It's simply stating that this evidence, in itself, is not sufficient to establish that. Other evidence well might.
What you did was link to a press release for a paper where they took 53 rats, removed their adrenal glands, injected them with corticosteroids, and saw they were more aggressive.
Few, if any, scientists would say that is a plausible hypothesis explaining the epidemiological results observed above. What you did was speculation with limited data and difficult transferrability (rats with adrenal glands are not representative of people exposed to air pollution).
Check out Robert Sapolsky's stuff on stress, disease and conflict in primates.
Then go and brush up on your research. Short term exposure to pollution is related to heightened inflammatory responses, which are in turn related to aggressive behavior... In humans and animals.
hey I'm not disagreeing with exposure to pollution being related to aggressive behavior. What i'm saying is that there is no reason to believe that in this particular paper, that you can go from the epidemiological linkage results to a physical mechanism. Normally complex epidemiological observations are multifactorial, have multiple causes, and the associations are still only weakly correlated with those causal mechanisms.
The paper doesn't seem to be freely available; there's just the abstract. The first sentence of the abstract mentions "short-term". So I think what they've detected is that in a given locality people get violent when the air is dirty, which is not surprising, of course, but it's interesting to have an estimate of the magnitude of that effect.
One would like to know for comparison what the magnitude of the effect of humidity and temperature is on violent crime. Those numbers are probably in the paper somewhere because the abstract mentions how they had to "address confounding variation between temperature and air pollution". It not being mentioned in the abstract makes me suspect that the effect of pollution on violent crime is probably rather small compared to the effect of weather on violent crime.
(Is there a positive correlation between people stating stupid banalities and violent crime? Probably there is, because people who parrot things like "Correlation does not imply causation" tend to get punched in the face. For the avoidance of doubt, that was a joke, not a veiled threat.)
EDIT: The last sentence of the abstract strikes me as highly questionable (though perhaps there's something in the paper to justify it). Just because there's a short-term correlation between pollution and violent crime doesn't mean that there will be a long-term correlation with a similar magnitude. In other words: if you reduce pollution globally over the next ten years people might just raise their expectations and be just as violent on the days on which pollution is relatively high as they were ten years previously.
Yes violence increases in animals too if the environmental stress (biotic and abiotic) ratchets up. It stands to reason the same applies to humans although we have more conscious control over our responses but we don't often exercise them unless we're aware. These are all subtle mechanisms by which destabilising effects can cascade and proliferate. This is why ecologists and conservationists warned against unrestricted exploitation and an appeal to the assurance of science and technology rescuing us out of the resulting mess, but this is exactly the path we're down on.
The article doesn't talk about mechanisms, but researchers have been looking at the effects of air pollution on the brain for a while, so there are at least some hints in the literature. My favorite specific result to cite in this general direction, which seems potentially relevant here, is that there is evidence that acute exposure to some forms of air pollution blunts or even eliminates the increase in serum brain derived neurotrophic factor (BDNF; basically a kind of growth hormone for the brain) otherwise caused by exercise [1]. Depending on what's mediating that effect, it could be that a whole complex of CNS regulation/recovery/maintenance mechanisms is altered by various forms of air pollution.
Can someone explain what "linked to" means in the context of a research study? Is it just a way to highlight correlation when it is insufficient to assert causation?
"linked to" is used as a synonym for "correlated with" in articles aimed at a layman audience because they think the latter is a word that isn't part of the layman's vocabulary.
It is debatable whether this type of presumed accommodation leads to making the articles more accessible, or just dumbs down colloquial language further in a perpetual spiral to the bottom.
It points out that they look at data for 99% of US counties, which rules out a simplistic attribution of crime to population density, and describes some of the data cleansing. They also claim it corroborated findings elsewhere.
Since this is a data review presumably other centres can repeat the analysis to confirm the correlation result at least, and medical researchers could conceivably find a real experiment to establish causality.
I'm not sure about how much causation is being captured here, but I don't doubt the effect directionally. Here's a paper that came to similar conclusions, which I think captures causality better
Basically it compares performance of students who transferred into schools upwind vs downwind of major highways. Unsurprisingly, those who ended up in downwind schools saw
> decreases in test scores, more behavioral incidents, and more absences
The article itself references many such studies correlating long term exposure to polluted environments with cognitive impairment.
This study however focuses on short term effects (within the same day).
I grew up in a part of the US that some of the most polluted air in the country, back in the 70s and 80s. I recall waking up on a particularly clear day and being shocked by a mountain in the background, not realizing how close we lived to the mountains. I also recall feeling depressed about the continually yellow sky. While there may be a physiological effect of air pollution, I'm certain there is also a strong psychological effect, which could explain the immediate effects described in the article.
From the Abstract - >"We find a robust positive effect of increased air pollution on violent crimes, and specifically assaults, but no relationship between increases in air pollution and property crimes." https://www.sciencedirect.com/science/article/abs/pii/S00950...
This could be data dredging since it seems they may have looked at lots of dimensions and picked the one that correlated. It doesn’t make sense to me the effect of air pollution would be so acute and at such low levels:
The effects are present in and out of the home, at levels well below Ambient Air Pollution Standards
Can someone go into the paper and list out the confounding variables that the authors accounted for?
e.g. weather (temperature and air pressure); day of week; time of year; whether auto-related crime was tagged as "violent" crime; ethnographics/demographics of a community (an aging community will pollute less and be safer) etc etc
While that sounds like a troll comment, I'm going to say that there's merit to this - at least as a class action lawsuit. When you get sick from second hand smoke and it's been shown to be bad for you, there's a case to be made. If people are being psychologically affected by 'second hand pollution' then there's a case to be made that the local or federal government isn't doing their job (which is to serve the people). A lawsuit can be very real if consistent results are shown in multiple studies.
Yes. Lead moreso that other pollutants. Areas with higher concentrations of ground lead (older, poorer parts of cities) are pretty strongly linked with violent crime.
Pretty well documented, don't have the citations handy though. It's been thoroughly discuss on HN.
Interested - what would be causing the high lead levels in those areas, not that lead has been removed from petrol? Is it legacy contaminants or other ways that lead is released into the environment?
I was warned off eating the food I grew in my garden in London as historically people threw their ash into their gardens, so the soil their can be pretty polluted.
This article is paywalled with no workaround. My understanding was that HN did not allow such links. Economist.com links, if they are all hard-paywalled like this, should be blacklisted.
A “workaround” doesn’t mean someone posting a link to Outline or a similar service. It means having a way to access the content on the site itself. Economist.com links should be blacklisted, because they have no workaround. If Outline is required to read a given article, then the main link on the submission should go to Outline.
You seem to be stickler for the rules, so I can’t understand your position on this.
If that is the case, why aren't major cities in China, or other major Asian cities like Singapore bastions of crime? Hint: they aren't. They are in fact some of the safest places in the world, and on average have air pollution levels much higher than cities in the US, even dangerous ones!
And yet look at all of the misinformation in the comments, as usual.
That's like saying that there's undocumented crime on the moon. You don't know that. But you have a theory that you most likely need more funding to look into. ;) On the other hand, if you have documentation to the contrary, I'll of course retract my own claim.
Reminds me of the organic food movement and # of autism correlation which is useful vs antivax arguments (as normally antivax people are heavy into organic [not that theres anything wrong with organic, yum!])
Somebody posts this every single time any statistical study is published, regardless of whether the study actually contains a claim or causation of not.
> Correlation is not causation of course (there may, for example, be a third variable affecting both pollution and crime) and the authors are cautious not to speculate about the precise mechanism by which contaminated air might lead to more rapes or robberies.
From the paper's abstract, though, "The results suggest that a 10% reduction in daily PM2.5 and ozone could save $1.4 billion in crime costs per year, a previously overlooked cost associated with pollution." This statement sounds like it's implying a causal relationship to me.
> Correlation is not causation of course (there may, for example, be a third variable affecting both pollution and crime) and the authors are cautious not to speculate about the precise mechanism by which contaminated air might lead to more rapes or robberies.
I'd like to assume that all the "easy" facors are already taken into account (population density, property value, education level, income level, weather, etc).
Even if it's just a correlation, it's not really a result worth publishing otherwise.
I do not get why it got downvoted. Maybe I should have elaborate more. I skimmed the paper and it does not seem to be the case. They only take into account weather.
It just doesn’t make sense intuitively to me. This feels more like an example to show correlation does not equal causation. So what if the violent crime is higher in highly polluted areas? How is that information useful for anything?
I'm not a biologist or any sort of doctor so this is just sort of a hot take stab in the dark. But here's my go.
It's all about oxygen.
If it's harder to breathe, less oxygen is getting to your brain, and that's never good. While it may not be life-threatening or anything, it might make you more irritable. Just constantly slightly fatigued and cranky. It takes less to set you off.
Another correlation, is it coincidence as in urban city usually has high pollution and high crime or a real link like pollution causes brain to go violent?
That's one dishonest headline. Air pollution areas are commonly associated with big cities, where the crime rate tends to be higher. The website clearly turned this into something else for no reason at all other than clicks.
Ironically, your accusation of them doing this for "the clicks" prevented you from clicking the link and finding out that your criticism is not relevant to this study.
So they are not comparing clean air neighborhoods to highly polluted neighborhoods, but the same neighborhood on different days.
https://www.sciencedaily.com/releases/2019/10/191003114007.h...