BackgroundThis post comes out of one point I raised in the Partisan Snark blurb, which further evolved in discussion over at Lucia's in the current open thread. My leading argument was my usual: nuclear fission has been historically less hazardous than coal-fired electricity generation. Using statistics I've bookmarked at the ready, the worldwide mortality rate is fully two orders of magnitude different. That's using the worst-case mortality estimate from nuclear against the best-case coal statistic.
Side note: Brandon Shollenberger finds that the Forbes article I so often cite has been silently changing the stats over time. Not kewl.
My arguments have long rested on noting that while there's certainly slop in both estimates, a two-order of magnitude of difference leaves a room for a lot of slop. A 95% confidence interval is 1.96 standard deviations under a Gaussian normal distribution. Must I really do a significance test when the lower bound of the higher risk factor is 100 times larger than the upper bound of the smaller risk factor?
Maybe I do.
Conclusions FirstIn comments here, Joshua turned me on to an interesting paper, Epstein et al. (2011), Full cost accounting for the life cycle of coal. One passage gives a good overview of the issues I've been recently discussing:
Epidemiology of air pollution.I don't have the reference for the Schwartz Harvard Six Cities study handy, however I have been referencing a follow-up study, Laden et al. (2006), Reduction in Fine Particulate Air Pollution and Mortality, of which Schwartz is a co-author.
Estimates of nonfatal health endpoints from coal-related pollutants vary, but are substantial—including 2,800 from lung cancer, 38,200 nonfatal heart attacks and tens of thousands of emergency room visits, hospitalizations, and lost work days. 85 A review 83 of the epidemiology of airborne particles documented that exposure to PM 2.5 is linked with all-cause premature mortality, cardiovascular and cardiopulmonary mortality, as well as respiratory illnesses, hospitalizations, respiratory and lung function symptoms, and school absences. Those exposed to a higher concentration of PM 2.5 were at higher risk. 86 Particulates are a cause of lung and heart disease, and premature death, 83 and increase hospitalization costs. Diabetes mellitus enhances the health impacts of particulates 87 and has been implicated in sudden infant death syndrome. 88 Pollution from two older coal-fired power plants in the U.S. Northeast was linked to approximately 70 deaths, tens of thousands of asthma attacks, and hundreds of thousands of episodes of upper respiratory illnesses annually. 89
A reanalysis of a large U.S. cohort study on the health effects of air pollution, the Harvard Six Cities Study, by Schwartz et al. 90 used year-to-year changes in PM 2.5 concentrations instead of assigning each city a constant PM 2.5 concentration. To construct one composite estimate for mortality risk from PM 2.5 , the reanalysis also allowed for yearly lags in mortality effects from exposure to PM 2.5 , and revealed that the relative risk of mortality increases by 1.1 per 10 μg/m 3 increase in PM 2.5 the year of death, but just 1.025 per 10 μg/m 3 increase in PM 2.5 the year before death. This indicates that most of the increase in risk of mortality from PM 2.5 exposure occurs in the same year as the exposure. The reanalysis also found little evidence for a threshold, meaning that there may be no “safe” levels of PM 2.5 and that all levels of PM 2.5 pose a risk to human health. 91
Thus, prevention strategies should be focused on continuous reduction of PM 2.5 rather than on peak days, and that air quality improvements will have effect almost immediately upon implementation. The U.S. EPA annual particulate concentration standard is set at 15.0 μg/m 3, arguing that there is no evidence for harm below this level. 92 The results of the Schwartz et al. 90 study directly contradict this line of reasoning.
The yellow highlight perfectly fits what I have already been arguing: that reduction will have almost immediate benefits. The detail of continuous reduction vs. peak days is a new one for me, filed for future reference. That Schwartz finds evidence of harm below the EPA concentration standard is a very interesting finding, very much filed for future reference.
The balance of this article is a bit of a ramble, more notes of my travels and readings. I don't find what I'm looking for, a probability distribution of estimated all-cause mortality from PM pollution, either due to coal specifically or all sources. The reason I'm looking for one is to be able to provide a less uncertain range of US mortality rates due to particulates than what I've been using (10,000 - 36,000 premature deaths/yr in the US due to coal particulates).
Should any reader know where I can find such a thing, I would appreciate the reference.
Some Standard ObjectionsIt seems obvious to me that breathing combustion products in "large" concentrations for extended periods of time is not conducive to healthy living. Start talking about that in a forum frequented by climate contarians, and it's often not clear that we agree on what's obvious. I paraphrase some comments from the thread at Lucia's:
- Suspicions that studies have used models to extrapolate small, unrepresentative samples as part of a political/ideological campaign against the coal industry.
- Researchers don't understand the models they're using, producing unreliable results which are then leveraged to scare the public into falsely believing that burning coal is dangerous.
- Models haven't been ground-truthed, implication being that researchers are careless and/or trying to hoodwink the public over a phantom menace.
- Models are required for obtaining reliable results. They're also required for obtaining unreliable results.
- Any data will yield up any answer one desires if one looks for it hard enough.
They're cause for healthy skepticism, to be sure. But lacking substantive evidence, easily dismissed. Or to put it another way, unless evidence that a specific study is somehow flawed or tainted, I've no duty to "prove" otherwise. Most of the crew at Lucia's seemed to agree with that when challenged.
An Argument With Some Legs... courtesy of Brandon Shollenberger. In his own words:
[...I]f the individual studies being used don’t come close to capturing the actual uncertainty, then it is difficult to justify trusting any range one might come up with. To a lay reader who hasn’t studied the issue, seeing one study give a result of 400,000 +/- 50,000 and another study give a result of 1,600,000 +/- 100,000 doesn’t make me think the actual range should be considered 400,000 – 1,600,000 (or 350,000 – 1,700,000 if you prefer).This was his second reply to me. My first one took the knee-jerk, "c'mon, enough with the standard objections" tack because a) it's Shollenberger and b) I'd already been getting similar pushback from others on the thread.
If a collection of studies don’t come remotely close to giving similar results, there’s no particular reason to think they’ve hit the high and low ends of the actual range. When you get very disparate results, it is quite possible the true value is lower (or higher) than any of them.
[...] I don’t think there’s any doubt nuclear electricity generation causes fewer deaths than the use of coal. I just don’t have any confidence in the numbers given for coal. The qualitative comparison seems safe to me, but any quantitative one seems quite shaky.
The second highlight is the thing which got through to me, causing me to consider having a closer look at his reservations. It also inspired the title and subtitle of this article. The first highlight is a reasonable point, and I expand on it in my response:
Agreed, especially if only two data points establish the range. In this case, the upper and lower bounds I cite differ by nearly a factor of four, so it’s tempting to see those as the “true” min/max. OTOH, that large a difference is cause to question reliability. Ideally we’d like to have multiple points clustering around some mean value, and use a PDF on the distribution to establish a CI for the range. Ye olde meta-analysis. (Cue the cautionary tales of publication bias, etc.)Before delving into it, I had two questions of relevance for Brandon:
I’ve no idea if such a thing has been done for coal specific studies. More likely it’s been done for particulates from any source. Onus probably on me to scratch around for one since this topic is more or less my baby.
1) What would tighter estimates for coal allow us to do that we can’t get from the current estimates?I have my own own answers to those questions, which part of why I'm writing this post. Unfortunately, I'm not finding them. But I did come up with some useful background and other tidbits of interest. I'll leave answering to the above question for another time, perhaps as a footnote or an update if not a full article.
2) Based on the present state of the coal estimates, what course of action would you recommend?
What's So Terrible About Particulates?... and don't you know that coal isn't their only source?
Yes, yes I do.
The US EPA provides some answers to the first question in this handy reference card. To sum up, the primary risks of premature deaths from particulate pollution are:
• AsthmaWhat? No cancer? Doesn't everything cause cancer?
• Bronchitis (acute or chronic)
• Premature aging of the lungs
• Coronary artery disease
• Abnormal heart rhythms
• Congestive heart failure
Wikipedia has an article on particulates, with a section devoted to health effects:
Health problemsSo there's your cancer. Happy now? Good, so am I.
The effects of inhaling particulate matter that have been widely studied in humans and animals include asthma, lung cancer, cardiovascular disease, respiratory diseases, premature delivery, birth defects, and premature death.
Increased levels of fine particles in the air as a result of anthropogenic particulate air pollution "is consistently and independently related to the most serious effects, including lung cancer and other cardiopulmonary mortality." The large number of deaths and other health problems associated with particulate pollution was first demonstrated in the early 1970s and has been reproduced many times since. PM pollution is estimated to cause 22,000–52,000 deaths per year in the United States (from 2000) contributed to ~370,000 premature deaths in Europe during 2005. and 3.22 million deaths globally in 2010 per the global burden of disease collaboration.
Note the estimated range for US deaths/yr for all PM pollution, vs. what I've been citing for coal: 10,000-36,000 for coal, vs. 22,000-52,000 total. I make two comparisons:
- Comparing lower to lower bounds and upper to upper, the estimate would be that coal causes 45-70% of all PM mortality.
- The high estimate for coal differs from the low estimate of same by a factor of 3.6. For all PM mortality, the high/low ratio is 2.4.
To provide more perspective, from the US National Center for Health Statistics:
Number of deaths: 2,596,993So in terms of percentages of all deaths, we get 0.39-1.39% for coal and 0.85-2.00% for PM pollution from all sources.
Death rate: 821.5 deaths per 100,000 population
Life expectancy: 78.8 years
Infant Mortality rate: 5.96 deaths per 1,000 live births
While I'm there, I may as well show this:
Number of deaths for leading causes of death:That's not 100%, but close. Yet, here's what grabs most of the news:
Heart disease: 611,105
Chronic lower respiratory diseases: 149,205
Accidents (unintentional injuries): 130,557
Stroke (cerebrovascular diseases): 128,978
Alzheimer's disease: 84,767
Influenza and Pneumonia: 56,979
Nephritis, nephrotic syndrome, and nephrosis: 47,112
Intentional self-harm (suicide): 41,149
All homicidesSo switching to nuclear power from coal would save about as many lives as banning guns
Number of deaths: 16,121
Deaths per 100,000 population: 5.1
Number of deaths: 11,208
Deaths per 100,000 population: 3.5
This pisses me off. Perhaps the less said, the better.
On the other hand, the number of potential years of life lost due to gun violence (or any murder for that matter) is probably much higher on a per death basis than potential years lost due to air pollution. That key point often gets shuffled under in these conversations because mortality rates are most commonly given as deaths per unit time or per capita. (h/t Steven Mosher for noting this technical point at Lucia's.)
Particulate Matter Mortality Meta AnalysisSay that ten times real fast. Or do what I did, Google it, and use Search Tools to narrow down the hits to results posted within the past year. Down the hitlist (the first several are too specific) is a bingo:
Hwashin et al. (2015), Meta-Analysis Methods to Estimate the Shape and Uncertainty in the Association Between Long-Term Exposure to Ambient Fine Particulate Matter and Cause-Specific Mortality Over the Global Concentration Range
As a bonus, it's open access. The abstract lays out a problem:
Estimates of excess mortality associated with exposure to ambient concentrations of fine particulate matter have been obtained from either a single cohort study or pooling information from a small number of studies. However, standard frequentist methods of pooling are known to underestimate statistical uncertainty in the true risk distribution when the number of studies pooled is small. Alternatively, Bayesian pooling methods using noninformative priors yield unrealistically large amounts of uncertainty in this case.Don't ask me to explain that in detail with any kind of authority. It's also a methods paper, quite interesting, but not what I'm looking for. Here's another one, also open access:
Fengying et al. (2016), Spatiotemporal patterns of particulate matter (PM) and associations between PM and mortality in Shenzhen, China
Still not what I'm looking for, but the body text has some interesting stuff:
BackgroundAdverse effects increasing with exposure is a no-brainer, but of course we want to quantify it. Not surprising that a low-end threshold hasn't been established, nor that low-pollution areas in China (if not most everywhere) have not been studied.
Airborne particulate matter (PM) consistently associated with adverse health effects at current levels of exposure in urban populations [1–4]. Air pollution has serious direct and indirect effects on public health in China [2, 5–8]. PM with aerodynamic diameters less than 2.5 μm (PM2.5) has become the fourth prominent threat to the health of Chinese people .
The range of adverse health effects of air pollution is broad [2, 10, 11]. Susceptibility to pollution may vary depending on overall health condition and age [5, 6, 12–14]. Risk of various effects has been shown to increase with exposure, but there is little evidence to suggest a threshold below which no adverse health effects can be anticipated [15, 16]. The lowest concentration at which such effects begin to manifest is not much greater than the background concentration, which has been estimated at 3–5 μg/m3 for PM2.5 in the United States and western Europe . Most studies on air pollution exposure and its effects on human health in China have focused on heavily polluted cities or mega-cities [8, 17–19], whereas studies on cities with relatively low air pollutant concentrations are rare.
It seems reasonable to question whether global PM mortality rates have thus been overstated due to under-sampling in more rural areas, and the fact that it's likely inherently difficult to establish a lower exposure threshold.
About here is where I ran out of gas in the Google search. And as there's already more than enough to chew on in this note, I'll end here. Whatever else I find will need to be taken up another time.