Doublethink: Two incompatible positions. S. Stanley Young, PhD - - PDF document

Slide 1

Doublethink: Two incompatible positions.

• S. Stanley Young, PhD

FASA, FAAAS stan.young@omicsoft.com

1

Doublethink is holding two different beliefs that are contradictory, perhaps in different

• situations. The beliefs may be firmly held as with some true believer, or they could be dishonest

for public and private consumption. When looking at an argument in support of the two positions, does one favor emotional and the other rational, perhaps self-serving, thinking? Or is there some combination of rational and emotional?

Slide 2

Decision making

• Appeal to authority.
• Appeal to emotion.
• Appeal to data/analysis.

2

We don’t have time to carefully evaluate everything in life so we often take shortcuts. If a good authority takes a position, it is fast and convenient to just follow along without much thought. Emotion can be used to force a decision without much though. Through much of human history authority and emotion were the primary modes of decision making. Logic/data/analysis are late comers to decision making. My question, mostly, is How dangerous is air pollution? Are we to believer what the EPA told a congressional committee? It can kill you right now! Air pollution is causal of acute deaths. By improving air quality we have saved 160,000 premature deaths each year.

Slide 3

NEJM 1993, EHP 1995

3

EPA chose Dockery (McCarthy, Nichols, Beale).

Where did the emphasis on particulate matter start? Two papers appeared in the mid 1990s. Dockery et al. (1993) claimed PM2.5 was the toxic component of the air. Styer et al. (1995) said they say no association of PM10 with mortality. The EPA funded both studies. NB: PM2.5 and PM10 contains PM2.5 and the two are highly correlated. The Styer data set is much larger than the Dockery data set. The EPA chose to go with Dockery. The decision was (likely) made by McCarthy, Nichols and Beale. McCarthy now heads the EPA. Nichols heads the California Air Resources Board. John Beale was lead policy person for water and air at EPA for many years. They were “rain makers” for the EPA. John Beale is now in jail for massive fraud. The EPA does not have the Dockery data set and Harvard refuses to make the data set public.

Slide 4

Journal of Risk and Uncertainty 2003

4

The Clean Air Act of 1970 and Adult Mortality

KENNETH CHAY CARLOS DOBKIN MICHAEL GREENSTONE UC Berkeley UC Berkeley University of Chicago cdobkin@ucsc.edu

“We find that regulatory status is associated with large reductions in TSPs pollution but has little association with reductions in either adult or elderly mortality.” Dobkin found and made this data set available to Young and Obenchain. Their re-analysis confirmed no association.

Slide 5

Journal of the American Statistical Association 2011

5

“Based on the local coefficient alone, we are not able to demonstrate any change in life expectancy for a reduction in PM2.5.”

This study was funded by USEPA. Greven et al. also say, that large effects noted from location to location are most likely due to confounding. Confounding occurs when two or more factors

• ccur together in a way that makes statistical separation difficult or impossible. Crazy example.

Ice cream consumption increases in the summer. Does ice cream consumption cause heat stroke? Obviously not. But if we measure ice cream consumption and not temperature, we might think so due to the association. The Greven et al. data set is not public. A letter from Dominici to EPA “walked back” the claim in the paper. Without the data, it is impossible to know which interpretation makes the most

• sense. The JASA paper was peer reviewed. Greven et al. have not retracted their paper. ??

Slide 6

BMJ Heart 2014

6

“This study found no clear evidence for pollution effects on STEMIs and stroke, ….”

Virtually all air quality/mortality papers point to heart attacks as the etiology for excess mortality due to poor air quality. This paper, using a massive data set, all of England and Wales, removes heart attacks and stroke as a possible etiology for air quality induced mortality. On the face of it, it seems absurd that very low levels of small particles can CAUSE heart attacks

• r stroke. Smoking one cigarette gives a dose many orders of magnitude higher than current air
• levels. Linking know substances in particle matter to specific fatal mechanisms of heart attack or

stroke has been very difficult. In epidemiology studies we are looking at statistical deaths. There are no autopsies.

Slide 7

BMJ Heart, Figure 2

7

• Orientation. Y-axis is % increase or decrease of mortality. X-axis list air quality components,

carbon monoxide, NO2, etc. A dot gives the mean change and the vertical lines give the 95% confidence limits. If the confidence limits do not overlap 0 then there is nominal statistical significance. Only 3/66, 4.5%, confidence limits fail to cross the no effect line, a result consistent with chance, 5%. There are no effects for all cause mortality, MI or stroke the primary response variables. We appear to be looking at purely random effects.

Slide 8

BMJ Open 2015

8

“…indicated that none of the air pollutants investigated —CO, NO, NO2,O3 and particulate matter (PM2.5)— showed consistent positive associations with increased risk

• f AMI hospitalisation.”

Canadian study found no effect of air components. This paper supports Milojevic et al. (2014).

Slide 9

Supporting Literature

1.1995 Styer No effect in two US counties. 2.2000 HEI repport. No effect in CA. 3.2003 Chay. No effect US. 4.2005 Enstrom. No chronic effect CA. 5.2011 Greven. No local effect US. 6.2013 Cox. Only temperature effect.

• 7. 2013 Young. No effect in West US.

8.2014 Milojevic. No heart attacks or stroke. 9.2014&2015 Young. No acute effect in CA

9

As early as 2000 it was reported that there were no excess deaths in California (a report funded by EPA). In 2003 Chay et al. reported that reductions in air pollution did not produce a reduction in deaths across the US. I’m happy to provide pdfs of these papers. These papers have been largely ignored by EPA and are usually not cited by EPA funded

• researchers. Not citing Enstrom 2005 is particularly egregious. The study is large and its result

calls into question causality. For example Dockery et al. (1993) has over 6,000 citations and Styer et al. (1995) has just over

• 100. The Styer data set is larger. The statistical methods are sound. The paper appeared in

Environmental Health Perspectives. Dockery used PM2.5 and Styer used PM10. PM2.5 is a component of PM10 and the two are highly correlated. Chay is cited ~100 times. It uses Total Solids Particulate, TSP. TSP contains PM10 and PM2.5. The three are highly correlated. For example, Integrated Science Assessment for Particulate Matter 2009 was searched for Styer and Chay and neither name was found.

Slide 10

Decision making

• Appeal to authority.
• Appeal to emotio

ion.

• Appeal to data/analysis.

10

If appeal to authority fails, then appeal to emotion is often effective. The emotions are complex but those of anger, fear, anxiety, etc. can cloud good judgement on the part of the decision

• maker. Induced emotions can cloud the judgement of individuals.

Slide 11

11

The Navy pilots of these F4B-4's are just having a little fun with the local farmer who is none too

• pleased. Scared cows don't produce milk.

EPA is just having a little fun trying to scare the public – maybe to their own advantage. They are true believers and anything goes, even unethical human experiments, or they are dishonest knowing that current levels of air quality are not causally related to death? Are they flimflamming everyone?

Slide 12

Emotion

The whole aim of practical politics is to keep the populace alarmed (and hence clamorous to be led to safety) by menacing it with an endless series of hobgoblins, all of them imaginary.

• H. L. Mencken

Global Cooling/ ……./Sustainability/Air Quality/etc.

12

Humans depend on emotional respond to dangerous situations. But emotion can be exploited.

• H. L. Mencken: “The whole aim of practical politics is to keep the populace alarmed (and hence

clamorous to be led to safety) by menacing it with an endless series of hobgoblins, all of them imaginary.” The medieval church used scare/emotion to control people. Global cooling, etc. and air quality kills are all likely imaginary.

Slide 13

Decision making

• Appeal to authority.
• Appeal to emotion.
• Gather data and do analysis.

13

The most difficult and time consuming method of decision making is to gather data and do analysis

Slide 14

Gather Data (Facts)

California, 8 air basins. Years 2000 – 2012, most recent. Daily deaths, air quality, weather. Over 37,000 exposure days. Over 2 million e-death certificates.

14

We secured an e data set from California of over 2M death certificates for the years 2000-2012. We looked at the 8 most populous air basins. We obtained air quality, PM2.5 and ozone, and weather variables, min and max temperature as well as relative humidity. We have over 37,000 days of data. This is one of, if not the largest, air quality/acute mortality data sets extant. It is clearly the largest publicly available data sets. This analysis was reported http://arxiv.org/abs/1502.03062, Feb 10, 2015. We made the data set available. The EPA knows or should know about the paper and the data set.

Slide 15

Do Analysis

1.Case Crossover / Moving Median 2.Time Series Regression 3.Leave-one-year-out and predict it.

15

Law is used to work out arrangements among people/legal entities. In science we use statistical methods to extract claims from data. The case crossover method looks at narrow time slices of data to try and equalize risk. We use a 21-day moving median and contrast the mid-day to other days in the interval. Time series regression is a complex regression technology for examination

• f a time series. With large, complex data sets cross validation, leave one year out and predict

it, is used to evaluate the reliability of claims.

Slide 16

Methods / Case Crossover

1.Remove seasonal trend. 2.Deviation from trend. 3.Cross correlate deaths with PM2.5 and ozone.

16

In a case crossover study, only cases are examine. Seasonal trend is removed. The gap between the deaths on the day of interest and the seasonal trend is a measure of the force of mortality

• n that day. We compute a similar gap for variables that might induce a change in the mortality

gap, PM2.5, ozone, min temp, max temp, relative humidity. We examine the correlations between these gaps. If the death gap correlates with the PM2.5 gap, we say there is an

• association. We also examine lags or 0, 1, and 2 days.

The case crossover method was invented in 1991 and has been used extensively in air pollution studies.

Slide 17

Mortality and Ozone

17

A B C D

The X-axis is day of year, DOY, 1,2, ….,365. Each dot is one day. A and B are all cause deaths for people 65 and older. A gives the raw data. We see that there is a strong annual effect. Deaths are higher in winter and lower in summer. We remove the seasonal trend from A to give B. C and D give ozone levels. In C we see that ozone is lower in winter, when deaths are higher, and higher in summer when deaths are lower. In D we give the seasonally adjusted ozone levels. We see that during the winter that ozone levels are more stable. The data in B and D can be used to examine the correlation between mortality (deviation from seasonal trend) and ozone (deviation from seasonal trend). If ozone “causes” increased acute mortality, then there should be a positive correlation.

Slide 18

Results : Case Crossover

18

We plot the deviation of daily deaths against the deviation of air quality, PM2.5 and ozone. Each dot represents a single day. IF there were an effect we would see the points sweeping from lower left to upper right. Clearly there is no obvious effect. These figures were computed for each air of the 8 basins and for lags of 0, 1, and 2 days. We see nothing. There are a total of 2 (air components) x 3 (lags) x 8 (air basins) = 96 tests.

Slide 19

Methods : Time Series Regression

1.Build variables for response, time lags, weather. 2.Compute regressions, testing various lags. 3.Compute for each air basin, consistency/replication? 4.Compute for NMMAPS data set (replication).

19

Time series regression methods are complex and are largely of interest to technical people. The details are given in a technical report which can be found at https://arxiv.org/abs/1502.03062. We applied the methods used on the large California data set to the California data from the NMMAPS study and we replicated our findings : there was no detectable effect in the California NMMAPS data.

Slide 20

Results : Time Series Regression, PM2.5

20

Each dot represents a day. The y axis represents deaths, below and above model prediction. The x axis gives PM2.5 exposure. The solid black line is no effect. The red line is a spline fit of the

• data. The dashed red line give confidence bounds for the line. As the confidence bounds include

no effect, there is no relationship of deaths to PM2.5. The no relationship continues down to the lowest observed levels of PM2.5.

Slide 21

Results: Time Series Regression, Ozone

21

Each dot represents a day. The y axis represents deaths, below and above model prediction. The x axis gives ozone exposure. The solid black line is no effect. The red line is a spline fit of the

• data. The dashed red line give confidence bounds for the line. As the confidence bounds include

no effect, there is no relationship of deaths to ozone. The no relationship continues down to the lowest observed levels of ozone.

Slide 22

Results : Leave one year out regression

22

We fit over 78,000 models to the data. Some are represented here. The red band gives an

• verlay of prediction models for the year left out. The reason the red line is thick is that multiple

models are essentially on top of one another. All the models essentially explain the data equally

• well. The model using only time is as good as any. Also note that the models capture only a very

small part of the daily death variability. We exhaustively tried to find an effect of PM2.5 or ozone, but we could find no effect that was consistently better than day of year.

Slide 23

Facts-Law / Data-Analysis

Fact : New, large, high-quality data set. Law : Best statistical methods. We find no association of air quality with mortality. There is no need to appeal to emotion.

23

There is a lawyer joke: In a case, if the facts are on your side use them. Absent facts, appeal to

• law. If both the facts and law are against you, appeal to emotion.

We have facts: a new, large high-quality data set of acute deaths in California. We use the best statistical methods to examine the data: Case control, Time series regression. And we use cross validation to evaluate the reliability of our data and methods. We find no association. I think EPA knows there is no association. There is need to appeal to emotion.

Slide 24

Summary (data and methods)

The empirical evidence is that current levels of air quality,

• zone and PM2.5,

are not causally related to acute deaths for California. Young, Lopiano, Smith

24

The empirical evidence is that current levels of air quality, ozone and PM2.5, are not causally related to acute deaths for California. Young, Lopiano, Smith (arXiv 2015).

Slide 25

My current position/hypothesis/theory.

25

1.There is no acute or chronic effect of PM2.5 or ozone on deaths in California (fact).

• 2. Any effects observed in rest of US are due to other factors.
• 3. There is new literature supporting no effect of PM2.5 and ozone.

There are no excess acute or chronic deaths in California. Fact. It is a complex argument (given by Greven et al. (2011) that any deaths seen are most likely due to other factors than air quality, confounding variables. The Greven work was funded by EPA so they know the results. New results given in BMJ Open and BMJ Heart find no effects. The EPA knows there is no effect of current air quality, or they should know. There was and is no reason for human experiments other than a human “fishing trip”.

Slide 26

Contact Information stan.young@omicsoft.com

26

You can email me with questions at the above email address.