p < .05, No One Reads the Science
31st October 2011
Time for Serious Science on climate change. We’ve got Nobel prize winning physicists releasing their science at a dot org web site. They will do what no one has done before. They will do Nobel level science on climate change data. Here’s (pdf) their Special Sauce.
B) the application of a quality control and “correction” framework to deal with erroneous, biased, and questionable data . . .
If you know anything about applied statistics you read that line with a chuckle. You can drive a truck through all the holes that process permits and anyone dumb enough to accept the premise will be trapped with the conclusions you manufacture. You just have to get them to accept that dubious assertion of correcting biased data which, as persuasion science teaches us, is easier when you have a Nobel prize, a dot org web address, and a press kit. The rest is just sophistical statistics. Think about this like a real scientist rather than somebody with a little time, a new web address, and a desire to debate.
You cannot execute experimental science on climate change. No one is in a position to randomly assign anything to any controlled and compared conditions. The nature of nature does not permit this gold standard scientific method.
The next best method is an observational approach that is most similar to well-done public opinion survey research. You create a well defined population, then draw random samples from that population with reliable and valid measurement of each element in the sample. Gallup, for famous example, begins with a population of all registered voters, then works like a dog to draw random samples of 1,400 people and asks each the same question under the same conditions. And, working hard with this kind of method, scientific pollsters still get the outcome wrong – remember 2000? The election was simply too close to discern with this good, but not ultimate, method.
Apply this thinking now with climate change. Throw a fishing net over Earth then within each opening of the net shoot thermometers into randomly selected locations in each opening. Now take a reading from each thermometer in all locations the same way over a long period of time.
While this fishing net approach is clearly not experimental, it is pretty good science and would be better than licking your finger and sticking it in the wind. You’d still be wary and skeptical with any results you obtain with this method, but you could build a good research literature that you might be able to tie into experimental results from lab or field settings. Then, taken together, you might not only know something, but also be able to change something. Good climate science can be done. Good public policy on climate change can be done.
But realize that the Berkeley Earth project does not have this kind of science. No one has yet to throw that net around Earth, shoot thermometer-arrows into randomly selected locations within each net hole, and then collect the same observation the same way at all randomly sampled locations. The only data anyone has on climate is from biased samples where the location of the hole is biased, the thermometer location is biased, the thermometer type is both biased and variable, and the data analysis is biased, variable, and not even statistically significant! The Berkeley Earth project blows past those limitations with this notation.
Robert Rohde, our chief scientist, obtained more than 1.6 billion measurements from more than 39,000 temperature stations around the world.
Nearly two billion data points! Imagine the size of that Excel spreadsheet! And nearly forty thousand thermometers! And, excuse me I didn’t quote this but, they have two hundred years of data!
Of course, the Berkeley Earth project did not obtain two billion data points from their thermometers. No new data here. They took old data with all the biases of sampling, measurement, and time, and invented some kind of Nobel applied statistics that takes a biased database of two billion data points and makes it True. Let me requote them on this.
B) the application of a quality control and “correction” framework to deal with erroneous, biased, and questionable data . . .
It is impossible to de-bias a biased sample of any data no matter what size the database. Prove this without any math. If anyone knew how to reliably and validly de-bias a biased sample he would be the Queen of Tomorrow who could know not only the outcomes with climate, but with the stock market, Presidential elections, and which person at the end of the bar would actually do this but not that on the first date. Only God knows how to de-bias a biased sample and because no one since Nietzsche believes in God, the claim must be ipso facto, summa cum laude, and dipsy doodle false.
If you are willing to work a bit harder than this Thought Problem approach, you need to peruse this paper and look carefully for all the “correcting” their new analytic method requires. The team moved between hypothesis and analysis as they derived their results. They did not start with one explicit system of “correction” and then run the data to see if it fit. Rather, they made some corrections, ran the data, compared the data to some model, then made more corrections, ran the data, compared the data to a model, and ran the this process until they got bored or ran out of corrections. You also have to remember that with the tremendous sample size, you can obtain results that have stupendous markers of statistical precision, yet may not explain that much or that well.
Realize also that this analysis is essentially a time series run on a dependent variable with no serious scientific independent variables like CO2 concentration or solar flare activity or even the number of slides in Al Gore’s Nobel prize winning PowerPoint presentation. It simply takes biased measurements over 200 years and tries to find a higher mean and a wider standard deviation during some point in the sequence.
Finally, you need to spot what I consider the weakest element of the Berkeley Earth project. They seem to be working towards confirmation rather than disconfirmation. In other words, the cycle of model-correct-test seeks to confirm the global warming hypothesis rather than disconfirm it. Merely taking a huge, but biased database, then applying various de-biasing tricks is not the same thing as actively trying to disconfirm the hypothesis.
We cannot do experimental science on climate change. We can do good observational science on it. But none of the data collected to date comes close to even acceptable observational research (or as Sheldon Cooper might put it, the slower, younger brother of experimental science). This is sophistical statistics where muggles correct questionable data with numerical magic.
Now, we could do better science with this, but who wants to pay the bucks to take the Earth’s temperature? Seriously, why does any rational person think that a single value for temperature can represent any meaningful knowledge about global climate? It would be like haphazardly sampling your heart rate with different methods throughout the day, almost every day, computing the average value each day you’ve got data, then assuming that one averaged number says anything about the state of your body. Think about that. The exercise is stupid on the face of it, yet people are trying to make that one crazily measured number into something meaningful.
Given the Bad Science here, we can now look for the Persuasion because with Bad Science, Persuasion is all you’ve got.
Start first with the communications effort from Berkeley Earth. The team is announcing their work to the public at large and making it available on a website with a dot org address. They describe their process and make available both their data and their reports. Nothing on the site to date is peer reviewed and published.
So what?
First, science don’t need no stinkin’ press releases and it is almost a Law of Persuasive Indication when you see scientists with a press kit.
Second, making data and reports available before publication is an invitation to debate. The normal course of scientific communication is to publish through peer review and then do something else with it.
Third, note that the team’s edu website affiliation is not the host for this information, but rather they purchased their own web hosting account and bought this particular domain name for their own purposes. And, they got that money from various interested sources who paid them to do this work.
If I’m doing science, I don’t give a rip about the press, the public, or my mom’s approval; I only care about peer review. Yet the Berkeley Earth project wants to weigh in publicly before obtaining any scientific opinion. They are taking it to the streets in a more polite and civilized form of Occupy Wall Street rather than working within peer review. Contrast the Berkeley Earth project with the team dealing with the Speed Of Light anomaly. That group of physicists spent three years showing their data to other physicists and only after that has issued a public communication about their data in a peer review journal where that data is now under standard scientific study. The Berkeley Earth team is not doing the same kind of scientific communication as the Speed Of Light team. Think about that.
Now look at this editorial. Richard Muller offers a summary of the team’s effort in, of all places, the Wall Street Journal. You need to read the whole thing because it is an excellent example of rhetorical science. Take ten minutes right now. I’ll wait . . .
Jeepers. Do you see the fabulous argument structure with Muller’s editorial. He begins by making the anti-change case, citing the counter-arguments scientific skeptics have raised with the climate change claim. Muller takes 1,025 words in the editorial and 511 of the words present the anti-change arguments. He then refutes those doubts in 311 words with the magic of de-biasing biased databases which arises when you have 2 billion data points, a Nobel prize, and a dot org website.
He thus avoids the classic persuasion error of the Climate Change Chorus – demonizing the opposition. Muller addresses skepticism with seriousness and politeness. He presents the opposition arguments correctly and without insult, implied or direct. He then counters the concerns with
deep looks . . . detailed papers . . . more scrutiny . . . esteemed scientists and statisticians . . . new analytic approach.
He simply asserts his team has refuted all the scientific concerns without providing anything remotely like a well reasoned argument which frankly and clearly describes the process. Muller essentially says, trust me, we’re doing science here and the science refutes the other side.
Now look at the press response to this paper. Pick any recent story and you immediately see the Stop The Presses! shout. We’ve Proved Global Warming!!! They note the high class credential of the Berkeley team (hey, there’s a Nobel in there, didja know?), those billions of data points, and, case closed.
Does anyone actually read this research? This is nothing more than an Observational Tooth Fairy Tale. But, since no one seems to read the reports, everyone merely follows the Rule: All Bad Science Is Persuasive.




