There was nothing in the world wrong with my scientific paper “Global Warming Increases Disastrous Music: A Scientific Paper.” Nothing, that is, that isn’t wrong with any paper which seeks, via statistical proof, to show a connection between global warming and some ill effect.
Let’s take the claim that global warming causes bad music and examine it. First, there is, after all, a lot of bad music around and something causes it1. It’s also indisputable that temperatures are increasing, or at least were increasing during the period 1946-2010. And FEMA says disasters are on the rise.
Do you say my theory is false because global warming “obviously” cannot cause bad music to be written? Well, prove it. Prove, in the form of a deduction, why it can’t. 
You cannot. At best, your counter evidence will be just another statistical argument, perhaps larded with suppositions of how music is created, how heat is a constructive creative force and not a destructive one, etc. Why would your suppositions trump mine?
in my paper, I gave quantitative evidence in the form of graphs and very small p-values. What quantitative evidence could you possibly provide? The most popular song of every year was worse in 2010 than in 1946 (melodically, lyrically, or harmonically)1. Temperatures increased. So did disasters.
If you argue against my theory you will be arguing against the empirical observations. You will going against the data. You will be an enemy of science!
I, a genuine PhD scientist, and therefore a soul pure and true, provided a causal linkage which shows how bad music could be the result of global warming. How can you disagree with that?
It’s true my linkage is somewhat rough, but that is nothing. I cold have easily padded on paper after paper, surmise after surmise, and made the journey from supposition to conclusion smooth. There are hundreds of papers published monthly that I could have bludgeoned you with.
There are authorities aplenty who do say that global warming causes disasters: I could cite a hundred easily. There are many more peer-reviewed authors who claim that disasters cause mental disease. And there are—they really do exist—scholars who claim that musical ability is harmed by mental illness.
I could have spent a week fine tuning the causal path so that the argument was as compelling as those in any peer-reviewed paper.
After that path came the statistical analysis. Of that, there was nothing untoward. Oh, sure, I could have improved it by, say, taking into account the auto-correlation of the data points, and so forth. But none of these improvements would change the statistical conclusion.
I could have been a Bayesian and not a frequentist, eschewed p-values and provided posteriors of the model parameters. Still, no conclusion would have changed. I could have then gone whole hog and used the techniques that I advocate as being superior to either frequentist of classical Bayesian analysis; i.e. the predictive techniques I’m always on about.
But they wouldn’t have changed anything either. The confidence I had in the conclusion would have gone down a bit (using these modern methods), but it still would have been high. I mean, just look at those pictures! Those are serious and compelling. That’s why any statistical procedure will show high confidence or “significance.”
What choice do you have except to accept my thesis?
If you need statistics to prove something, and you have no proof except statistical proof, then what you have proved probably isn’t true. Statistical evidence is the lowest form of evidence there is.
What a depressing conclusion.
—————————————————————————————–
1I used yesterday the increase in the number of words in a pop song as the measure of badness. But that is only a proxy for what I prefer: the decrease in unique words in a song. Repetitiveness is positively correlated with badness. Correlated, I say, not determinative. Anyway, it would not take much effort to shore up the quantification of musical badness, to eliminate repetitiveness, and to instead quantify badness in learned language, speaking of melodic, harmonic, and rhythmic aspects in highly technical terms.
Briggs, your paper suffers from a fundamental “fatal flaw”…though that minor defect is easily rectified:
Lack of consensus! Such papers must have consensus of agreement. That’s what its all about. Until your paper is sufficiently consensified by peer review it may remain a mere also-ran.
Let the consensification begin:
So I hereby apply my formal concurrence on your paper & its conclussions.
Now you/your paper is formally in the game!
Mr. Briggs,
Again and again you ignore the fact that climate research is mostly physics, not statistics. As long as you don’t propose a physical mechanism linking rising temperatures with music quality (whatever it is and however you quantify it), your paper is rubbish — because correlation is not causation. Your attempts at satire would weigh a bit more if anyone actually used statistics in climate research the way you are parodying. I can’t wait for you to actually get to review IPCC papers. You’d have to read them then, you know.
Grzegorz Staniak,
Your lack of knowledge of statistics and of the massive non-physics literature of claims of untoward events said to be due to global warming is truly impressive. You have to actually read these papers to understand, you know.
Start at Numbers Watch.
Mr. Briggs,
I didn’t really count on a sensible answer. You’ve used that patronizing tone and arbitrary assumptions before, and I’m still not impressed. Anyway, no wonder you don’t know what you’re talking about if you spend time reading pages like the one above and think they are about climate research. You’d better start reading this:
http://www.ipcc.ch/publications_and_data/publications_and_data_reports.shtml
and references you find there. You’ll need it.
Grzegorz,
Read the Numbers Watch-linked papers and report back here. Otherwise, please no more irrelevant, evasive answers (including answering this by making claims I am being evasive: it’s wearying).
Mr. Briggs,
I’ve seen this page before, and by the looks of it hasn’t improved. If you think it links to papers on climate, you probably haven’t read it yourself.
Grzegorz,
Your last comment was foolish and displays either a irredeemable stubbornness of profound inability to grasp simple logic. I made a claim about the statistics in papers that claim untoward effects due to climate change. These claims I made are independent on whether or not the climate changes or, if it does, why.
Now, in an effort to teach you one last time, go and look at these papers (say a dozen or so peer-reviewed ones) and then come back and comment.
Mr. Briggs,
Haven’t I mentioned before that your patronizing tone doesn’t impress me much? I’d swear I have. You know nothing about climate research that you could teach to anyone, and you’re unable to support your claims with evidence. No, the gossip page that links to press clips and denialist blogs is not suitable evidence in this case.
I’m sorry, but I’m not gonna fall for an online version of the Gish Gallop. If you have a peer-reviewed paper that could be an inspiration for your parody, just link it. Can you find one on the AR4 reference list?
My dear Grzegorz,
If you believe I have nothing to teach, then go away.
But if you’d like to learn, then as a matter of fact, I have several examples of atrocious peer-reviewed papers (as does Numbers Watch, which in your stubbornness you close your eyes to). One of the most egregious is this: Italian suicides.
It’s a paper that comes to the direct opposite conclusion of the data.
My Dear Mr. Briggs,
I had a piece of coconut cake for breakfast. How about you? (Apple pie is my favorite breakfast food.) Evidently the cake didn’t do me any good, and I was still in a funk after eating the cake. But now… hahaha! Sorry, I know it’s not funny. But for an untellable reason, I found the exchange between you and Mr. Staniak hilarious!
So there is a statistically significant linear correlation between two seemingly unrelated variables, therefore you conclude that statistical evidence is the lowest form of evidence there is!
Of course, if I have opinions, why would I need data evidence? ^_^
Let me cheer you up! One way to avoid the situation is to do some research first and then collect quality data. Even if foot size sounds like a ridiculous predictor for student’s GPA, in addition to hours of study, go ahead and add foot size as one of the independent variables… and see what happens! You would love Statistics for this!
So anyone want to guess which patronizing commenter above is seriously over-invested in AGW and can’t stand it because the issue has begun falling apart world wide? Anyone? Sorta reminds me of a “scientific” Meghan McCain. Just whistle and shout loud enough and the marks will never notice mistakes and lies.
I heartily disagree. Global warming causes GOOD music, viz., http://www.youtube.com/watch?v=Vx-t9k7epIk
Noblesse,
with one example ( most likely an outlier) you are trying to undermine the finding uncovered by the ironclad multiple imputations method. The method was around for quite some time and was verified over and over again.
Mr. Briggs,
How selective of you. I’m sure you’ve got a lot to offer as far as your area of expertise is concerned, and I’m honestly glad that you’ll (hopefully) review IPCC papers, since statistical methods have been a weak spot for many researchers (not only climatologists). However, it’s quite clear that your area of expertise doesn’t cover climate research as such, where you propagate cliches and myths, and sometimes simply grossly misrepresent science (like in the case of the CLOUD paper). The satire you produced might apply for papers like the one you linked, but last time I checked Journal of Affective Disorders wasn’t a source of references for IPCC reports. So that would restrict your critique to non-climatologists, right?
Noblesse & 49er,
Briggs has jumped the causation wagon again. It is a well known fact that bad music causes clouds which lessen GW. The concert depicted in the post actually works. Temps are stabilizing and even coming down. In any case, the music per se has little effect — it’s LISTENING to it that matters — ask any RIAA member.
I commend to your attention On the Non-existence of Music. Bishop Berkley would be proud.
We also have the fascinating experience here of hearing the sound of someone not listening. John Cage would be proud.
Dear Mr Staniak
.
I am afraid that it is you who have no clue about physics in general and climate in particular.
You claimed among other silliness that:
“Again and again you ignore the fact that climate research is mostly physics, not statistics”.
.
You are clearly not a physicist and indeed William is absolutely right, the disciplin that is by far mostly used in climate papers is statistics.
And if one leaves the WG1 and enters the WG2 of IPCC, we enter the kingdom of random and often absurd claims which are only supported by statistical “studies”.
.
Even if you write that you don’t want to learn (and seem to be strangely proud about it), let me explain WHY there is so little physics in climate papers.
From the physical point of view, climate science is a field theory.
The dynamics of the system is the dynamics of the coupled fields – temperature, pressure, density, velocity etc.
If one is able to specify those fields at every point for all times then the problem is solved.
Unfortunately the system is so complex that part of the field equations is unknown and the part that is known (f.ex Navier Stokes), can’t be solved..
By default it is then statistics that kick in as in this example of paper : http://www.pims.math.ca/files/kleeman_6.pdf
There are hundreds of similar papers that I could link and everybody can see that what happens here is statistics.
Of course the predictive skill is weak as expected and none of these papers can explain even such basic questions as to why is the ENSO frequence what it is.
PCA and EOF methods abound and these are again statistical tools and sophisticated ones at that.
Number of scientists have bungled PCA or EOF and Mann is just one prominent example.
About the only serious physics in the current climate “science” is the radiative transfer and it is notoriously not enough to tackle the system’s dynamics based on physical field equations.
.
A word about “models”.
They are all degenerated variations of weather models.
Of course they do NOT solve any field equations because of the inadequate resolution.
From the physical point of view, the numbers produced by the computer cannot be proven as converging to the solutions of the underlying equations.
They cannot do so anyway because the system is governed by spatio-temporal chaos.
This can be readily seen in the obvious fact that the dispersion of “results” of different models is atrocious.
That’s why the “climate science” invented the highly dubious method of “ensemble averaging”.
And what is this method? Again statistics.
.
So clearly William has exactly the necessary skills (e.g statistics) which are relevant to most papers dealing with the system’s dynamics and with the “ensemble averaging” of models and to the quasi entirety of papers relating climatic fields (humidity, precipitation,temperature etc) to economy and biology.
These latter papers are precisely of the kind “Global warming causes cancer,poverty (or whatever else you fancy)” and you have a large selection of this kind on the link William provided.
Perhaps G. Staniak, instead of making broad generalized comments, might address the following — fundamental uncertainties that induce heavy reliance on statistical tools to extract trends from what amounts to very limited data:
*
In addition to major unknowns about the Carbon Cycle, solar effects & clouds involve substantial uncertainties. The solar link to cloud formation (with cosmic ray involvement) is known to occur with certainty; however, as the latest CERN CLOUD experiment showed, how that actually occurs remains partly mysterious.
Some of the solar effects uncertainties are profound, with declining solar activity effectively nullifying anthropogenic effects, such as from 2002 to 2008 where CO2 increased steadily but warming was not observed.
The Carbon Cycle is uncertain & substantially unknown. Plants are consuming CO2 faster than plant-related decomposition discharges CO2 into the atmosphere, making plants a net CO2 scrubber overall – and warming leads to longer growing seasons and with that greater CO2 absorption by plants.
Over half of all CO2 produced, year after year, is absorbed – however, knowledge of the nature & locations of CO2 sources & sinks, as well as the processes that will affect their future evolution continues to be limited by lack of precise measurements of atmospheric CO2 (the processes responsible for year-to-year fluctuations in CO2 are largely unknown) – that knowledge is needed to accurately predict how CO2 sinks will change, how the changes will affect the rate of CO2 buildup, the impact on climate, and measure the effects of any carbon energy policy.
All such uncertainties are modeled based on empirical trends bolstered by statistical analyses of varying degrees of rigor, or lack thereof.
Thus, model-based climate forecasts have significant limitations. On time-scales of years to a decade, naturally induced surface temperature changes can dominate current anthropogenic warming—with many locations/regions showing locally induced variations that smother the global trend by an order of magnitude. On time scales of 10 years or more climate forecasts are difficult to make with general circulation climate models due to their many uncertainties. That illustrates how crude & generalized any given climate model is.
Further, developing (“2nd Worldâ€) countries are polluting, with CO2 being a substantial greenhouse gas among many others that first world countries like the US & Europe have long ago substantially stopped emitting. This 2nd World pollution show no signs of abating. If China, as an individual example, stopped producing additional such pollutants and used “cleanâ€/â€green†energy sources for all additional power generation (an impossibility for a very long time to come) its dependency on existing dirty (e.g. coal) power is such that CO2 emissions will continue to increase even if the U.S., Europe, and other First-World countries dramatically reduce CO2 emissions…and that too is an impossibility for the near-term & foreseeable future.
But does that really matter? Consider CO2, the now infamous anthropogenic gas (& now an official pollutant per the U.S. EPA!): 3.62% of green house gasses are CO2, and only 3.4% of that 3.62% is from human activity. That’s 0.12% of all CO2 emitted annually into the atmosphere is from human activity. That’s not much any way one examines it. Let’s put that in perspective, here’s a picture of the Big House (Michigan Stadium during a recent football game): http://blogs.courant.com/photo/pictures/The-Big-House.jpg About 110,000 people in attendance, and most are visible in the photo—so for comparison purposes & simple round numbers, that’s about 100,000 people in that picture. If each person represents a molecule of “air†(most being a nitrogen molecule, then the next plurality being an oxygen molecule, etc.) 38 would represent CO2 molecules. That’s what some are so worried about…and with about 0.12% of the annual increase due to human activity, how significant a problem is this, and more importantly, how much can be done to offset it? Especially when we have a very poor understanding of the mechanisms involved??
I suspect that Mr. Staniak is on the payroll of “BIG MUSIC”
Tom: ensemble averaging wasn’t invented by climate scientists. Its actually a pretty good technique. All of the face detection you ever see in consumer electronics and photo software use an ensemble of classifiers (minus the averaging). Random Forests use ensemble averaging for regression quite effectively.
Where the climate models fail is in how they are tested.
The correct way to test an ensemble is with a validation set never seen by the model, and measure its predictive ability. You can run a bunch of other tests (like predictions made with inputs that were also predictions or guesses) to see how the MSE grows under different circumstances.
Testing the model against the data that was used to build it usually only tells you how good a data compression algorithm you’re model is.
There is a glaring contradiction between your claim that you are a genuine PhD scientist and your conclusion that statistical evidence is the lowest form of evidence there is.
https://www.wmbriggs.com/blog/?p=4390#comments (comments are closed)
“Prior sensitivity is not problematic—and in any case, in simple situations like this give the exact same answers as the frequentist solution, as you know. We know that 1.92 is the actual risk of the observables, given the model is true and the data observed.â€
If you know the actual risk, why would you want to do Bayesian or frequentist estimation? No, in simple situations like this don’t give the exact same answers as the frequentist solution.
http://www.ccs.neu.edu/home/rjw/csg220/lectures/MLE-vs-Bayes.pdf
Why all the hostility towards our host? Sheesh.
Tom S: What part of “correlation does not equal causation” do you disagree with? Warm sunny days correlate well with drownings. Is this evidence of the sun causing people’s lungs to spontaneously fill with water?
Will
Tom: ensemble averaging wasn’t invented by climate scientists.
Sure, it was a sarcasm 🙂
Having some expertise in the ergodic theory, I know well that ensemble averages in the phase space are defined and have been used for more than 100 years.
However using this concept for numbers coming out of computer models makes no sense.
So just for the sake of accuracy, what I wanted to say is :
.
“That’s why the “climate science†invented the highly dubious method of using “ensemble averaging†on outputs of the computer models.”
Tom S,
Comments are always closed after one week. In basic linear regression, flat priors give the same answers as frequentist analysis, as is well known.
I’m troubled that your analysis appears to have overlooked the fact that early in his career, Al Gore (along with his then spouse Tipper) provided considerable vocal opposition to bad music (granted, they meant “bad” in the moral sense, not the aesthetic sense, or even the urban sense, where “bad” really means “good”).
I’m fairly certain that constitutes a major failing in your paper.
(In considering this reply, please excuse any patronizing tone, real or imagined).
I wonder how trustworthy statistical evidence is that fails to show significant results (i.e. p > 0.5 according to the general lore). Intuitively, it seems far more likely to stumble over any accidental correlation by chance and cry out “significant result!” than that a genuine causal relationship has been masked by a random correlation of exactly the same magnitude but opposite sign. So, if a study couldn’t find a significant result, we should be pretty sure that there is indeed no relationship, while statistically significant results are only very minor evidence that a relationship exists and have to be confirmed by further studies.
This reasoning also fits with the interpretation of the outcomes of medical tests: If the test is positive, the probability that I’m sick may still be quite low, depending on the prevalence of the disease; further tests are necessary to determine wether I contracted the disease or not. On the other hand, if the test shows negative, then I can be pretty sure that I’m healthy. Or are these considerations too simplistic?
nick,
Excellent question; exactly the right one to ask. The answer is: relationships can be predictive and useful even though they do not reach “statistical significance”. Just as variables/relationships which are “significant” can be of little or no use predictively, as explanations, or of use in making decisions.
I hope (if I have time) to have some examples of this.
Briggs,
looking forward to your examples! Apart from a certain general curiosity, the very serious personal backround of my question was the task of convincing my girlfriend of the futility of ever more significant results like “eat more broccoli, get no leukemia”. I tried my best by quoting null studies showing no relationships, only to run into the deadly reply: “If you don’t trust significant results, why do you trust results that are not even significant?”. So, you see, a lot is at stake here.
Talk, talk, talk. Everybody talks about the whether but no one ever does anything about it.
My intuition tells me that the global warming -> bad music link is backwards; clearly it’s the other way around. For example, I replaced the roof on my house a couple weeks ago, and it hasn’t stopped raining since. I forgot my hat one day, and sure enough it rained. I could go on and on – the evidence is strong that my actions affect the local rainfall probability.
Instead of just talking about the weather (and I include creating computer models as a form of talking), pro-active action is required to truly understand the magnitude and direction of causation. This means designing an experiment to actually measure the relationship.
Here is my proposed experimental procedure: Hire a million people. Their job is to listen to either good music or bad music on command. Command all of them to listen to good music at the same time, then at a randomly chosen interval (say, a period of time with uniform distribution between nine and seventeen days), command them all to listen to bad music. Do this for ten years, noting the exact time and date of each transition from good-to-bad music and bad-to-good music.
While this is going on, archive all the publicly available temperature and IR radiation data. After ten years, simply slice up the temperature/IR data into segments that all start at a transition from good-to-bad music, then add up all the slices. Any patterns or trends or spurious relationships in the data that are not caused by good versus bad music will tend to average away, while causally related data will be reinforced.
If no causal relationship rises out of the ‘noise’ after ten years, then we must assume you are correct and that global warming does indeed cause bad music, not the other way around. Until then, I remain a skeptic/denier.