Philosophy

Don’t Use Statistics Unless You Have To

It's catching.

It’s catching. (Image source.)

We’re finally getting it, as evinced by the responses to the article “Netherlands Temperature Controversy: Or, Yet Again, How Not To Do Time Series.

Let’s return to the Screaming Willies. Quoting myself (more or less):

You’re a doctor (your mother is proud) and have invented a new pill, profitizol, said to cure the screaming willies. You give this pill to 100 volunteer sufferers, and to another 100 you give an identically looking placebo.

Here are the facts, doc: 72 folks in the profitizol group got better, whereas only 58 in the placebo group did.

Now here is what I swear is not a trick question. If you can answer it, you’ll have grasped the true essence of statistical modeling. In what group were there a greater proportion of recoverers?

This is the same question that was asked [before], but with respect to…temperature values. Once we decided what was meant by a “trend”—itself no easy task—the question was: Was there a trend?

May I have a drum roll, please! The answer to today’s question is—isn’t the tension unbearable?—more people in the profitizol group got better.

Probability models aren’t needed: the result is unambiguously 100% certain sure.

As before, I asked, what caused the difference in rates? I don’t know and neither do you. It might have been the differences due to profitizol or it might be due to many other things about which we have no evidence. All we measured was who took what substance and who got better.

What caused the temperature to do what it did? I don’t know that either. Strike that. I do know that it wasn’t time. Time is not a cause. Fitting any standard time series model is thus admitting that we don’t know what the cause was or causes were. This is another reason only to use these models in a predictive manner: because we don’t know the causes. And because we don’t know the causes, it does not follow that the lone sole only cause was, say, strictly linear forcing. Or some weird force that just happened to match what some smoother (running means, say) produced.

Probability isn’t needed to say what happened. We can look and see that for ourselves. Probability is only needed to say what might yet happen (or rather, to say things about that which we haven’t yet observed, even though the observations took place in the past).

Probability does not say why something happened.

I pray that you will memorize that statement. If everybody who used probability models recited that statement while standing at attention before writing a paper, the world would be spared much grief.

In our case, is there any evidence profitizol was the cause of some of the “extra” cures? Well, sure. The difference itself is that evidence. But there’s no proof. What is there proof of?

That it cannot be that profitizol “works” in the sense that everybody who gets it is cured. The proof is the observation that not everybody who got the drug was cured. There is thus similar proof that the placebo doesn’t “work” either. We also know for sure that some thing or things caused each person who got better to get better, and other causes that made people who were sick to stay sick. Different causes.

Another thing we know with certainty: that “chance” didn’t cause the observed difference. Chance like time is not a cause. That is why we do not need probability models to say what happened! Nothing is ever “due” to chance!

This is why hypothesis testing must go, must be purged, must be repulsed, must be shunned, must be abandoned, must be left behind like an 18-year-old purges her commonsense when she matriculates at Smith.

Amusingly for this set of data a test of proportions gives a p-value of 0.054, so a researcher who used that test would write the baseless headline, “No Link Between Profitizol And The Screaming Willies!” But if the researcher had used logistic regression, the p-value would have been 0.039, which would have seen the baseless headline “Profitizol Linked To Screaming Willies Cure!”

Both researchers would falsely think in terms of cause, and both would be sure that cause was or wasn’t present. Like I said, time for hypothesis testing to die the death it deserves. Bring out the guillotine.

Since this is the week of Thanksgiving, that’s enough for now.

Categories: Philosophy, Statistics

65 replies »

  1. Happy thanksgiving to you as well!

    Hot off the presses from the Statistical Times:

    The stock of Bayesless Pharmaceuticals, best known for producing Profitizol (TM), jumped several points after a recent shareholder’s meeting. At this meeting it was voted that the company should consider p-values below 0.1 to be significant. Traditionally, 0.05 has been considered significant, marking the first departure from the consensus. The president of Bayesless, Mr. Fisher, commented that “Our competitors’ products are now demonstrated to be inferior, as more of ours are significantly better than placebo.”

    The news is rocking the pharmaceutical industry, which has long been suffering from trying to perform an infinite number of experiments in order to demonstrate their significance. An employee who wishes to remain anonymous commented to ST that “We’re going to attempt to compete by creating statistical modeling that can generate the wee-est p-values seen in the industry.”

    Market analysts have begun speculating if it is possible to patent the significance level, and if this could be used as a competitive tool against Bayesless. Here at ST, our in-house modelers have yet to determine if the jump in stock value is due to time, however, as time and price have been discovered to be highly correlated within the last few hours of the market.

    What is certain, though, is that statistics in the industry are more important than ever before. Only a well-dressed

  2. My laptop trackpad submitted the story before it’s copy editor could complete it. The story finishes below:

    Only a lone, well-dressed detractor could be found on Wall Street carrying a sign that read “Probability does not say why something happened”. Analysts could not determine the meaning of the sign, and left the man to his rantings.

  3. Bob,

    Rather, to the extent the analogy works, the other way around. Probability is used to express uncertainty. Because we are already certain of what has happened (mostly!), we use probability to say what might happen.

    Of course, “what might happen” could be past events, like murder mysteries.

  4. Matt, by “descriptive” I mean describing what COULD happen and what HAS happened; by “prescriptive” I mean saying this is what WILL happen. Accordingly I’d judge probability/statistics to be descriptive rather than prescriptive. ….
    Later, looking at web definitions I see my definitions are out of the foul line. As applied to grammar and linguistics, descriptive means talking about language as it is commonly used, while prescriptive means telling you how you should speak properly (Henry Higgins and Eliza). There’s also an interesting article on the terms applied to information technology:
    http://www.informationweek.com/big-data/big-data-analytics/big-data-analytics-descriptive-vs-predictive-vs-prescriptive/d/d-id/1113279

  5. When looking at “statistically significant” and medical type things, one that is interesting to me is how both heat and cold can be used to decrease migraines, sometimes in the same person. From the studies, some of which show great improvement with a specific treatment, it becomes fairly obvious that we know nothing about the cause. The idea that two opposite treatments both work shows that correlation and statistical significance really don’t address causes. Even if the results are “statistically significant”.

  6. Sheri wrote:
    “When looking at “statistically significant” and medical type things, one that is interesting to me is how both heat and cold can be used to decrease migraines, sometimes in the same person. From the studies, some of which show great improvement with a specific treatment, it becomes fairly obvious that we know nothing about the cause. The idea that two opposite treatments both work shows that correlation and statistical significance really don’t address causes. Even if the results are “statistically significant”.”

    To the degree you are also discussing climate change, the analogy is imperfect. The human body is very complicated — the CO2 molecule is not. Tyndall discovered in 1859 that CO2 (and some other gases) absorbs infrared radiation. We know humans have emitted a huge quantity of CO2 from burning fossil fuels — because they’ve been billed for it. We know CO2 is an important greenhouse gas, because without it you can’t explain the Earth’s pre-Industrial surface temperature.

    So we know a lot more than we do about a patient. We also know that the laws of physics REQUIRE anthropogenic climate change. And, as expected, the globe is warming up, with no other apparent cause in sight.

    The scientific surprise would be if global warming WASN’T happening.

  7. David

    The human body is very complicated — the CO2 molecule is not.

    That’s Appell and Oranges.

    The human body is very complicated — The climate system is very complicated.

  8. John: The climate system is complicated. The CO2 molecule isn’t. It absorbs infrared radiation, and the Earth emits it. Give that, it’s just a matter of working out the response and how much additional warming additional CO2 creates.

    The CO2 part is really the surest part of climate science, because it can be solved using fundamental laws of physics (thought the detailed equations can only be solved numerically). It’s the rest of the warming — due to the feedbacks — where the complication arises. But the CO2 part is solid.

  9. The scientific surprise would be if global warming WASN’T happening.

    It only takes looking at the past record to see that warming has happened. What can’t be seen is the supposed anthropogenic cause. It’s far from clear that has warmed at all in the last 18 years or so.

    AGW proponents are often too quick to conflate any warming with human caused warming as in “AGW is real because icebergs are melting”.

    The CO2 part is really the surest part of climate science, because it can be solved using fundamental laws of physics (thought the detailed equations can only be solved numerically). It’s the rest of the warming — due to the feedbacks — where the complication arises.

    Effectively base on an assumption which can’t be verified at all.

  10. Terry Oldberg wrote:
    “Correction: Fallacious arguments beloved by climatologists create the appearance that “the laws of physics REQUIRE anthropogenic climate change.””

    Not if you really understand the laws of physics. The Earth emits IR upward. Atmospheric CO2 absorbs it, and when it re-emits IR, some of that is downward.

    The only question then is, how big is this effect. Scientists have been understanding that for well over a century.

  11. DAV wrote:
    >> The CO2 part is really the surest part of climate science, because it can be solved using fundamental laws of physics (thought the detailed equations can only be solved numerically). It’s the rest of the warming — due to the feedbacks — where the complication arises.<<
    "Effectively base on an assumption which can’t be verified at all."

    Actually not — direct measurements confirm that increasing greenhouse effect due to CO2 (and other GHGs):

    "Increases in greenhouse forcing inferred from the outgoing longwave radiation spectra of the Earth in 1970 and 1997," J.E. Harries et al, Nature 410, 355-357 (15 March 2001).
    http://www.nature.com/nature/journal/v410/n6826/abs/410355a0.html

    “Comparison of spectrally resolved outgoing longwave data between 1970 and present,” J.A. Griggs et al, Proc SPIE 164, 5543 (2004). http://spiedigitallibrary.org/proceedings/resource/2/psisdg/5543/1/164_1

    “Spectral signatures of climate change in the Earth's infrared spectrum between 1970 and 2006,” Chen et al, (2007) http://www.eumetsat.int/Home/Main/Publications/Conference_and_Workshop_Proceedings/groups/cps/documents/document/pdf_conf_p50_s9_01_harries_v.pdf

    “Radiative forcing – measured at Earth’s surface – corroborate the increasing greenhouse effect,” R. Phillipona et al, Geo Res Letters, v31 L03202 (2004)
    http://onlinelibrary.wiley.com/doi/10.1029/2003GL018765/abstract

    “Measurements of the Radiative Surface Forcing of Climate,” W.F.J. Evans, Jan 2006
    https://ams.confex.com/ams/Annual2006/techprogram/paper_100737.htm

    "A method for continuous estimation of clear-sky downwelling longwave radiative flux developed using ARM surface measurements," C. N. Long and D. D. Turner, Journal of Geophysical Research, vol 113, D18206, doi:10.1029/2008JD009936, 2008
    http://onlinelibrary.wiley.com/doi/10.1029/2008JD009936/abstract

    "Satellite-Based Reconstruction of the Tropical Oceanic Clear-Sky Outgoing Longwave Radiation and Comparison with Climate Models," Gastineau et al, J Climate, vol 27, 941–957 (2014).
    http://journals.ametsoc.org/doi/abs/10.1175/JCLI-D-13-00047.1

    "Evaluations of atmospheric downward longwave radiation from 44 coupled general circulation models of CMIP5," Qian Ma et al, JGR Atmospheres, Volume 119, Issue 8, pages 4486–4497, April 27, 2014.
    http://onlinelibrary.wiley.com/doi/10.1002/2013JD021427/abstract

  12. David

    If the “climate system” were simply the components of the atmosphere, I would agree.

    But the “climate system” by definition includes more than just the atmosphere – it includes the land and everything growing on it plus or minus everything not growing on it, the oceans and everything growing in it plus or minus everything not growing in it, the sun which causes things to grow or not grow. We won’t even discuss geological issues.

    We know the “system” utilizes the components of the atmosphere, we have no idea how much. We know the system utilizes the sun, we don’t know how much. While we canna change the laws of Physics, without a clear understanding of the entire system, the laws of Physics as it relates to the CO2 module alone cannot represent the “climate system”.

  13. John B wrote:
    “But the “climate system” by definition includes more than just the atmosphere…”

    I wrote above that the complication is in the feedbacks.

    “We won’t even discuss geological issues.”

    Good, because they’re not relevant to AGW.

    “We know the “system” utilizes the components of the atmosphere, we have no idea how much. ”

    Not true — we know a lot about the global carbon cycle, and where the extra carbon is going.

    “We know the system utilizes the sun, we don’t know how much.”

    Not true; the global energy budget is now well known, better known that the carbon cycle:
    http://www.cgd.ucar.edu/cas/Topics/energybudgets.html

    “While we canna change the laws of Physics, without a clear understanding of the entire system, the laws of Physics as it relates to the CO2 module alone cannot represent the “climate system”.”

    I never said it didn’t.

    A doubling of CO2 causes, at present temperatures, a 1.2 K rise in average global surface temperature. Past episodes of warming show how the planet reacts to that extra heat, and the feedback is definitely positive (up to a point, of course). In fact, you don’t even need climate models to determine climate sensitivity — you can get its value from paleoclimate research.

  14. Actually not — direct measurements confirm that increasing greenhouse effect due to CO2 (and other GHGs):

    Those aren’t direct measurements. There is no way to measure dT(CO2) because it can’t be separated from all of the other causes of dT. Your first link got it right: “Increases in greenhouse forcing inferred from the outgoing longwave radiation …”

    IOW: They measured the amount of radiation and guessed at the surface temperature that has been “caused”. Not an actual measurement.

  15. DAV wrote:
    “Those aren’t direct measurements. ”

    They are direct measurements. No, they don’t measure dT/dCO2 (that should be a partial derivative). But they definitely show it’s positive, because they show radiation being absorbed at the top of the atmosphere and more radiation coming down to the surface.

    There will never be a clean measurement of ∂T/∂CO2, because climate can’t be reduced to just one variable. That hardly means we don’t know anything about CO2 and its effect on climate.

    The human body is complex, but scientists have figured out that smoking has negative health consequences. But there’s no equation that relates the properties of a cell to the properties of tobacco, nor an experiment that can measure that directly.

  16. DAV wrote:
    “IOW: They measured the amount of radiation and guessed at the surface temperature that has been “caused”. Not an actual measurement.”

    No, they measured amounts of radiation. Then the FIrst Law of Themodynamics requires that radiation’s energy be conserved.

  17. No, they measured amounts of radiation. Then the FIrst Law of Themodynamics requires that radiation’s energy be conserved.

    So then they didn’t measure the atmospheric temperature change caused by CO2 but estimated it by proxy from the radiation. That means they took no direct measurement and applied a model.

    Yes, indeed the human body is complex as is the Earth’s thermodynamic system. And because the human body is a complex system, it doen’t repond to heat inputs in the way a metal rod might. You can have a person stick their feet in a pot of very hot water without noticeably altering their core temperature but get different results when inserting a metal rod into the same pot. Likewise, you can’t assume merely because lab experiments resulted in a rise in air temperature from CO2 in a simple system like a bottle the same thing will happen in the more complex system that is the Earth. It is a erroneous to assume you can. You would be ignoring all of the other factors and their interactions within the system. Without knowing what they are, how can you possibly ignore them?

    They are direct measurements. No, they don’t measure dT/dCO2 (that should be a partial derivative).

    You can’t separate out dT(CO2) without resorting to a model that purports to know dT/dCO2. That is not a direct measurement. In fact, it’s entirely circular: you know the effect of X because the effect of X is being used to calculate the effect of X. At no time was the effect of X ever validated through demonstration except as it might apply in a much simpler system. It is necessary to show that the effect in the simpler system can be applied to the more complex one prior to claiming you know what its impact is within the more complex system. You can’t just claim it is directly applicable.

  18. Matt Briggs, it seems that the thread is being diverted away from your well-chosen topic of what probability and statistics is all about, to conjectures and disputes about a subject of dubious importance (AGW)–discussions not even as substantive the purported scholastic disputes about “how many angels can dance on the point of a pin”. Is there any way to divert this stream to a forum where the AGW evangelizers and the non-believers can dispute? It takes more psychic energy than I can spare to sift through the comments about the global warming nonsense. But it’s your call.

  19. Mr. Briggs,

    We should all be very worried if modern clinical trials were as trivial as the example provided by you in this post.

    Also, whatever prescription drugs you are taking, you better stop taking them because they are stained with the blood of p-value.

    (I sincerely hope that you are not taking any medications though.)

    Right,“probability itself does not say why something happened.” Chance doesn’t cause anything, either. A layman understands that “due to chance” simply means “cause unknown.” Can you offer more in the book you plan to write?

    Modern, competent statisticians are free to use whatever techniques, Bayesian or frequentist analysis, and they are having fun with the challenge of critically thinking and sorting through about the big data! Which doesn’t seem to be what you are interested in doing.

    So, may I recommend that you google for the work of D. Rubin and P. Holland. They both have written about “statistics and causal Inference.” (Evidently, econometricians are familiar with their work, or so I was told by an econometrician speaker just last week.) You might also enjoy the paper “Philosophy of statistics” by D. Lindley. Literature review often helps researchers organize their thoughts better and see things in a clearer way. You might just find them more enjoyable and rewarding than the articles in some blogs or MailOnline. Well, you know what to do if you want your peers to value your work!

  20. Actually, the CO2 sidebar is pertinent to the topic. The Arrhenius Effect is not sufficient to account for the observations, and so the models used by the IPCC incorporate various “amplifier” feedback mechanisms in order to match the observations, and these amplifiers are rather less-well established than Arrhenius’ original work. That one must use partial derivatives ∂T/∂CO2 indicates there are additional factors in play. Otherwise, since CO2 has been increasing for the past 7000 years, one would have expected steadily increasing temperatures for the past seven millennia; but the temperatures have only been notably increasing for the past 400 years, since the end of the Maunder Minimum and the onset of the Solar Grand Max. And even then, there have been notable reversals: the Dalton Minimum, the Global Cooling of 1940-1970, etc.

    Dr. Briggs’ main point is that no associative law can be a causal law, and statistical evidence can only provide associative relationships. The Arrhenius law is actually causal, since a physical mechanism is being asserted. But it is also known that as water gets warmer, less CO2 can remain dissolved in it, so warmer sea surface temps lead to more CO2 coming out of solution.

    A less contentious example, since no political amour propre is involved, is a test of battery life involving two production lines, one using fresh hydroxide and the other recycled hydroxide. Of course, there was a difference in battery life between the two, and it tested significant under the usual procedures. But all that meant was that the two lines did not produce batteries with the same mean life. It did not mean that the difference was due to the hydroxide, since there were a great many other differences between the two lines. That’s why in industrial problem solving, such statistical analyses are useful only to indicate rocks that may be worth rolling over to see what crawls out.

  21. Bob: I just skip over the problem children.

    JH: While I am not opposed to medications (medications are why I am still alive, so I’m a bit prejudiced), the FDA approval process is terrifying. I read everything I can before I take any medication.

    I think that people may be misinterpreting some of this. I learned “wee-p values” in college as a way to identify human behaviours and their causes. Yet, I understood that even with the wee-p value, there was a very real chance of missing the actual cause. Experiments had to be very carefully designed. There are places where statistics are necessary and can be valuable. Sorting through data–which is not the same as finding causality. However, it remains true that virtually every statistic can be countered with an opposite one. If you look through studies, you find studies with statistical significance that are diametrically opposed. That’s a problem and confusing. It needs to be addressed—how to know when statistics are valid and when they are not. As of late, the entire discipline seems out of control at times. By the way, one of textbooks for my stat class was “How to Lie with Statistics”.

    The paper by Lindley looks interesting. I will check it out.

  22. DAV wrote:
    “So then they didn’t measure the atmospheric temperature change caused by CO2 but estimated it by proxy from the radiation. That means they took no direct measurement and applied a model.”

    Describe to me how you yourself would measure dT/dCO2. You can assume whatever planetary conditions you want.

  23. DAV wrote:
    “So then they didn’t measure the atmospheric temperature change caused by CO2 but estimated it by proxy from the radiation. That means they took no direct measurement and applied a model.”

    No.
    Why don’t you read the paper and learn what they did and didn’t do?

  24. Sheri wrote:
    “My blog has an open thread for discussion of whatever you want. Feel free to go there and have it out.”

    Warning: She censors comments that contain science she does not like.

  25. Ye Olde Statisician wrote:
    “…but the temperatures have only been notably increasing for the past 400 years.”

    That’s also wrong. See: Little Ice Age. Read Marcott et al, Science 2013
    http://www.sciencemag.org/content/339/6124/1198.abstract

    “And even then, there have been notable reversals: … the Global Cooling of 1940-1970, etc.”

    The cooling during those decades was slim-to-none: -0.1 C at best:
    http://data.giss.nasa.gov/gistemp/graphs_v3/

  26. DAV commented:Those aren’t direct measurements. There is no way to measure dT(CO2) because it can’t be separated from all of the other causes of dT. Your first link got it right: “Increases in greenhouse forcing inferred from the outgoing longwave radiation …””

    Yes. Also, did you know you can’t measure the mass of the Earth? Fly into a black hole? See a quark?

    That doesn’t mean we don’t know about these things. Same with AGW. All of science is heavily dependent on inference.

    IOW: They measured the amount of radiation and guessed at the surface temperature that has been “caused”. Not an actual measurement.

  27. The cooling during those decades was slim-to-none: -0.1 C at best

    If -0.1º C is “slim-to-none” why is +0.2º C catastrophic?

    One must also ask whether these things appear in the data before or after they have been homogenized. The point is that none of this is actual data, but always the output of a model. You cannot have a derivative like ∂T/∂CO2 unless you have a function, i.e., a model.
    +++++++++
    http://www.climate4you.com/images/GISP2%20TemperatureSince10700%20BP%20with%20CO2%20from%20EPICA%20DomeC.gif

  28. Describe to me how you yourself would measure dT/dCO2. You can assume whatever planetary conditions you want.

    I have no idea. For one, that’s not a measurement — it’s a calculation. You have a value, T, that is an amalgam of A*X +B*Y+ C*Z+D*whatever. How does dT/dX get you the values of A, B , C and D? It seems that you are fitting a model and expressing a correlation without really knowing what effect T,X,Y,etc. have on each other, if any. The form of that amalgam is not really important here. If it is more complicated then it just makes the job that much harder.

    It’s not even really known what the temperature was umpteen millennia ago A problem with most proxies is they are rarely simple relationships that are hard, if not impossible, to separate. A good example of this is using tree ring width to arrive at temperature. The width is affected by more things than temperature. Recent data (after 1960) shows a negative correlation to temperature (hide the decline and all that) which calls into question all of past reconstructions.

    Further complicating the picture is heat does not always get expressed as air temperature. Some. maybe a lot, of it goes into oceans perhaps to be released at a much later time. So, even if you could somehow measure the incoming and outgoing energy, you still need to account for its distribution recognizing that some of it will not be in the atmosphere.

    What would be a big help is a decent predictive model which I don’t see having been achieved yet. With that, you might be able to claim to have a handle on the problem. It would help your case immensely. Until then please stop being so adamant about have all the answers.

    I’ve been having trouble posting comments. I keep getting timeouts. It’s not as simple as swapping as I’ve tried different machines. The site just seems to be generally slow.

  29. Ye Olde Statisician wrote:
    “If -0.1º C is “slim-to-none” why is +0.2º C catastrophic?”

    Do you just make up numbers when you need one? Seriously. Because every number you’ve supplied here has been wrong.

    Warming since about 1900 is +0.8 C, and laterly +0.9 C by some datasets. And it’s hardly going to stop here and now.

    You can calculate a derivative numerically — you don’t need an analytic function. That’s what climate models do to calculate climate sensitivity, except instead of CO2 amounts they use radiative forcings. Then

    climate sensitivity = ∂T/∂F

    where F is the forcing under discussion (CO2, solar, CH4, etc.).

  30. DAV wrote:
    “It seems that you are fitting a model and expressing a correlation without really knowing what effect T,X,Y,etc. have on each other, if any.”

    Have you read a climate textbook yet?

  31. DAV wrote:
    “What would be a big help is a decent predictive model which I don’t see having been achieved yet.”

    Haven’t you read and understood anything here?????

    AGAIN: it is IMPOSSIBLE to “predict” with a climate model, because no one knows necessary sociological factors such as fossil fuel use for the future. Will China’s peak CO2 be 2030? Will they then switch to gas or solar? Will their cars run on gasoline or electricity? How much fossil fuels will India burn between now and 2100? Will the world’s population in 2100 by 9 B or 12 B? What energy sources will they use between now and then?

    I-M-P-O-S-S-I-B-L-E to predict. The best that can be done is project, based on assumed scenarios, none of which will be exactly right.

  32. AGAIN: it is IMPOSSIBLE to “predict” with a climate model

    Well, then you’re stuck then aren’t you? You will never be able to demonstrate you have a handle on the problem. The models will always be descriptive. Who needs a function that can only give you the past? Far easier and cheaper to just look up the answer.

  33. DAV wrote:
    “Well, then you’re stuck then aren’t you? You will never be able to demonstrate you have a handle on the problem.”

    1) Do you know when (or even if) your house will burn down?
    2) Do you buy fire insurance ?

    Do you not stop smoking because a doctor can’t give you the exact month you will develop lung cancer?

  34. Sheri,

    Yet, I understood that even with the wee-p value, there was a very real chance of missing the actual cause.

    Do you have a better option that would never miss the actual cause? Yeah, there was a real chance, just like any decision made under uncertainty could be wrong. But… is the p-value supposed to somehow identify the actual cause? No.

    Drugs are approved often due to small p-values or practical significance in drug effectiveness. There is a difference between the cause of an effect and the effect of a cause (i.e. a treatment in this context)!

    If the medications are why you are still alive, than p-value has helped you. It has done you right!

    However, it remains true that virtually every statistic can be countered with an opposite one. If you look through studies, you find studies with statistical significance that are dilametrically opposed.
    Virtually every statistic?! Evidence, please.

    As of late, the entire discipline seems out of control at times.

    The entire discipline? Evidence, please.

  35. JH: Of course I don’t have a method that would never miss the actual cause. That’s not even humanly possible, so far as I know.

    My biggest concern with the FDA is the number of studies needed to prove effectiveness versus how many trials are actually ran. I have read that only two studies have to show significance and it does matter how many do not. That’s scary.

    Evidence? This blog has multiple examples of statistics run amok.

    * Also, please note I use the word “seems”, not “is”. I was stating a personal opinion, not a fact. I don’t need evidence for an opinion. Why is English such a difficult language for people to understand. “Seems” is an opinion word.

  36. Appell,

    I have no idea what you are going on about. Looks rather emotional. The only way to validate a model is to see how well it predicts. It is IMPOSSIBLE (your word) for a model to predict so it follows they can never be verified and are thus no better than Ouija boards, tea leaves, Tarot cards, horoscopes, etc. but whole lots more expensive. Why should we keep funding these climate models?

  37. I like your post and thought about a minimal statically time series reconstruction. Instead of converting everything into anomaly, us used the crazy assumption that research building temperature proxies from ocean cores might actually have a clue what they we doing. Then I selected similar local reconstructions using the same basic methods with about the same uncertainties and used the actual temperatures the had estimated. Then if a reconstruction was long enough for the period, I combined it with another short one from the same location so the reconstructions all ended close to the same time, If there wasn’t a another reconstruction, I used the last know value to fill the gap.

    https://lh4.googleusercontent.com/-2zeYv85NJLc/VHQFtxF-QLI/AAAAAAAAL1U/i_zfEv7Bqqc/w689-h439-no/Tropical%2BHolocene%2BOceans%2Brecon%2Bmax%2Bsolar%2Bat%2Bequator.png

    I ended up with a reconstruction of the tropical ocean SST for the Holocene that has things like MWPs, LIA and tends to indicate that the boring solar precessional cycle has an influence on tropical sea surface temperatures. For some reason I don’t have a “non-robust” attention grabbing hockey stick and the temperature of the reconstruction is within about a half degree of the observed temperature of the region in the overlap period. The most powerful statistical tool I used was a function called AVERAGE.

    Obviously I must have screwed up since there are bunches of other statistical functions I could have stuck in there some where 🙂

  38. DAV wrote:
    “The only way to validate a model is to see how well it predicts. It is IMPOSSIBLE (your word) for a model to predict so it follows they can never be verified and are thus no better than Ouija boards, tea leaves, Tarot cards, horoscopes, etc. but whole lots more expensive. Why should we keep funding these climate models?”

    Do you understand yet why future sociological information about population and energy use is needed to make a prediction?

    Do you understand the difference between a prediction and a projection?

    You often hear things like Medicare will go broke in 2037 (or whenever). How do they calculate that, when they don’t know the population between now and then, or future tax rates, or the number of people who will be on Medicare, beyond 2014?

  39. Terry Oldberg wrote:
    “There is no reason to fund them. They are worthless.”

    And your analysis of potential climate change is what?

    Or must we wait around for 3900 years to get enough data to keep the logicians of the world are happy?

  40. Yes oracles in one form or another have been with us for a very long. The successful ones (those that managed not to be ripped to pieces) made sure to leave themselves an out. How the oracles arrived at their proclamations had many forms such as consulting the gods, reading entrails and tea leaves. The modern equivalent is to consult with “Science” and “Computer Models” and the proclamations are called “projections” and “scenarios”.

    The IPCC even calls for sacrificing the economy instead of a few virgins. Sacrifices are popular time honored practices with oracles but this may be the first time sacrificing everybody was called for. The latest batch of oracles are quite cheeky.

    Until you can demonstrate that a climate model really is reliable — and the ONLY way to do that is by observing how well it predicts — then these scenarios and projections are nothing more than what oracles have done for all time but with more fashionable names.

    Or must we wait around for 3900 years

    Longer maybe. Whatever it takes for climate “science” to get its act together.

  41. Sheri,

    Whenever you use the word “seems” or the phrase “it seems that,” you are just expressing your out-of-thin-air, non-evidence-based opinion. Also, you need no evidence to support your opinion. Gotcha!

    Not sure why English is such a difficult language! It sure is much easier than Chinese.

    This blog has multiple examples of statistics run amok.

    So, the fault is not in the killers but in their chosen weapons.

    Multiple examples? How many examples would allow you to make a blanket statement?

  42. JH: Gotcha what? That was my meaning. I told you it was a “out-of-thin-air, non-evidence based opinion” and that is exactly what I meant. What part of “I meant it that way” do you not understand?

    I gave an opinion and you are free to agree or disagree with it. Should you want to take it to an evidence level, feel free to jump in an put some kind of statistical measure on it. Then we can move to evidence and supporting positions. Assuming I want to actually debate the issue, which I clearly did not or I would not have expressed an opinion rather than a position.

    (Statistically speaking, the number of examples needed to make a blanket statement varies from individual to individual and situation to situation. It is not generally based on actual counting of the examples but rather a memory of how many times one encounters the errors. In my case, well over half of the statistics I find, according to what may be a failing memory, are improperly applied. I picked “over half” because it seemed appropriate. Another may require more, some less. The probably of someone actually being able to articulate an exact number–well, you’re the statistician.)

  43. DAV wrote:
    “The modern equivalent is to consult with “Science” and “Computer Models” and the proclamations are called “projections” and “scenarios”.”

    You have yet to demonstrate how to do better. It’s a simple question. Answer it.

    “The IPCC even calls for sacrificing the economy instead of a few virgins.”

    Where did the IPCC call for that? Please provide the relevant passages, with a link that can be verified.

    “Until you can demonstrate that a climate model really is reliable — and the ONLY way to do that is by observing how well it predicts — then these scenarios and projections are nothing more than what oracles have done for all time but with more fashionable names.”

    First you need to supply the economic and energy future of the world. When will that be coming…..?

    “Whatever it takes for climate “science” to get its act together.”

    As if you’re in any way qualified to judge climate science. You’re not.

  44. That last post was juvenile, even for you.
    You repeatedly hide behind the fact that some of the influencing factors cannot be predicted. Got it. But how well have the models created at times X done, say at X +10 or 20 years once the influencing inputs are kniwn. Without of course changing the parameters. From what I have seen didley.

    Oh BTW your hijack if this thread is quite OCDish

  45. Nope — Ch11 indicates explicitedly that assumptions are still needed. From the very first page of that chapter:

    “Projected Changes in Near-term Temperature:
    “The projected change in global mean surface air temperature
    will likely be in the range 0.3 to 0.7°C (medium confidence). This
    projection is valid for the four RCP scenarios and assumes there will be
    no major volcanic eruptions or secular changes in total solar irradiance
    before 2035. ” (emphasis mine)

Leave a Reply

Your email address will not be published. Required fields are marked *