Losing the plot

Lucia lost the plot some time ago, mostly by cherry-picking her time period, using a weird data-fitting method and failing to understand what she was looking at. Now RP Jr follows her down the rabbett hole and bizarrely describes her post as “clear”. Well, when people are telling you what you want to hear you’re apt to approve.

As usual, you’re better off avin a larf with James . Though if you’re tired of slapstick, maybe reading the truth at RC would be more useful. I prefer bluetooth myself nowadays 🙂

[Update: it gets worse. Roger is losing his temper, and unfortunately hasn’t found somone to ask about stats. Although in fact its not really a stats question, its a climate modelling question.You can’t compare the trends from different model realisations with 5 different estimates of the same observational period. To try to understand this, suppose all 5 obs estimates were really really close together – they could be, if all the methods were near equivalent. Then the SD would be very small. But the models, because they aren’t simulating the same real years, will maintain a large spread. Their statistics will be different, and unknowing black boxes will declare them different -W]

27 thoughts on “Losing the plot”

  1. I suppose that’s not exactly unexpected. Shine on, Breakthrough Institute. Mayhaps someone will mistake your offerings for diamonds.


  2. William,

    I lurk here to learn and be entertained but, this is really beneath you. This is a silly, sloppy reiterative attack. You also pointed to a RC post as the truth, in reading it and the comments not many took it as the truth.

    steven mosher said here on a previous subject:
    William. C-0 is Unheard of in climate science? Do you want an opportunity to check on that?

    Enviroentat statisics. Chapter 7. page 269


    I havent had the time to run down all the cites in that
    chaper but I did see one reference to Tom karl.

    There are more, of course


    So, not unheard of.

    The technique is standard cookbook statistics. The method has been used and explored in many areas of climate science.
    That does not make it a correct method for all time series, but it is surely not unheard of. present company excepted.

    You should really take a lesson in humility from Lucia. She goes to great lengths to answer question about what she has done. Your yourself have engaged her and she dealt with your forth rightly and curiously. In your own exchange with Lucia I found that you did not really want an answer, your criticism were easily answered (if they were not answered why did you drop off so quickly).

    Jon –
    I read your exchange with RP JR with interest, it is amazing how he lead you around the poll and used you as an example of just the short sightedness he was pointing out in his post, it was laughable as you took praise and patted yourself on the back here for being a sheep. Thus your name calling supprises no one.


  3. Dear Bob

    If we are supposed to deal with all kind of claims when are we ever supposed to work? Just go over to James and se for your self what is wrong or tell Roger to publish his claims… that is how science is done, not by doing not so good statistics on a blog then expecting the scientists to answer. If you really are after the facts in this story you already have enough data to see what’s right and wrong here… but ok I guess it could be made even clearer, but who has the time? I guess it will come eventually.


  4. Here’s a shorter version of Lucia and RPjr:

    Projection: The Red Sox are very likely to go to the playoffs this year.

    RPjr: The Red Sox lost 3 of 4 to the Twins last weekend. They’re a half game behind the Devil Rays. The prediction has been falsified!

    Commenter: But Roger there’s still 5 months and 120 games left to play.

    RPjr: So you’re saying there’s no way your projection can be falsisifed, hufff, that’s not science!


  5. “bob”

    The irony of you calling others ‘sheep’ and talking about being ‘lead around a poll’ whilst being impressed by this nonsense is astounding.

    The fact that they cannot grok the distinction between multidecadal projections of the forced component of climate vs. short term predictions of the real world, or are can and are carrying on as if they can’t is far more damning than any riposte I care to offer in response to your post.



  6. Magnus, bite your tongue. Based on recent experience Nature might well be willing to publish it.

    bob, I would suggest you examine the cvs of JA and WMC vs. those of RP Jr. and Lucia. Of course that does not by itself make the former right about this, but what it does make them is experts operating in their own field (discounting William’s job change since it was just a few months ago). Adding to that the qualifications of the rest of the RC co-authors doesn’t 100% guarantee which side of this argument is right, but I won’t be putting my money on the mechanical engineer and political scientist.

    Regarding C-O, mosher demonstrated very nicely that it’s a rarity in climate science.


  7. Well that might not be such a bad thing then he could read the answer in nature later and he just might understand and accept it…

    (ignoring the fact that he don’t have anything to publish unless he frames it as a policy issue which is a completely different thing)


  8. Pielke: Also, his claim that the models don’t simulate the same “real years” is contradicted by his colleagues at RC, see the X axis in the following figure:

    You can’t make this stuff up.

    Call me crazy, but I took the meaning of “same” to be in reference to the different initializations and thus the different realizations the runs would produce rather than the same *time period* (i.e. 2000-2007 or 2001-2008)- so that even over the same time period (same years) the runs are not attempting to achieve the same (or even extremely similar) year-to-year realizations, or “years”. The numerical value of the years of the time period would be the same, but the runs are not attempting to simulate the “same” 2001-2008.

    Comparing five observations produced in a similar manner of the same single “realization” of reality for 2001-2008 is going to give you an inherently smaller spread than 55 model realizations of “not same” 2001-2008s.

    I’ll admit to having had a bit of wine, but it looks like he’s assuming that you are talking about different time periods (i.e. not the ‘same real years’) rather than different initializations and thus realizations per run.

    Care to *hic* clarify for a feller?

    [I think you’re right. This has always been a fairly simple concept but hard to say in a few words. Climate models simulate, say, the 20th century. In that sense, they run through “real” years. But you do not expect them to track year-to-year variability – if you run 10 GCMs, they will all have ENSOs in different years, for example. For long term trends that doesn’t matter. If you look at – oh, take a random example, 7-year trends – then you expect a far wider scatter of trends from 10 GCM simulations of that period, than you do from different observational estimates, which you would hope would be nearly the same -W]


  9. Bingo. It shouldn’t be about stats at all. Neither set of data is applicable to using standard deviations. For the observations, 5 separate obs are simply not enough to apply a standard deviation (a basic concept that everyone ignores). But since they are very close to each other you could do one of two things; a)Use a mean value with an outlier envelope, or b) Note that one set of obs is inconsistent with the four others, bin it, and conclude that the others are so close one can say the average represents reality; no envelope required. With the model outputs you still can’t use the standard deviation because these are not real results – they are far too dependent on parameter bias. If a random sampling had been used for the inputs then the outputs might pass the basis for assuming a normal distribution, but then the output spread would be so large that it would be equally meaningless. So all you can really do is plot the forecasts, leave stats out of it and use your brain.

    Similarly the Douglass et al results should have been a statistics free zone. All you need to do is plot the obs and results together and reject the models that are obvious rubbish. The big problem is that the ensemble-is-best argument totally undermines what should be the main effort, which is to to separate good models from bad ones. In that regard IPCC and RC are both putting a roadblock in the path of sound scientific methods. Worse still for science, is that I’m sure that is their deliberate intention because being vague is much more advisable when providing policy support, ideally vague enough so that you can never be proven wrong. Much of this current back-and-forth about model capability is like watching “Yes Minister” but without the jokes.

    [I disagree with that. You can validly use SD for either. You just have to know that the thing that is varying isn’t the same in the two cases. If you apply a test for “are the statistics of these two series the same” you should expect it to say “no” -W]


  10. I think Mark Hadfield put it pretty well. Since the spread of the models in the eight year predictions is so large, the main thing it tells that just eight years of measurements is not a very good measure whether climate models work or not. And also that the models don’t do very useful predictions for eight years. They make more useful predictions for longer ranges.

    And in the big picture, the whole community and IPCC is worried about longer time scale phenomena anyway.

    I don’t know how much Roger really understands of all this, as he seems to have no mathematics back ground. But boy he sure can go around accusing people and get everyone to listen to him.


  11. Roger did an undergraduate major in mathematics, a fact he proclaims quite proudly in the train wreck thread. He has now finally agreed that “that 8-year trends in global average surface temperature provide a weak test for the IPCC ensemble of AOGCMs.” I’m not sure whether this is full circle but I agree with Mr. Connelley that the Real Climate post is quite clear and convincing.

    The comments thread on the RC post now has another train wreck in progress. Gerald Browning jumped in with a recommendation that we all read and heed Pat Frank’s article in Skeptic. Browning is another Colorado mathematics/weather modeling emeriti weighing in on climate models. (Is it something in the mountain air?) He is convinced that the climate models are nonsense and proved it mathematically a long time ago.


  12. First lets start with the notion that Lucia cherry picked the start date. In the IPCC simulations the start date for the “projections” was 2001. If you want to test the claims of the projections what date makes sense to pick as a starting point for your observation data set? 1892? 1964? 1979?

    I don’t see how there is any credible claim of cherry picking. The climate of the past hindcast GCM runs ended in 2000. The “projection” runs started with 2001 as their first year. So, if you want to test the models skill in “projecting” you start in 2001. It wasnt Lucia’s choice. It was the IPCC choice.

    It probably was an unlucky choice.

    Now, here is an interesting question. If the IPCC had said
    for the first decade of this century the warming trend will be .2C per decade, and if the warming trend were .4C per decade ( after collecting data for 7 years) what would people say?

    oh, yes, I think the warming trend should be about .2C for 2001-2011. presently the data doesnt support that conclusion very well. But I wouldn’t give up that belief just because of a bad patch of weather. neither would I ignore or discount the current “inconsistency” or wave it away.

    The interesting bits are always found where thre is divergence between expectation and observation.

    Like humour.


  13. hi Bloom,

    Yes, O-C is rarely used. That however is not the claim I was questioning. William said it was unheard of. Well, its not.

    There were other sources and other references, but that is beside the point. The point is there are cook book technqiques,typically from econometrics, that could be of use.

    Could be. So it might be good thing ( MIGHT) if climate science had a look approaches developed in, say, econometrics.

    In fact, there are efforts to this.

    The point is this. dismissing Lucia because she used a method William had not a clue about, is just boys club garbage.

    is o-c the right method? i dunno. explain why ols is best


  14. Now, here is an interesting question. If the IPCC had said
    for the first decade of this century the warming trend will be .2C per decade, and if the warming trend were .4C per decade ( after collecting data for 7 years) what would people say?

    We’d all be saying “it’s short-term variability in the climate system”, just as we are now.

    The difference, of course, is that lucia, CA, and the rest of the denialsphere would be saying the same thing, while now of course they’re saying the IPCC projections have been proven wrong to a 95% confidence level.


  15. “We’d all be saying “it’s short-term variability in the climate system”, just as we are now.”

    Balls. You’d be saying “we clearly understimated warming and now were doomed. DOOMED! And those lying denialist LIARS are lying again with their short-term variablity lies. They’re in their denialist blogs right now laughing. Laughing and lying!”

    Or summat like that.


  16. you got that right Alan.

    The issue is that warmists do not want to specify in advance of testing any of the following.

    The predictions they will validate against.
    The statistical test they will use.
    The alpha.
    The beta.

    It’s all ad hoc.

    Quite simply, the GCM make predictions, say of GSMT, that cannot be resolved for many years. so skill cannot be assessed. so action can’t be taken.


  17. But you do not expect them [climate models] to track year-to-year variability – if you run 10 GCMs, they will all have ENSOs in different years, for example. … -W

    In which case, only compare model outputs for those runs that have the ENSOs in the correct place. Then you’ll have much lower variability and you’ll quite likely find that the last 8 years are inconsistent with the model predictions.

    Any model that has a long-term trend mixed with low probability, high-amplitude events will have inherently high variance. The way you resolve the problem and get better (ie reduced variance) estimates is via importance sampling of the form I just described. (Non-climate) modelers have been doing this for decades.

    But of course, the huge variance in the climate models suits the alarmists (and the realclimate sneer brigade) because they can then make the lofty (but ultimately vacuous) claim that the models are consistent with almost any observed climate, at least over the next couple of decades while they try and impose their environmentalist agenda on the rest of us. So don’t look for any serious variance-reducing studies of the models anytime soon.

    I love the way James and William and co sneer from their ivory towers, when in fact it is they who do not have the basic statistical understanding.

    [You still don;t understand. ENSO is only an example. The point is that there is, and there is expected to be, year-to-year noise -W]


  18. You still don;t understand. ENSO is only an example. The point is that there is, and there is expected to be, year-to-year noise -W

    That’s pretty funny. I understand this an awful lot better than you think, William. Modeling is my field.

    Yes there is year-to-year noise. But it is mostly the result of big amplitude effects like ENSO. Those could easily be factored out giving much lower-variance trend estimates. Or, as I said in my previous post, fix the ENSO dates but otherwise leave the runs the same (if you can’t manipulate the model at that level, just pick the runs with ENSOs at more-or-less the correct time. If there are other big sources of variance, pick those to match the observations over the last decade.)

    These are computer simulations: you can generate as many as you like. If you really cared whether the models confirm the last decade’s cooling, you’d do these kinds of experiments.

    Relying on the enormous variance in the models to claim consistency with widely varying scenarios is – to put it charitably – scientifically lame. If you think it isn’t, then I have a new theory of gravitation for you, way better than the old ones. It’s consistent with General Relativity, Newton’s laws of gravitation, *and* Ptolemaic epicycles. What more could a good scientist want?


  19. Not that anyone will see this post again, but I was tracking down the “Megan the undergrad” Pielke trainwreck, and saw mugwump’s last post:

    “These are computer simulations: you can generate as many as you like. If you really cared whether the models confirm the last decade’s cooling, you’d do these kinds of experiments.”

    Hee! “Generate as many as you like”! “Modeling” may be mugwump’s field, but “climate modeling” clearly isn’t. How many hundreds of years of computer time would it take to generate sufficient runs to get a few runs with ENSOs at more-or-less the correct time? *falls down laughing*



Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s