Frankly, I’m having difficulty making sense of your posts. I have the feeling that you are not a native speaker of English.
As far as computer modeling goes, I don’t see how an increase in computer power undermines my point about survivor bias.
The fact is that you could take a computer 10 times as powerful as the most powerful computer available today; put together a climate model; and have it turn out wrong – in prediction, postdiction, or both.
So clearly there’s still a potential problem with survivor bias.
Guilty as charged, when in a hurry the problems I have do show up, but in my experience calling attention to that is a symbol that there is not much to say against my main points.
The history (already posted) of how computer power was used to find the evidence of GW and then AGW, shows that survivor bias was not much of a factor.
Does that means you only want to continue playing games? Get the evidence to deal with the study not just your say so. Clearly you did refer to most or all of the study.
I can’t respond to a point that I don’t understand. I don’t know anything about you in real life, so the fact that I questioned whether you are a native speaker of English should suggest to you that I’m not dodging your arguments, I simply don’t understand some of the points you are trying to make.
In fact, I urge you to consider the possibility that you yourself don’t understand many of the points I have made and you aren’t really responding to many of my arguments.
How?
Sorry, but you are the one who is playing games. It was a simple and reasonable question.
No I did not. I referred to a particular subset of the study.
I don’t think so, because others besides me also found fault on your methods, the evidence is here.
As for my grammar I said before that it is a crime against nature, all other dopers do mind from time to time and I do thank them when they are specific on what did I do wrong.
Related to the matter at hand, you are only contradicting yourself now. So I will wait for your evidence, no need to discuss more when nothing but your opinion against AGW is what is left to defend.
For I was here also to learn from the evidence and unfortunately the lessons from the anti AGW faction here concentrated in showing that it did not matter if their evidence was wrong, it only counted that it defended their rhetoric. I prefer to defend the idea with the most evidence, not just empty rhetoric.
I have argued with numerous people in this thread. You might have a point if I had made the same claim with respect to all, or even most of those people.
But I haven’t.
Why do you think that is? Are your arguments just so uniquely brilliant that, rather than respond to them, I luckily guessed that you are not a native English speaker and used that as an excuse?
I think the problem is that you don’t even realize that you don’t understand the points I’ve been making. So you conclude that it must be “empty rhetoric.”
For example, if you had understood my point about “survivor bias,” you would see that the amount of computer power available is completely irrelevant to the point.
The point was that I did not see evidence of the “survivor bias” as the reason climate models were judged. What I see so far is an evolution, the earliest models to check for the CO2 effects in the atmosphere came to be because the calculations to find if CO2 was a problem could not be made just with pen and paper, the computer’s results then had little to do with survivor bias. And now more computing power is applied to the issue, it seems to me that bigger errors can be easily found now by critics thanks also to the increase in computer power available, the fact that only small variations (and no computer debunking of the CO2 effects in the atmosphere) are the counter evidence so far presented by the opposition is telling. As it is becoming routine here, dismissing the current effectiveness of computer modeling as “survivor bias” remains your say so.
In a nutshell it means that some researchers in the field will find statistically significant results by chance and then use those mistaken results. Now, Applying this to climate science I have to conclude, based on the evolution of the models, that “survivor bias” could be relevant to the discussion if only **some ** researchers had confirmed the evidence of AGW and that then for the modelers that would be the it. We need to assume no more checks were made, or the data was not confirmed, or that the formulas used are not checked again and by many. Indeed, to assume “survivor bias” is here is too assume too much. (and this is why generalizing and dismissing AGW modeling with “survivor bias” is not effective)
In reality most researchers and modelers in different countries came out concluding AGW was an issue.
In a nutshell, it means that some researchers in the field will find statistically significant results by chance and then use those mistaken results.
Now, Applying this to climate science I have to conclude, based on the evolution of the models, that “survivor bias” could be relevant to the discussion if only **some ** researchers had confirmed the evidence of AGW and that the modelers then decided to call it a day. We need to assume no more checks were made, or the data was not confirmed, or that the formulas used are not checked again and by many. Indeed, to assume “survivor bias” is here is too assume too much. (and this is why generalizing and dismissing AGW modeling with “survivor bias” is not effective)
In reality most researchers and modelers in different countries came out concluding AGW was an issue.
First of all, they are based on well-understood physical principles (such as Newton’s equations of motion). Second, they are checked against how much fidelity they have in reproducing the current general climate, e.g., distribution of temperature and pressure both across the globe and in the vertical, to name just one thing. The amount of data available to check against provides many more degrees of freedom than the number of parameters so that it is impossible to fit to all these things.
Also, as has been pointed out, numerous groups have independently developed climate models and the modeling results can be compared with each other. They are also compared against specific events like the eruption of Mt. Pinatubo. And, in at least one case (climateprediction.net), the predictions of the climate sensitivity have been studied by systematically varying the parameters that do exist in the model over physically-plausible ranges and seeing how much the climate sensitivity changes…with the conclusion being that low sensitivities are almost impossible to get. Most values of the parameters yield “moderate sensitivities” (of, say 2-4 C per doubling of CO2) and there is a tail going off at high sensitivities with a small range of parameters actually having climate sensitivities above 8C (although many people argue that climate sensitivities this high can be ruled out from paleoclimatology).
They are not statistical models and the concept of “survivor bias” has no intelligible meaning when applied to climate modeling. All you are proving by talking about this is that you have no idea what you are talking about…which makes it not at all surprising that your beliefs are in conflict with almost all climate scientists and the conclusions of scientific authorities such as the National Academy of Sciences.
As far as faster computers go, they allow for higher resolution in the models which means more fidelity in reproducing the actual physical processes, as well as resolving topography, land vs. water, etc.
Do you agree that, in general, it’s better to test a prediction ex ante than ex post?
Do you agree that if a set of climate models is tested by comparing the individual models to historical data, the researcher should disclose both failures and successes?
It depends on the type of modeling being done. Obviously, all things being equal, it is nice to be able to test it ex ante. However, it is not always realistic to do this because of the time you must wait to verify whether the prediction was accurate or not…and it is also not always really necessary as I have explained above.
They do. If you read most papers on climate modeling, the authors will discuss what aspects of the climate the model reproduces well and what aspects it does not do as a good a job on.
Well, it varies with what sort of model you are using. In the limit that you are using a purely statistical or phenomenological model and essentially fitting to the data and, particularly in the limit where the number of parameters is not very small compared to the number of degrees of freedom in the data, then it is quite vital to test the model on data that was not considered in the fitting. In the limit of a purely mechanistic model with no parameters used to fit to the data, it doesn’t really matter whether the data was there at the time you developed the model or not…although there is still usually something psychologically-fulfilling, if nothing else, when your model can actually predict data that was not even available at the time it was developed.
Climate models are much more toward the latter limit than the former limit. They do have a few parameters for those processes that cannot be treated in full-blown detail at the available resolution but those are not tuned to fit to the data of interest (e.g., historical global temperature record) because there are many other important basic things to get right in terms of basic climatology or of the basic behavior of the specific process being parametrized. In other words, the number of degrees of freedom in the available climate data are many times larger than the number of parameters. Furthermore, most of the parameters wouldn’t tend to tune very effectively how the model would predict the historical temperature record anyway because they are not designed for that purpose.
I’m not sure in what context you are talking about failures. I am talking about it in the context of, say, giving some details of what aspects of the climatology the model seems to simulate well and what aspects are still problematic. Another example would be that if you run the model several times with perturbed initial conditions and get curves with somewhat different wiggles as a result, then you should not just choose to show the one that gives the closest agreement with the historical temperature record to show, say, but you should show them all or show the ensemble average.
My feeling though is that you are talking more about failures in the sense of purely statistical studies, such as are done in medicine, where if you are testing a drug, you can’t just cherry-pick the time that the drug gave a good result and leave out the times when it did no better than the placebo. This sort of issue is one that is certainly very relevant for those sorts of studies or modeling but is essentially irrelevant for climate modeling.
Assuming that the historical data was not considered in the fitting (or modeling), then, are you saying that the only benefit to testing ex post is psychological fulfillment? Why should scientists care about psychological fulfillment?
In this context it means testing a model against historical data and getting a poor post-diction. Are you seriously claiming that you didn’t understand that?
Here’s the question I asked before, which I will ask again:
Well, I am not saying they should necessarily. But, like I said, it’s a continuum and that is one end (and the end that climate models are pretty close to although probably not completely at).
Well, as I explained, in the context you are talking, this doesn’t really make sense. It is the sort of thing that is important to do for statistical modeling but not for a mechanistic physical model. In mechanistic modeling, it is more important to look at things that are somewhat analogous to this but different. For example, it is good to study and report how sensitive your result is to the parameters that do exist in the model…i.e., do you still get a good postdiction for other values of the parameters. Or what happens if you take a certain process out of the model (like Soden et al. have done in regards to the water vapor feedback in modeling the cooling following Mt. Pinatubo or various groups have done in terms of various natural and anthropogenic forcings when they look at the global temperature record over the last hundred years or so)…then how does it change how well it represents that historical data?
And, like I said in my post, to the extent that the model get slightly different results when the initial conditions are perturbed, it is important not to cherry-pick a particular member of the ensemble that is better than the others at postdiction.
These are all things that are well-understood in modeling such systems.
