Thanks to Bill Shockley I’ve been made aware of new paper by James Annan about recent developments in Bayesian estimation of climate sensitivity. It covers many things I’ve been wondering about, so I found it very interesting. I don’t need to say much, since it’s pretty readable, but I’ll highlight a few things.
In Bayesian analysis one has to specify a prior, either a Subjective prior, or what is called an Objective prior. Early Bayesian analyses of climate sensitivity used uniform priors which were intended to be objective, but they gave unrealistically high weights to high climate sensitivities. There are other Objective priors (such as the Jeffrey’s prior used by Nic Lewis) and as James Annan’s paper says
[t]he priors used by [20–22] decay strongly for high values of S, and the resulting posterior pdfs have median estimates a little lower than 2oC with 5–95 % credible intervals that vary from about 1–3 to 1–4.5oC.
So, these newer Objective priors resolve the issue with the uniform priors, but bring the mediam estimate to below 2oC. However, as many have pointed out,
[w]hile these results are rather lower than many of those reported earlier in the literature (e.g. [10, 11]), they are not so dissimilar to recent results using the subjective paradigm [1, 17, 26, 30].
So, even the relatively low values returned by some Bayesian analyses using Objective priors, are not that different to results using more Subjective priors. Although James does discuss the problems with uniform priors, what would have been interesting would have seen some discussion of the other Objective priors, such as the Jeffreys’ prior used by Nic Lewis. This appears to peak at 0, which too seems physically implausible. It would be interesting to know more about the implications of this. Presumably, this could partly explain the difference between these results, and those using Subjective priors.
James’s paper also discusses the models that are normally used, which are pretty simply and – typically – assume that the feedback factor is constant. However,
[f]or many state of the art climate models, the “effective” feedback (that is, the value of (
) at a specific point in time) can change, typically (though not always) decreasing in standard scenarios of increasing greenhouse gas forcing [2,5]. A decrease in this feedback implies that the effective sensitivity early in the warming will be lower than the equilibrium sensitivity, and this suggests that methods which use the historical period for estimation may underestimate the true equilibrium sensitivity.
Hence, the assumption of a constant feedback factor may not be entirely realistic.
Also, if you combine other estimates, such as those from paleo-climatology and climate models, you get that
these two additional lines of argument both point to a sensitivity around the canonical value of 3oC [9]—perhaps a little higher than estimates based on the observational warming, but certainly highly consistent with them—but each approach has significant uncertainties and inbuilt assumptions.
The paper ends with the following, which seems quite reasonable to me
While estimates based on the recent observational record are increasingly converging to a moderate value with a best estimate rarely far from 2 to 2.5oC, and a range which is confidently bounded between about 1 and 4.5oC (or less), these estimates are themselves conditional on approximations that are now recognised to introduce significant additional uncertainties (and perhapsa bias) into the results.
There’s more to it, but those are just some things that caught my eye. One comment I will make, is that if this was not such a contentious topic, that one can get reasonable agreement using various different estimates (some complex, some simple) would normally give some confidence that we’re in the right kind of ball park. However, since it is quite a contentious topic, the tendency is – unfortunately – to pick an estimate (and sometimes just a value) that suits the message that is being presented. Maybe, as James’s paper suggest, we should be considering the
possibility of synthesising different lines of research, all of which inform on the equilibrium sensitivity. Such an analysis has the potential for generating a more precise and credible result.
...and Then There's Physics 
Something that I meant to add to the post, but forgot, is that the Equilibrium Climate Sensitivity (ECS) is specifically defined so as to include fast feedbacks only (water vapour, clouds, lapse rate). It doesn’t include the slow feedbacks. Hence, even if ECS is somewhere between 2 and 2.5oC, this does not mean that we cannot warm more than this might suggest given that we might expect some slow feedbacks to also contribute to the overall warming.
I don’t really like the term “objective prior” for the Jeffrey’s prior, calling it an “invariant prior” would be a rather better description of what it actually achieves (invariance under re-parameterisations, e.g. changes in the units of measurement).
There is a natural temptation in science to associate “objective” with “good” and “subjective” with “bad”, which normally has some merit, but in this case it is rather more subtle. If we genuinely had no idea of the value of climate sensitivity, and only the particular set of observations to work from, then it would be entirely sensible to use a Jeffrey’s prior. But that is not the position we are in, we do actually know quite a lot of the physics, and we have other sources of information, e.g. paleoclimate, which also provides information. So while it would be “objective” to use a Jeffrey’s prior, it doesn’t actually represent our actual state of knowledge, and if it contradicts what we actually know (e.g. zero is not a plausible value for CS), then it is obviously flawed. A “subjective” approach, using an informative prior would give a better representation of what we know a-priori, and more plausible result, but it would be contingent on acceptance of the prior, which is the problem with informative priors (subjective or objective). Annan’s suggestion of combining multiple approaches is a promising approach, if you used a Jeffrey’s prior, but used a likelihood that covered both observational and paleoclimate data, that would be a much more satisfactory “objective” approach.
As I mentioned on a previous thread, the value of those models that provide estimates at the lower end of the range is that they pretty much demonstrate that climate sensitivity cannot plausibly be less than 1, i.e. negative feedback is pretty much ruled out. That seems like progress to me. What we really need though are studies that rule out high values for CS that are currently within the plausible range.
Dikran,
Thanks. I thought of mentioning some of that, but the post was getting long and I assumed someone would highlight that issue. The point you make about ruling out an ECS below 1C is a very good one. Jeffreys’ prior peaks at 0, so if it rules out an ECS less than 1C, that seems like a pretty strong constraint. As you say, trying to further contrain the high-end may well be the way to go. That may even be consistent with what James’s paper suggests at the end – combining different methods to try and further constrain it.
“which is confidently bounded between about 1 and 4.5oC (or less).”
I am no statistician.
However, doesn’t this widen the range from 1.5 – 4.5C by 1/2 a degree C?
We have had 30 years to narrow the range from 1.5 – 4.5 and we end up at 1 – 4.5?
I would love to narrow the range – but is seems like a pretty tough nut to crack given our 30 years of failure.
RickA,
I think he’s just using different terminology. The 1.5-4.5oC range is the likely range, which means at least a 66% chance of lying in that range. What I think he’s saying there is that we can be even more confident than that that it does not lie outside 1oC to 4.5oC. In fact, the full IPCC statement is
So, I think the paper here is saying the same at the bottom end (i.e., extremely unlikely less than 1oC) but maybe a stronger statement about the upper end (i.e., something like very unlikely greater than 4.5oC).
About 3C. Read the post.
ATTP
We crossed – I was responding to RickA.
Or put another way, RickA – sensitivity is not low enough to provide any policy wriggle-room on emissions reduction. And then there’s ocean pH.
We know enough to know that we have to do something about the CO2. End of story.
We haven’t done that for a while 🙂
Only because I haven’t been posting very much 😉
“There’s more to it, but those are just some things that caught my eye. One comment I will make, is that if this was not such a contentious topic, that one can get reasonable agreement using various different estimates (some complex, some simple) would normally give some confidence that we’re in the right kind of ball park. ”
Yup.
RickA We have only had 30 more years worth of additional observations in that time, which isn’t as much as you might think, just about enough to reliable estimate the rate of warming, provided the forcings are relatively stable. We have also found more things we didn’t know we didn’t know, which nevertheless need to be taken into account. Same interval, better reasoning/understanding is still substantial progress.
One issue that I do find noteworthy relates to this paragraph in James’ paper which you’ve cited yourself:
Ref [1] and [30] (Adrin and Skeie et al) are methodologically similar to each other and suffer both from the problem of (considerably) underestimated NH temperatures with their fitted model. With more recent OHC data, their model will most likely also suffer from underestimation on yet another “front”. Having a posterior ECS mean of 0.9-3.2K [30] is hence beyond possible and imho two rather questionable references to back up the above-mentioned statement. Reference [17] and [26] are in the usual 2.5-2.8K ECSballpark and as such not particularly supportive of the argument either.
I tend to be a bit underwhelmed by work that seriously suggests ECS values <2K. But then again, that's just my 50 irrelevant Cent 😉
Nope, Karsten, it just sounds as though you know what you are talking about 🙂
> [W]hile it would be “objective” to use a Jeffrey’s prior, it doesn’t actually represent our actual state of knowledge, and if it contradicts what we actually know (e.g. zero is not a plausible value for CS), then it is obviously flawed.
This is not so dissimilar with what Radford Neal said:
http://www.cs.toronto.edu/~radford/res-bayes-ex.html
This, in turn, is not so dissimilar with what was discussed around then:
I’d rather speak of Speedo-Bayesians, which is not so dissimilar with “pseudo-Bayesian.”
I’m sure Josh would too.
ATTP,
thanks for the mention. JCH actually posted the paper and BBD made me aware of James Annan as a person who may have first recognized the primacy of a 3C ECS estimate, i.e., even before Hansen championed the paleo conclusion. Credit where it’s due…
My 2 cents is ECS is 3C +/- 0.5C because paleo data say so. According to the statistics and confidence of the measurement data. I would really appreciate if someone could explain to me why +/- 0.5C doesn’t mean what it says in this context, i.e., where are the hidden assumptions.
Also, regarding sensitivity varying with forcing strength, Hansen has this graph from this paper that looks interesting but is a bit technical for me.
bill shockley
It’s basically about water vapour feedback and temperature. From Hansen et al. (2013):
Oops. If ATTP or the duty mod 🙂 would be so kind as to de-bork my html I’d be very grateful.
Thank ‘ee kindly.
Thanks BBD, I couldn’t tell if he was talking about a modeling experiment or if there was anything in there about the real world, and especially the modern world that we’ve seen so far. My guess is yes to the real world, no to what we’ve seen so far.
I posted a link to the Annan paper here because you guys understand this stuff far far better than I do and I was hoping to see some discussion on it, so this is like an early Christmas.
Annan et al have great photographs on their website, but I don’t think he has ever mentioned this paper. I found it searching Google Scholar for new papers.
The other one I linked to placed ECS at 3.9 – I think.
Here is the other one:Jet Propulsion Laboratory
bill
I’m sure I’ve read something about this enhanced sensitivity effect wrt the Cenozoic hyperthermals but I am damned if I can find the reference. IIRC along the lines of enhanced sensitivity makes the amount of carbon released to trigger the hyperthermal more geophysically plausible (smaller) than some estimates.
Maybe you’ll find it in the reference section of the Hansen paper? I think I’ll benefit from a slow reading of the paper. Thanks for the help.
RTFRsFRs 🙂
Good thought.
I try to be humble. I guess I need to try harder.
Not on my account 🙂
The way I see it, we have 4 main ways of estimating climate sensitivity, please point out if I missed something:
1) Simply look at what GCMs say
2) Palaeoclimate: see what happened in the far past.
3) Emergent constraints: diagnose relationships between sensitivity and physical processes, then combine observations of the physical processes with GCMs to get information about sensitivity.
4) Observational energy-budget based: combine observational time series of, e.g. temperature or its pattern of change with simple models and estimates of forcing. May be Bayesian or non-Bayesian.
1) and 2) aren’t really distinguishable from each other, and tend to favour values of around 3 C for a best estimate of equilibrium response. 3) often tends to favour higher sensitivity (e.g. see papers by Laue, Sherwood, Dessler, Fasullo and the Zhai paper JCH linked to), 4) tends to favour lower sensitivity.
But it’s very hard to be sure they’re measuring the same thing. 1) uses global air temperatures, but as far as I remember method 4) always relies on some observational temperature series and as Cowtan et al. (2015) showed, these are expected to underestimate global air temperature change. If we apply 1) and 4) to the same system, 4) will therefore always report a lower value than 1).
And that’s before you go into issues like state dependence and how the response seems to change with time… If you put together some of the papers since ~2012 by Knutti, Soden, Armour & Gregory then it’s very believable that the methods used in 4) will underestimate the equilibrium response.
The only ways I can see that they would underestimate the climate response are if 1) forcing estimates are biased in the right way or 2) there’s stronger internal variability that’s well correlated with forcing changes. This would need to be demonstrated and quantified, and shown to be of similar size to the likely negative biases in order for the methods in 4) to provide convincing evidence to shift down estimates of climate response.
Mark R,
The advantage of the paleo derivation of ECS is we see the full response to a known forcing, so a lot of the guesswork is removed. Certainty is relative to our confidence in the proxies. Models have a wide formal certainty range, but intuitively, because some are so mature and well understood, the intuitive confidence is much better. I believe Bayesian theory can be understood as trying to formalize this intuition. Same way statistics formulize intuition.
This is why Hansen places paleo knowledge at the top of the knowledge hierarchy. It’s no coincidence that his 3.0C paleo number matches Annan’s 3.0C Bayesian number.
Maybe an expert can help.
Does the Jeffries prior pile up at 0 just because 0 is an arbitrary lower bound? On the face of it 0 might seem reasonable, but if there is such a thing as a ‘no feedback’ sensitivity then 0 already implies negative feedback.
The divergence in the prior at zero then creates a sharp gradient in the prior at lower positive temperatures.
If we were to allow a Jeffries prior with negative sensitivities then it may actually result in a higher posterior estimate as the data kills the lowest sensitivities and the prior would be much more uniform at higher sensitivities. Only if the peak in the prior is in just the right place is the peak in the posterior low? Doesn’t in itself mean it’s wrong, but I think it does go to robustness.
If we do rule out negative sensitivities then on what grounds? Physics? In a nonlinear, non equilibrium system, physics is enough to say that an ‘objective’ totally non informative prior is absolutely rules out negative values? But if you allow enough physics to say sensitivity must be positive why not incorporate more, or more sophisticated physics? Or is there a compelling argument that says a sensitivity of -1 femto K (Jeffries prior = 0) is in principle more unphysical than +1 femto K (Jeffries prior peaks)?
At any rate, I think it’s worth making explicit what the assumptions here are, and a lower limit of 0 seems to be one of them.
“the Jeffreys’ prior […] appears to peak at 0 […]. It would be interesting to know more about the implications of this.”
By construction, the Jeffrey’s prior is non-informative in the sense that the prior has no effect on the posterior — or in the sense that the posterior is fully determined by the likelihood.
Therefore, the implications of using a Jeffrey’s prior are exactly zero. The implications of not using a Jeffrey’s prior are non-zero, of course, so switching from a uniform to a Jeffrey’s prior is not without consequences.
JK,
Ii don’t have a deep or extensive knowledge of statistics and yet that didn’t keep me from surmising from a rudimentary experience sample the connection between intuition and formality and their order of creation. I know of at least one expert whom I think would agree with me.
Anybody can figure out the equations and do the tests, but it takes a good intuition, a good feel for why things might be happening, to notice some of the more subtle effects that can change the result entirely
and I would surmise further that this generalization can be extended to almost anythng. Science begins with a hypothesis and proceeds to proof. Art begins with an idea and proceeds to form. Even a search dog is given a clue before it’s set loose. It’s good to know what you’re looking for before you go searching.
I’m sure you mean well, but thanks anyway.
Richard,
You’re just playing semantic games again. Clearly using a Jeffreys’ prior that peaks at 0, gives a different result to what you would get if you used a subjective prior that was both small at 0 and small at high S. You do get this, don’t you? We are trying to understand actual reality here, not some kind of statistical reality that doesn’t properly represent the real world.
Come, come, ATTP, don’t let’s be naive. In certain quarters, the objective is not to understand physical reality, it is to obtain a specific type of result.
BBD,
Every now and again I make the mistake of giving the benefit of the doubt 🙂
@-“In certain quarters, the objective is not to understand physical reality, it is to obtain a specific type of result.”
Don’t be so cynical.
The goal is much more high-minded. The objective is not to understand physical reality but to conform to a dogma or ideal of mathematical/statistical form. This is the reason why economeretrician have been unable to make accurate predictions of the real world economy, it is more important that the mathematical equations that are meant to describe reality are ‘correctly’ formed and structured than that they are skilful in describing the real world.
@-“Science begins with a hypothesis and proceeds to proof.”
To be pedantic, it is mathematics that goes from hypothesis to proof. Science goes from speculation to theory. Very occasionally it gets as far as a Law…
izen writes: “… it is more important that the mathematical equations that are meant to describe reality are ‘correctly’ formed and structured than that they are skilful in describing the real world.”
In economics this may be true of the freshwater (Chicago) school of economists, but the saltwater (MIT) school is interested in models that reflect reality.
So, for instance, Paul Krugman can tell us more about reality with a simple IS-LM sketch than many economists can in 10,000 words.
Given the topic of the last few comments, it seems worth suggesting that people listen to the podcast that Grant McDermott highlights in this comment.
izen: To be pedantic
Not at all, this being a technical discussion. There’s an attractiveness to precise language.
MarkR,
Hansen’s oft-used basis of understanding slide from this video [10:20] and also an excerpt from this paper
Paleoclimate, changes of climate over Earth’s history, provide valuable insights about the effects of human perturbations to climate, even though there is no close paleoclimate analog of the strong, rapid forcing that humans are applying to the climate system. International discussions of human-made climate change (e.g., IPCC) rely heavily on global climate models, with less emphasis on inferences from the paleo record. A proper thing to say is that paleoclimate data and global modeling need to go hand in hand to develop best understanding — almost everyone will agree with that. However, it seems to me that paleo is still getting short-shrifted and underutilized. In contrast, there is a tendency in the literature to treat an ensemble of model runs as if its distribution function is a distribution function for the truth, i.e., for the real world. Wow. What a terrible misunderstanding. Today’s models have many assumptions and likely many flaws in common, so varying the parameters in them does not give a probablity distribution for the real world, yet that is often implicitly assumed to be the case.
There’s also this, from Hansen & Sato (2012):
And:
The fact that ice sheet and GHG boundary conditions are actually slow climate feedbacks is irrelevant for the purpose of evaluating the fast-feedback climate sensitivity.
This is something I don’t understand. Please don’t answer the question if you’re going to whine about it. 🙂
Bill,
If I understand what you’re asking, the ECS is formally defined as the Equilibrium Climate Sensitivity with fast feedbacks only. These are typically water vapour, clouds, lapse rate. It doesn’t include things like ice sheet retreat, and permafrost release. The reason that Hansen’s estimate of 3C +- 0.5C is an ECS is because he essentially includes albedo changes due to ice sheet changes as a forcing, rather than as a feedback. If you treat that as a feedback, rather than a forcing, then you get the ESS, which is higher than the ECS. Hope that is what you were asking.
> [T]he Jeffrey’s prior is non-informative […] Therefore, the implications of using a Jeffrey’s prior are exactly zero.
There’s a premise missing, RT.
ATTP yes, exactly. Neat trick. I’ll have to do the math myself to really get it. Thanks, kindly. 🙂
It doesn’t include things like ice sheet retreat, and permafrost release. The reason that Hansen’s estimate of 3C +- 0.5C is an ECS is because he essentially includes albedo changes due to ice sheet changes as a forcing, rather than as a feedback.
Err…
JCH,
Errr…? You might have to explain. If you consider what BBD quotes, then the 6.5W/m^2 is the combined change in forcing due to GHGs and surface albedo.
If the zero lower bound is absolute it should correspond to no CO2 in the atmosphere. Since it assumes an arbitrary concentration it is arbitrary and not objective
aTTP – it sounds contradictory, but I just need to slow down and read as it apparently is not.
Bill Shockley,
For me the palaeo evidence 1) shows that the consensus sensitivity estimates are plausible, and 2) undermines the implied sensitivities behind much of the science denial movement says.
But there’s also evidence that sensitivity is state dependent, and if the climate sensitivity today is different from what it was in the past then the comparison is harder. Iirc Dan Lunt has done some nice studies trying to calculate how things like changing geography affect climate response.
Well, yes, in the sense that in hot climate states with a diminished or absent cryosphere, ECS is much closer to ESS as the slow ice sheet feedback is effectively absent.
As far as I know, nobody has argued that the empirical estimate derived from 800ka of Pleistocene climate (Hansen & Sato 2012) is affected by state-dependent sensitivity.
So the palaeoclimate evidence pointing to a present ECS of ~3C (best estimate) is unaffected.
@wotts
Not playing semantics. The Jeffrey’s prior is that prior that does not affect the posterior — by definition — the rest is math. If the Jeffrey’s prior looks strange to your eyes, that is because you are not properly calibrated.
bill shockley
As it turns out, I didn’t find it there but I *did* find it in the references in Lund et al. (2013) 🙂
It was Kiehl & Shields, (2013).
Richard,
Whether or not you’re playing semantics, the Jeffreys prior peaks at 0. This is almost certainly unphysical. Therefore the posterior is clearly influenced by a prior that appears to be unphysical. If you don’t get this, maybe you should consider that your basic calibration is physically unrealistic.
Richard Tol wrote: “By construction, the Jeffrey’s prior is non-informative in the sense that the prior has no effect on the posterior — or in the sense that the posterior is fully determined by the likelihood.”
No, that is achieved by the uniform prior (in which case the posterior is just the likelihood). The Jeffrey’s prior is uninformative in the sense that it maximises the divergence between the prior and the posterior (i.e. it mazimises the information provided by the data and minimises the effect of the prior in a particular technical sense). Note Jeffreys’ himself didn’t consider the prior as uninformative, but minimally informative. I’d agree, as it encodes the prior knowledge that the analysis should be invariant to some re-parameterization (e.g. a change of units).
“Therefore, the implications of using a Jeffrey’s prior are exactly zero.”
No, that is clearly incorrect, using a Jeffreys’ prior in situations where it contradicts prior knowledge is likely to have rather negative implications. “Objective” is not synonymous with “good”.
“The implications of not using a Jeffrey’s prior are non-zero of course, so switching from a uniform to a Jeffrey’s prior is not without consequences.”
Yes, in this case a uniform prior (even though doesn’t provide invariance and it is informative for a scale parameter) gives a better representation of our prior knowledge, and hence more plausible posterior. That is also a consequence. There are negative consequences as well, but that is why statisticians need good intuition/common sense as well as mathematical rigour.
RT’s best buddy chimed in a while ago:
Click to access jeffreys.pdf
MarkR,
Modeling science shows to me that people go with what they know. Hansen’s been jumping up and down for years proclaiming the superiority of paleo ECS estimates, and it’s nearly obvious he’s right. But modelers don’t think like that. They don’t have those tools. Margin of error isn’t as clearly defined in modeling. They’re not dishonest or stupid, they’re just like everyone else who ever had training and an occupation. There’s a million examples of this and it would make a good blog post, but one that I like is Joe Romm vs Jim Hansen in 2009. Romm was a technology policy wonk in the federal government for many years (trained in physics) and then when he became a climate advocate journalist he favored government hands-on direct management of technology issues to manage emissions. Things like the Princeton stabilization wedges. Romm was adamant and threw down on Hansen, denouncing a carbon tax. I like Romm and think he’s a smart guy. People just think they know more than other people when they are extremely expert in one area. They’re more focused on what their tool can do than on finding the best tool to accomplish a task. American foreign policy is like that — who can we bludgeon next?
Hansen won the Romm argument without even responding. Nowadays, seems about every 6th article on ClimateProgress has to do with a carbon tax. Hansen is extremely frustrated but he thinks it’s coming. Reminds me of the civil rights days in the 60s.
Thanks BBD, I’ll put those 2 on top my slowly compressing pile. And thanks for your answer to MarkR about climate state, I was wondering about that too.
ATTP
“Jeffreys’ prior peaks at 0, so if it rules out an ECS less than 1C, that seems like a pretty strong constraint.”
and
“the Jeffreys prior peaks at 0. This is almost certainly unphysical.”
Not so. You really don’t properly understand Bayesian estimation. But you are not alone – nor do the vast majority of climate scientists.
It isn’t the Jeffreys prior that rules out ECS much lower than 1 K, it is the likelihood function, which has an extremely sharp cutoff at the low ECS end.
Nic,
The point I was making was that an ECS of 0 is unphysical. This has nothing to do with Bayesian estimation. We have prior information that indicates this.
Yes, I realise this, but that the prior peaks at 0 still means that you give more weight – through the prior – to these low ECS values.
Also, if you are going to go around highlighting other peoples lack of understanding of Bayesian statistics and – I assume – your supposed expertise, maybe others can go around highlighting your lack of understanding of basic physics and of the scientific method. It works both ways, Nic, but maybe you think that appeals to your authority somehow trump other peoples appeals to their authority. If so, why would you think this?
Nic,
In fact, I think I will respond to this
and simply change it to
If you give this a little thought (which I suspect you won’t bother doing) you might even get the point I’m trying to make.
niclewis wrote “It isn’t the Jeffreys prior that rules out ECS much lower than 1 K, it is the likelihood function, which has an extremely sharp cutoff at the low ECS end.”
Nobody is claiming that it is the Jeffreys prior that rules out ECS lower than 1K, quite the opposite. The point being made was that even with a prior that favours low ECS (at least more so than the alternatives that have been considered) the likelihood still overpowers it sufficiently to effectively rule out an ECS below 1K. That is why it is a useful lower bound that everyone can agree on, even if they are not comfortable with the Jeffreys prior.
BBD
“There’s also this, from Hansen & Sato (2012):
The empirical fast-feedback climate sensitivity that we infer from the LGM-Holocene comparison is thus 5°C/6.5 W/m2 ~ ¾ ± ¼ °C per W/m2 or 3 ± 1°C for doubled CO2.”
Or this more recent estimate of ECS = 1.7 C using the same calculation, but better estimates of temperature and forcing changes:
“A first order estimate of the equilibrium climate sensitivity could be generated from the ratio of temperature change to radiative forcing. Our new temperature anomaly of 4.0±0.8 C, combined with estimated forcing of 6–11Wm−2 (Annan et al., 2005; Jansen et al., 2007) would suggest a median estimate for the equilibrium climate sensitivity of around 1.7 C, with a 95% range of 1.2–2.4 C.”
From – guess who – James Annan ( http://www.clim-past.net/9/367/2013/cp-9-367-2013.html )
So much for observationally-based ECS estimates using instrumental-period data being rules out by paleoclimate estimates. And before you look for some earlier period paleo ECS estimate, I’ll point out that IPCC AR5 concluded that lECS estimates from earlier than the LGM were difficult to directly compare with ECS in today’s climate state.
Nic,
You’ve done this before. You continually leave out the next bit. I’ll – once again – quote the relevant section in full
This post is also relevant.
==> “…nor do the vast majority of climate scientists.”
Interesting scientific assessment from an (non-activist) independent scientist.
Nic, could you describe the methodology you used to make that assessment?
And while you’re at it…
Since you’ve described the “vast majority” of climate scientists as not properly understanding Bayesian estimation, it shouldn’t be hard for you to pick a few names as illustrations. I notice a number of well-known climate scientists were involved in writing the following paper:
Click to access 2010%20Hargreaves%20WIREs%20Clim%20Chang.pdf
Would you include any or all of Annan, Mann, Schmidt, J. Nieslsen-Gammon, G. Foster, and Hargreaves along with the rest of that “vast majority” who don’t understand Bayesian estimation? If so, could you describe how you assessed the level of understanding of the individuals in that group?
Thank you, ATTP.
@ Nic Lewis
Palaeoclimate won’t go away, however much one might wish it to. It is unlikely that sensitivity is low enough to provide policy wriggle-room. And then there’s ocean pH.
Speaking of “self-professed lukewarmers”:
http://www.theguardian.com/science/blog/2015/nov/18/antarctic-ice-sheet-collapse-sea-level-rise-whats-new
Perhaps if Tamsin says what AT keeps saying will it lukewarmingly sink in?
Larry Wasserman chimes in:
Willard, he must not understand… probably needs to consider a career in climate science.
Nic Lewis – intentional or did you misinterpret Annan?
ATTP
“More recently, Hargreaves et al. (2012) used the relationship found in the PMIP2 ensemble between the tropical temperature change at the LGM, and equilibrium climate sensitivity, to estimate the equilibrium climate sensitivity to be around 2.5 C with a high probability of lying under 4 C, although this result is subject to several important caveats.”
Yes, I know Annan went on to say that, but it is not relevant to the apples-to-apples comparison I was making, which was to Hansen’s simple estimate.
Also, you haven’t made the relevant points re Hargeaves et al (2012) that:
a) the 2.5 C estimate uses a subjective Bayesian method with a prior having a median of 3.1 C. Their underlying non-Bayesian, regression based estimate has a mean of 2.3 C;
b) they state that dust cooling omitted in the PMIP2 experiments biases the results, and attempt to correct for this, reducing both the regression and the Bayesian ECS estimate to 2.0 C.
c) they used a method that fails when applied using the more recent CMIP5 models – maybe you are ignorant of that finding.
Nic,
It is relevant to our understanding of the climate system, something you appear to not yet understand. The goal of science is to gain understanding, not blindly apply a method so as to obtain a result that you then promote without much thought. Your ability to apply Bayesian methods may be beyond compare, but your understanding of the scientific method, and why we do research, appears sorely lacking.
Hmmm, this also appears rather ironic
Maybe I’ll rephrase this to
Care to comment?
ATTP
“Also, you haven’t made the relevant points re your own work
a) the 1.6 C estimate uses a rule-based Bayesian method with a prior having a peak at 0C.
Care to comment?”
Wrong. You are showing your ignorance of my work, as well as of the logicality of the Jeffreys’ prior.
Nic,
And you’re illustrating your ignorance of the scientific method, or a lack of understanding of best scientific practice.
Two questions.
1. You use a Jeffreys prior. The Jeffreys prior in your work peaks at 0. Is this correct?
2. Do you think an ECS of 0C is physically implausible?
Guys:
This is math. You can argue all you want, the equations won’t change. Jeffrey’s prior is Jeffrey’s prior.
Dikran:
Jeffrey’s prior is uniform IN THE APPROPRIATE DIMENSION, and therefore often non-uniform in the parameter of interest.
Richard,
Yes, I realise this. This still doesn’t mean that Jeffreys’ prior does not peak at 0C. Are you trying to illustrate the point that many are trying to make? Simply properly applying some kind of technique does not mean that what you get is necessarily a reasonable representation of reality. This is essentially physics, not econometrics. The goal isn’t to blindly apply a technique and to do so without any arithmetic errors; the goal is to gain understanding of how our climate will respond to changes.
I think the non-informative prior used by Nic Lewis in Lewis (2014) is shown in slide 12a in the PDF of his Ringberg presentation “Pitfalls in Climate Sensitivity”), as opposed to the distribution shown in the Fig.4 of the paper itself, although I may be wrong – I don’t properly understand the technical details of the paper. The former, however, has no peaks at 0 sensitivities, and is similar in form to what Pekka Pirilä suggested would be a reasonable choice of a subjective ‘expert’ prior.
Where is Pekka by the way, I hope all’s well with him.
oneuniverse,
If that is the case, then Nic could have simply said so. The figure below is Figure 4 from his paper and is described as Noninformative (Jeffreys’) joint prior for S and Kv for inference from TA and CH likelihoods.
ATTP
Yes, but your wrote:
“the 1.6 C estimate uses a rule-based Bayesian method with a prior having a peak at 0C.”
Lewis (2014) doesn’t estimate ECS at 1.6 C – it is a methods paper that uses data from an old attribution study and gets an ECS estimate a bit above 2 C. And the prior shown is to illustrate the point that a prior can be highly peaked but nevertheless noninformative.
Richard Tol wrote “Dikran:Jeffrey’s prior is uniform IN THE APPROPRIATE DIMENSION, and therefore often non-uniform in the parameter of interest.”
When you make a mistake, such as writing “By construction, the Jeffrey’s prior is non-informative in the sense that the prior has no effect on the posterior”, it is better to admit it, rather than double down. The only prior that has no effect on the posterior is the one that is uniform on the parameter itself.
Bayes rule gives us
P(theta|data) = p(data|theta)p(theta)/p(data)
only if p(theta) = 1 does the prior have no effect on the posterior, as then p(theta|data)=p(data|theta)/p(data) and the prior is absent entirely.
“Guys: This is math. You can argue all you want, the equations won’t change. Jeffrey’s prior is Jeffrey’s prior.”
No, the issue is one of science, not maths. The Jeffreys prior is the Jeffreys prior, but it still does not represent our state of knowledge, and an ECS of 0K is still unphysical.
Nic,
Wow, you really are making it very difficult to keep to my promises. Seriously, do you do this on purpose, or do you really not know how to engage in a worthwhile discussion? I really can’t tell, but it seems that you are utterly incapable of having a discussion in which you don’t end up focusing on an irrelevance. The 1.6C really wasn’t the point. The point was that your prior is highly peaked at low climate sensitivity.
So, can you answer the questions. Does the Jeffrey’s prior peak at 0C? Do you regard an ECS of 0C as physically implausible?
Joshua, really good paper about IPCC, modelling, certainty and Bayesian theory. Still reading… 🙂
Explains (to me) a lot about why you don’t find “Hansen” linked to “Bayesian” in a google search. Thanks for posting.
niclewis,
I’ll point out that IPCC AR5 concluded that lECS estimates from earlier than the LGM were difficult to directly compare with ECS in today’s climate state.
Could you provide a link? I should point out that Hansen defied this advice and, in fact, went ahead and narrowed his original certainty of +/-1.0C to half that by doing exactly what the IPCC (purportedly) said not to do.
Or this more recent estimate of ECS = 1.7 C using the same calculation, but better estimates of temperature and forcing changes:
Better or worse? You neglected to support your contention.
More recent… by how much? Is a nominal year significant?
oneuniverse,
Someone has just emailed me a link to this. Unfortunately it seems that Pekka passed away on the 24th of November. My condolences to his family and friends.
That is sad news indeed, my condolences to his family and friends are added to ATTP’s.
Richard Tol:
And then there’s physics. Our state of physical knowledge tells us that an ECS of zero is physically implausible.
Nic,
You wonder why I sometimes get a little frustrated discussing things with you. I’ve looked through some of your papers. This says
However, it appears to have a prior that does not peak at zero, but from here it suggests that The plot is probability-averaged over all values for aerosol forcing, which was also being estimated, accounting for the turndown in the prior at low ECS values. It might be nice if you could explain that and the significance of that.
This paper – as you point out – has median ECS values above 2C, but appears to have an prior that peaks at 0K. However, there is no mention of probability-averaging this over the aerosol forcing.
This paper has a best estimate for ECS of 1.66K and apparently uses the same method as your earlier paper, but doesn’t appear to show the form of the prior.
So, it seems that a number of your recent papers do indeed produce best estimates for ECS of around 1.6K and appear to use an objective Bayesian approach which – if I understand it – uses a Jeffreys prior. Given this, are you at all interested in discussing the implications of this, or not? For example, do you accept that an ECS of 0C is physically implausible?
ATTP, thank you, I’m very saddened by that news.
I guess I’m stupid, but if perfect math in an observations-based analysis of reality arrives at an implausible answer, haven’t they sort of proven the observation are incomplete? If so, can’t you just plug in 3.0C and get Hansen’s answer for where a complete set of observations of the system should be?
That’s very sad news, ATTP. I add my condolences.
JCH,
Technically, if you arrrive at an implausible answer, then that would be true. On the other hand, Nic Lewis’s analyses do not produce implausible results. If anything, they are still largely consistent with other estimates. Given the various assumptions, that some of the datasets may suffer from coverage bias, that the prior may give undue weight to low ECS values, one might argue that it provides some confidence that we’re in the right kind of range. I suspect that if this wasn’t such a contentious topic, that is how people would intepret this. James Annan’s paper seems to hint at this kind of idea.
> Jeffrey’s prior is Jeffrey’s prior.
And yet this identity leads you to claim that “the implications of using a Jeffrey’s prior are exactly zero.” There’s something missing in your inference, RT.
This is logic, BTW. Not maths.
I notice that Nic Lewis deftly avoided acknowledging what I actually wrote earlier, so I will repeat it:
Please, Nic, be sure and tell the good people at the GWPF of their error.
Hooking on the final words,
“possibility of synthesising different lines of research, all of which inform on the equilibrium sensitivity. Such an analysis has the potential for generating a more precise and credible result.”,
dikranmarsupial made an important point early on in the thread which I think deserves to be repeated: “objective” is not equivalent to “good”. It is useful that everyone can agree and that it is fully transparent how you derived at the numbers, but you cannot really escape the judgement part anyway (not even Richard Tol! 😉 ), and it is still rather easy to make explicit what is subjective or not. It seems to me that progress of the issue tends to be a bit blocked by this fallacy.
For example, no-one seems to really have taken on the famous IPCC “likely” ECS estimate of 1.5-4.5K which also have been pretty invariant over decades. One standard division of lines of evidence (LOE) is the one MarkR gave above.
1. GCM runs
2. Paleo evidence
3. Observed feedback analysis
4. Observed energy balance modelling
“Likely” usually means 2/3. Assume just for the argument that the best assessment of each individual LOE is a likely 1.5-4.5. Now, if unrealistically, all the LOE:s are deemed completely dependent on each other (e.g. a bit like 4 different teams used the same data and model), the combined assessment becomes a likely 1.5-4.5. If instead, a bit more realistically, they are fully independent, the 1.5-4.5 interval becomes an 99% (“Virtually certain”, from 1-1/3^4) interval which gives a hint that truly different lines of evidence really is having a strong effect. I would judge that in the best analysis they are at least more independent than not. Using more correct LOE:s interpreted from IPCC, PALAEOSENS (2012), Dessler presentations, etc., the 1, 2, and 3 should probably have the lower end raised from 1.5 to 2 or so. A resulting 90% (“Very likely”) 1.5-4.5 interval seems pretty reasonable (even if I didn’t do any detailed calculations).
In any case, I’m curious if anyone can present a similar analysis landing near the IPCC assessment.
Speaking of maths, RT, here’s a question that involves additivity. In a recent comment, Nic said that:
http://www.bishop-hill.net/blog/2015/10/5/puffed-rice.html?currentPage=3
One of these teams was referred to it by “Otto et al.” The “et al” of that team comprises Nic himself. Which means Nic counted himself in two different teams.
Would you say that there’s double accounting going on, RT?
***
More interestingly, this was in response to James’ remark that:
http://www.climatedialogue.org/climate-sensitivity-and-transient-climate-response/#comment-901
Nic’s response to this comment was that he’s not claiming that all work on climate sensititivity is wrong. It’s rather all work except his and the one of three other teams, one in which he belongs.
Yet another exchange where Nic picks on a nit, i.e. James’ “one.”
That would stand to reason, as the “vast majority” of climate scientists don’t properly understand Bayesian estimation.
Presumably, we can determine whether climate scientists are wrong about Bayesian estimation on the basis of whether or not they “properly” understand Bayesian estimation, and we can presumably determine whether they “properly” understand Bayesian estimation on the basis of whether or not their findings are in agreements with Nic’s (I would say are consistent with Nic’s, but I think that would be less accurate)…
Barring further information from Nic about his methodology of assessment, it seems that there’s little to go on here other than Nic’s appeal to his own authority.
Perhaps Nic could mansplain this to AT and the climate science community:
ATTP
Very sorry to hear that Pekka has passed away.
Of course I think an ECS of zero is physically implausible. But the question has no relevance to the shape of the Jeffreys’ prior.
I’m afraid I don’t have time to answer the multiple other questions directed at me. Let me recommend you and your denizens to read the book Bayesian Theory by Bernardo and Smith, also texts by, Jeffreys, Jaynes, Phil Gregory, Box and Tiao, Zellner, etc.
Bill Shockley
“Could you provide a link?”
Go to the IPCC AR5 WG1 website, obtain Ch. 10 and look at the penultimate paragraph of section 10.8.2.4.
> [T]he question has no relevance to the shape of the Jeffreys’ prior.
This would imply that the shape of Jeffrey’s prior has no relevance to a discussion of the implications of the fact that an ECS of zero is physically implausible. Or, as AT said in the post:
Or, as AT said in the comment to which Nic responds:
Clarifying the “this” would reduce wiggling room.
Nic writes:
And directs our attention to AR5 WG1 10.8.2.4 (penultimate paragraph), where we find this:
But this leaves unaffected the ~800ka of Pleistocene climate from which Hansen & Sato’s empirical estimate is derived. There’s plenty to go on there, and the evidence from the relatively recent geological past strongly suggests an ECS most likely value somewhere around 3C.
As I keep saying, too high to allow for policy wriggle room and that’s even if we ignore the effects of ever-increasing CO2 on ocean pH.
One wonders what the GWPF is thinking of, really.
* * *
For the interested reader:
Lunt, D. J., A. M. Haywood, G. A. Schmidt, U. Salzmann, P. J. Valdes, and H. J. Dowsett,
2010: Earth system sensitivity inferred from Pliocene modelling and data. Nature Geosci., 3, 60–64
Pagani, M., K. Caldeira, R. Berner, and D. J. Beerling, 2009: The role of terrestrial plants in limiting atmospheric CO2 decline over the past 24 million years. Nature, 460, 85–88
Royer, D. L., 2008: Linkages between CO2, climate, and evolution in deep time. Proc. Natl. Acad. Sci. U.S.A., 105, 407–408.
Royer, D. L., R. A. Berner, and J. Park, 2007: Climate sensitivity constrained by CO2 concentrations over the past 420 million years. Nature, 446, 530–532.
Willard
Yes, it would, which makes NL’s apparent reluctance to engage all the more puzzling.
BBD
“Estimates of ECS from other, more distant paleoclimate periods”
If you read the preceding paragraph of AR5 10.8.2.4, which is concerned just with LGM studies, you might realise that this means “more distant than the LGM”.
BBD,
If you read GWPF’s pamphlet, you can see that this:
Click to access A-Sensitive-Matter-Foreword-inc.pdf
does not cohere Nic’s assumption that there is a best estimate. There are 97 occurences of “best estimate” in that pamphlet. Sometimes it’s “best estimate” and some other times it’s “best observational estimates.” Some other times, it’s
– best observational evidence [an interesting pleonasm]
– best observational studies
– best observationally-based estimate
– ‘best observational’ estimate
Best. Observational. These two words somehow seem to go hand in hand in Nic’s work.
***
It’s also interesting to see when Nic associates himself with Otto and al, and when he doesn’t.
Paleo estimates of sensitivity strike me as ridiculous.
We lack accurate or precise estimates of albedo for even today.
Estimates of albedo from the paleo record with no measurement or even proxy would appear to be pure speculation.
Yeah. They are inconvenient, aren’t they?
Still, an argument from your incredulity should suffice to dismiss them.
Science? Why bother when you can just ask TE?
Nic
Well, my reading of the text is as I commented above: more distant than the Pleistocene. Not the LGM, specifically.
Or do you suggest that state-dependent sensitivity has changed during the last ~800ka? If so, can you point to a reference?
To be more correct, I should have written ‘more distant than the Mid-Pleistocene Transition. Not the LGM, specifically’.
From AR5 WG1 10.8.2.4:
The reason I detailed the four references AR5 WG1 10.8.2.4 provides to define what it means by ‘more distant paleoclimate periods’ was to illustrate that it does not refer to the last ~800ka. The most recent of those more distant palaeoclimate periods referenced was ~3Ma. The next, ~24Ma, then 420Ma and on to deep time.
Nic,
I realise that it has no relevance to the Jeffreys’ prior, but it does have relevance to the general idea of Bayesian analysis. I may not be an expert at Bayesian analysis, but I do know that one of the strengths is the ability to incoporate prior information. So, you still haven’t answered my question as to whether or not the Jeffreys’ prior in your analysis peaks at 0C. The obvious reason for asking this is that if it does, then that would suggest that it provides more weight to very low ECS values than is physically plausible.
So no chance that you could amaze us with your obviously expert understanding of this extremely complex topic?
As I understand it, this figure summaries the various paleo ECS estimates.
Willard
It is. For example, I can’t recall Nic’s ringing endorsement of this statement by Otto:
That’s not much policy wriggle room.
> Paleo estimates of sensitivity strike me as ridiculous.
TE,
If we add to your strinkingness Nic’s characterization of “observational-based evidence” (that pleonasm again) as
wouldn’t it be ridiculous to contest the IPCC’s claim that there’s no best estimate for ECS too?
Many thanks!
BBD,
If you read Otto’s report at the MET Office’s, you might have noticed this claim:
http://www.metoffice.gov.uk/research/news/alex-otto-article
If you read the GWPF’s pamphlet, you may also have noticed:
That’s an interesting way to interpret scientific development.
Besides, would you think that returing to the previous ECS’ range is “good news”? Here’s Otto’s response:
http://www.metoffice.gov.uk/research/news/alex-otto-article
Do you think it’s possible to bury good news that is not really news?
Willard
More ermine than weasel 😉
Willard,
Indeed, there seem to be plenty of examples of people putting these kind of studies into what seems to be quite a reasonable context.
You want context, AT? If you read what Otto says about the AR4, you’ll see this:
http://www.metoffice.gov.uk/research/news/alex-otto-article
Where Alexander says “difficult,” Nic & Marcel say “suspect.”
I suspect it may be difficult to say if Nic & Marcel’s “suspect” is suspect.
You guys are rude to guests. Even Willard, but especially the host.
Wwzd
People are treated according to the merits of their arguments and the level of intellectual honesty displayed in presenting them.
You of all people should know that, Steven.
Steven,
Really? I thought I was more restrained than I have been in the past, and probably more restrained than was deserved. I would argue that the guest (who wasn’t really a guest) you’re probably talking about could try harder to actually engage in a discussion, rather than appealing to their apparent brilliance and everyone else’s ignorance. It does get rather tedious after a while and does give the impression that they’d rather dodge the questions and nitpick, than actually engage in anything serious.
This was a particularly annoying response. Also, having gone back to find that, I’m also struggling to see where anyone was particularly rude. Is it possible that you’re becoming a bit like those ex-smokers who suddenly become rabidly anti-smoking?
Pingback: Use of Jeffreys prior in estimating climate sensitivity - Statistical Modeling, Causal Inference, and Social Science
Steven,
one comes into another’s territory knowing (or at least having the opportunity of knowing) ahead of time what the climate is like and how “dissenters” are treated — in the case of ATTP’s blog, I would say with uncommon fairness and consideration. The moderator makes the rules. By entering, you agree to play by them. It goes without saying that nic is aware of the treatment he will get, by having been here more than a few times.
Andrew Gelman has provided some thoughts. Be interesting to get Nic’s views about Andrew Gelman’s post.
Climate scientists are too stupid…
I expect most people have seen James Annan’s posts on ‘objective’ priors, but here are the links anyway, for completeness’ sake:
Coverage.
Or, why Nic Lewis is wrong.
Objective probability or automatic nonsense?
Poignant to see Pekka in the comments.
While I’m thankful for Moshpit’s concerns, playing the ref is boring. Oilman (!!!!!!) is actually helping me find gentler ways to deal with snobbishness. Cordonnier mal chaussé perhaps, but so far his advices are good.
I’ll remind everyone that RT is, well, RT and that Nic’s first comment was:
Please don’t play the ref furthermore.
BBD
It’s more accurate to say that people are treated according to one’s perception of their arguments etc. However, perhaps your comment provides insight into Nic Lewis’ evaluation of how (and whether) to respond to the comments here. This kind of thing works both ways.
Off to get my popcorn before reading them 😉
oneuniverse,
I agree, it does. I would like to give everyone the benefit of the doubt and to always be polite and reasonable. However – as Willard points out – starting with “you really don’t properly understand Bayesian estimation” doesn’t help. Technically, it’s a true statement, but I don’t think one has to understand something in exquisite detail, in order to discuss it.
oneuniverse
Read the Annan links if you think NL’s approach merits the confidence NL has in it. Interestingly, NL’s confidence in his methodological choices does not extend to a defence in depth.
BBD,
Yes, true, but in a sense, I’m certainly not trying to trash his method, so it is a pity that Nic seems reluctant to discuss this in any depth. I think what he does is quite interesting and that the results are reasonable, given the caveats (and if you consider the overall result, rather than simply the best estimate). The problem is more that Nic doesn’t seem to want to discuss the possible caveats with his work. Virtually all scientific analyses have caveats. It’s not a bad thing to discuss them. It helps to improve overall understanding.
==> “You guys are rude to guests.”
Selective concern is, well, selective.
ATTP
That is because he is a polemicist at heart. Worse, he is collaborating with such as the GWPF to push misinformation into the public discourse, which means he is well beyond the pale as far as I am concerned.
Even if true, by responding with a polemic like this, you are choosing to play by his rules.
“Selective concern is, well, selective.”
All behavior is selective. The issue isnt whether or not the concern is selective. of course it is.
The point is pretty simple. I saw the same thing at WUWT, at climate audit (piling on), at Judith’s, even at Lucia’s when willard showed up.
As I explained to Mc one day, when a visitor shows up to defend something they have done or explain it, the group tends to use a very bad tactic. They start a barrage of good questions, questions with insults attached, stupid questions, rudeness, ( man I slammed curry the first time she showed up) basically everybody goes into their standard mode of engagement.
All the while there may be one or two critical questions that everyone wants answered, but since there are 52 questions on the board, and a few slams or insults, the visitor can effectively avoid the 2 inportant questions by answering other questions… or by responding to insults or by taking off in a huff and saying that a scientist should not have to put up with this stuff.
It happens everywhere.. at stupid skeptic blogs and at smart physics blogs. It’s not about you
its about how relatively insular groups of people treat “outsiders” and its about the relative effectiveness of this.
As someone who has polluted the dialog I’m well aware of the methods both intentional and unintentional. That is why I asked WWZD.. what would zeke do?
That’s all.. carry on.
But I would like to hear marsupials comments on how we can incorporate our knowledge that ECS cannot be zero ( it physicall cant ) into a prior?
Set aside Nik’s beliefs..
Then after that I’d like to learn more about paleo from BBD…
This has been done, hasn’t it? There are analyses that use subjective priors.
As I understand it, reference priors are formed by maximising e.g. the Kullback-Leibler divergence between the prior and the posterior, so in principle we could follow the same procedure, but with the constraint that p(0) = 0. However reference priors are complicated enough to discourage most statisticians from using them in the first place, and there is no guarantee that the problem is tractable. Nic Lewis is probably the best person to have a go at implementing this, and it would address that criticism of his approach and hence improve its acceptance.
Personally I would probably opt for a subjective prior, but one that addressed the criticisms of the existing subjective priors, as the objective weakly informative prior is likely to be more effort that its worth. Perhaps something like a log-normal with a mode of about 1C, which is the “no feedback” theoretical warming from a doubling of CO2, and some suitable dispersion parameter sampled by MCMC?
The real point is that in a situation where we actually do have substantial prior knowledge, a posterior from an analysis with an “uninformative” “objective” prior needs to be interpreted with some care, and certainly can’t be taken as the most plausible or satisfactory analysis. We do have prior knowledge, and much of it suggests that ECS is rather higher than the estimates provided by Lewis’ analyses. How much higher is a mater of conjecture at this point, but it needs to be acknowledged that it is likely to be higher. However like ATTP, if find Lewis’ work interesting and a useful contribution, if nothing else it provides a lower bound we ought all to be able to agree on!
“This has been done, hasn’t it? There are analyses that use subjective priors.”
of course, then again I;m not sure subjective priors exist. Outside my knowledge that ECS cannot be zero I’d be hard pressed to specify what my prior was . Now I might pick something
and then defend it as ‘my prior’ , Let’s put it this way. what is your subjective prior, where does it peak?
ECS cannot be zero ( it physically cant )
There are ways other than surface temperture rise that can restore radiative balance.
( lapse rate change without surface change, albedo, etc. ).
But these are not observed nor very likely compared to processes which warm the entire troposphere.
ECS is highly likely to be positive, but could physically be zero or negative.
==> “The issue isnt whether or not the concern is selective. of course it is.”
As you have been mistaken about many times in the past, you have no authority to determine what “the” issue is. You are certainly entitled to determine which issues are of most importance to you. I happen to think that the selectivity in your concern about how people interact in climate-o-spheric exchanges is a good example of how the basic discourse in that forum is characterized by bad faith dialogue.
Next, judging a website as some sort of collective entity seems more a projection of motivated reasoning than anything else. Websites are not coherent entities. If you have a constructive criticism, with some specificity, to offer to Anders or anyone else, have at it. That might bring about an outcome (perhaps improved discussion between Anders and NIc) that could have some merit. Observations about a website as a group amounts to tribalism – and will not bring about an outcome of value (unless you value sameosameo).
Of course all behaviors are selective, to at least some degree. However, not all behaviors are equally selective.
==> “As I explained to Mc one day, when a visitor shows up to defend something they have done or explain it, the group tends to use a very bad tactic. They start a barrage of good questions, questions with insults attached, stupid questions, rudeness, ( man I slammed curry the first time she showed up) basically everybody goes into their standard mode of engagement.”
It’s up to the “visitor” (of course, the distinction between “visitor” and “denizen” is basically meaningless other than as projection) to determine how s/he wants to engage with that. What you call a “visitor” could simply engage with the good questions and ignore the others. Blaming the decision-making of the “visitor” on the reaction from “the website” is not holding the “visitor” accountable for his/her own actions. We all have the option of being adults here.
==“All the while there may be one or two critical questions that everyone wants answered, but since there are 52 questions on the board, and a few slams or insults, the visitor can effectively avoid the 2 inportant questions by answering other questions… or by responding to insults or by taking off in a huff and saying that a scientist should not have to put up with this stuff.”
See above. People who respond as you described are not, IMO, likely to be motivated by good faith exchange. It is incredibly easy to ignore slams and insults and to focus on good faith exchange on interesting and meaningful questions. Nic certainly had the opportunity to do so here. If he chose not to, that’s on him. People can always find an excuse for their lack of good faith exchange if they’re looking to find such excuses.
==> “It happens everywhere.. at stupid skeptic blogs and at smart physics blogs. It’s not about you
its about how relatively insular groups of people treat “outsiders” and its about the relative effectiveness of this.”
Wrong. Group dynamics exist. We can choose to react to them in any variety of ways. As an individual who knows that blogospheric exchange is characterized by juvenile identity-related behaviors, and responds accordingly, or as someone who likewise, allows identity orientation to strongly influence his/her own behaviors. It’s about individual accountability, not the banality of group behaviors.
BTW, with respect to the dynamics of group behaviors, my own personal favorite of Nic’s comments was the following (as I indicated earlier):
==> “…==> “…nor do the vast majority of climate scientists.”…”
Perhaps you might consider talking to Nic about accountability for how individuals interface with group behaviors?
[Chill, please. – W]
[Let’s stop this, please. -W]
Steven Mosher wrote “Let’s put it this way. what is your subjective prior, where does it peak?”
In principle, you don’t have to specify where it peaks. It is possible to construct a heirarchical prior, where the prior itself just dictates the form of the prior (e.g. log-normal), but the parameters of the prior are not directly specified, but instead they are treated as nuisance parameters and integrated out using a hyper-prior (which would normally be an uninformative one with no hyper-hyper-parameters… ;o). In some situations this marginalisation (integrating out) can be done analytically, but there are good software tools for this sort of thing, like BUGS or STAN (neither of which I have had time to play with), for when it can’t.
TE, take a moment to think.
ECS <0 means that temperature falls as radiative forcing rises, and vice-versa.
What implications would this have for the climate system?
Stephen,
Excuse my ignorance, but would not the most obvious way to proceed be to build a prior based on the pdf from am alternative methodology for ECS eg paleo?
Or have I misunderstood something fundamental?
vtg,
I think that is what this paper discusses. There is a whole section about expert priors.
Rules-based priors like Jeffreys’ aren’t really priors in the conventional sense. The point, as dikran noted, is to maximize the influence of the experiment according to the Fisher information (there are other objective priors using different information measures), and this doesn’t always lead to physically reasonable results.
Jeffreys’ prior depends, entirely, on the expected variance distribution of the experiment. Lewis has to use models to derive the error distribution, and, as his conclusion states, “there are many other subjective choices to be made, such as the observables, datasets, error distribution assumptions and model”. So I think Lewis hasn’t really removed subjectivity, he’s made it more opaque by shuffling it off into the model that gives the error distribution.
@-“ECS <0 means that temperature falls as radiative forcing rises, and vice-versa.
What implications would this have for the climate system?"
Large volcanic eruptions would warm the climate for a few years.
But I am still trying to envisage conditions in which –
"lapse rate change without surface change,"
Is possible.
@-" So I think Lewis hasn’t really removed subjectivity, he’s made it more opaque by shuffling it off into the model that gives the error distribution."
If that has resulted in the lowest possible estimate of ECS which is 'consistent' with plausible models then it at least it provides a good informative lower bound on any future subjective prior.
izen,
Yes, me too. All the energy being transported via evaporation, but that is temperature dependent too, so that doesn’t sound physically plausible to me.
Yes, that’s an interesting point. I did wonder the same myself, but I wasn’t sure if there was some kind of argument against an iterative Bayesian analysis. If you’re essentially doing the same analysis with updated data, can you use the posterior from an earlier analysis as your new prior?
“But I would like to hear marsupials comments on how we can incorporate our knowledge that ECS cannot be zero ( it physicall cant ) into a prior?”
Can’t you just estimate the logarithm of ECS instead? And then take the exponential to get ECS such that ECS > 0?
Thanks for the link to Andrew Gelmans blogpost and paper. I agree with much in his subjectivity/objectivity analysis. A natural demarcation criteria for using a distinction is that it helps. I’d say the objectivity/subjectivity distinction in this discussion doesn’t pass the mark, it mostly misleads.
Then I think you should question the entire approach of “Bayesian approach to climate sensitivity” as described for two reasons, even if it might be technically correct to apply Bayes formula in the right place:
1. Conceptually, no line of evidence is more prior than the other. You just have different pieces of information to combine, which more naturally can be modelled as the issue of combining a number of more or less dependent stochastic variables.
2. Involving the prior before sorting out (including judging) model bias and structural uncertainty makes the analysis and its interpretation unnecessarily convoluted. This is analogous to the common sense when you are doing normal scientific experiments: You try to avoid interfering or letting your prior guesses affect the outcome (aside the model you are using to interpret the data of course, but that could be considered part of the experiment setup).
A consequence of 2. together with the mess of choosing prior is that the actual resulting numbers and pdf:s given as result in Lewis’ and others’ similar studies are very difficult to make sense of. A practical approach is to not take them seriously. (But the studies might of course contain other valuable content.)
ATTP:
I don’t get the same impression. I’m not sure which caveats you have in mind, but in his papers, he writes about some of the issues and caveats related to the choice of prior, including ‘non-informative’ priors. He’s also discussed this at Climate Audit.
He also acknowledges the possible underestimation of equilibrium CS using the EBM method :
“Implications of recent multimodel attribution studies for climate sensitivity” (Lewis 2015)
However, he (like the IPCC AR5 report) is not perfectly consistent in always differentiating between effective and equilibrium climate sensitivity (something perhaps not helped by the shared acronym).
But I am still trying to envisage conditions in which –
“lapse rate change without surface change,”
Is possible.
So, if the lapse rate increases, ( some warming at the surface and greater warming aloft ),
that reduces the imbalance imposed by GHGs ( Soden and Held 2006, LR negative feedback ).
Similarly, if the lapse rate increases, ( zero warming at the surface, but even greater warming aloft than the above case ), that also reduces the imbalance imposed by GHGs.
Given that the opposite has occurred ( lapse rate has decreased instead of increasing ),
and processes act to distribute imbalances, this is improbable. But it remains quite possible.
In case it’s of interest, there’s further discussion by Nic Lewis at Climate Audit of the possible underestimation of EqCS. The main post is reproduced in the “Pitfalls of Climate Sensitivity” linked upthread, but there’s some additional dialogue in the comments.
re: zero-peak prior for sensitivity
I find most of Dikran’s comments here very relevant and helpful. However, in the main post, ATTP wrote that the prior used by Nic Lewis peaks at zero sensitivity, and linked to an earlier statement of Dikran’s (“The most physically unrealistic aspect of the prior seems to be that it suggests the most plausible value for climate sensitivity is zero.”). Figure 4 of “An Objective Bayesian Improved Approach for Applying Optimal Fingerprint Techniques to Estimate Climate Sensitivity” (slide 12a of NL’s Ringberg presentation) appears to show a partial representation of the shape of the joint prior, and, by eye, the peaks are around 2 K or so (certainly not zero, and indeed higher than Dikran’s suggestion above of the no-feedback 1 K). I assume Dikran agrees that his earlier statement was wrong. Also, this seems to make Dikran’s suggestion that NL’s estimate be used as a lower bound less applicable.
looking at that very nice figure of the prior nic used
the following question occurs. As long as the likelihood is non zero at low ECSs, the prior, being higher the lower one goes, will bias the answer toward the low end. It can be small, even very small, but the bias will exist.
Which raises the following point, has anybunny ever use priors that are uniform btw 1 and 6 ECS, 2 and 5, etc? James was trying to kill off Myles Allens flat 0-20ECS not look at such questions.
TE
ECS <0 is not remotely possible. It implies that night should be hotter than day, winter warmer than summer, and switching off the sun would increase the temperature of the earth.
It is simply unphysical.
AT, thanks for the link.
oneuniverse,
I’m trying to get Nic to clarify his prior. He still hasn’t answered my question about the Jeffreys’ prior. As Eli shows, there is indeed an example which indicates it peaks at zero. The figure on slide 12 actually says The plot is probability-averaged over all values for aerosol forcing, which was also being estimated, accounting for the turndown in the prior at low ECS values. I don’t really understand what this means and why the prior shown on slide 12 is different to the one that Eli shows. It would be nice to get this clarified.
TE,
Firstly, you don’t know that it’s decreased (the certainty with which you says things that are probably wrong is quite impressive) and, no, it does not remain quite possible.
I second Steven Mosher’s call for a perspective piece on Bayesian theory by dikran or any ambitious/inspired contributor. Topics could include:
— How would modern Bayesian theory have treated Newton’s law of gravity at the time of its emergence (or any other major scientific theory that would serve as an instructive example) — what would have been the priors, etc?
— What have been the major successes of Bayesian theory? In what branches and types of science has it been the most beneficial and how do those examples resemble or differ from the current case of ECS estimation?
— Comment on the article posted by Joshua covering the IPCC’s suggestion (AR4) that certainty be handled by Bayesian concepts and how that suggestion was spurred by the difficulty in defining certainty for modeling results.
— How Hansen’s approach to modeling differs from the mainstream and why Hansen has never found the need for Bayesian tools and how that is in a way a Bayesian confirmation of Hansen’s methods.
A non-informative prior does not have any information. It also omits the information that the climate sensitivity cannot be zero, or a million K.
If you want to use an informative prior, use one — just call it by its name.
But ff you use an informative prior, you should make sure you only add the information you want to add. The good thing about a Jeffrey’s prior is that it does not move the posterior to a climate sensitivity of 1K, 2K, 3K, 4K, 5K, 6K …
Gelman’s thoughts about weakly informative priors may help.
@-ATTP
” If you’re essentially doing the same analysis with updated data, can you use the posterior from an earlier analysis as your new prior?”
My only exposure to Bayesian stats is a slim book which should have been called ‘Bayes for Dummies’ explaining how Bayesian analysis is used in some research to asses the efficiency of diagnostic tests in the clinical field. That seems to use iterative analysis with improved priors to reach a result.
But I have read it twice and am still deep in D-K land on this issue so would also appreciate a clear exposition of the matter.
However a personal bias makes me highly suspicious that much of this is a matter of various mathematical formalisms being waved around without much relevance to how it might actually improve our legitimate level of knowledge about the system being examined.
Richard,
Yes, I think we get that the non-informative priors technically carry no information – although see Andrew Gelman’s comment that I’ll copy below. However, it clearly influences the result. Hence, it would be interesting to better understand the form of the Jeffreys’ prior that is used by Nic Lewis as it appears that it gives more weight to very low ECS values than may be physically plausible.
What about these thoughts of Andrew Gelman’s
And what about James Annan
Playing semantic games with statistical terminology does not change that the form of the prior is relevant and suggesting otherwise seems utterly bizarre.
@Wotts
“[a Jeffrey’s prior] clearly influences the result”
No, it does not. The result is the posterior. For any prior but the Jeffrey’s prior, the posterior combines the information in the prior and the information in the likelihood. For the Jeffrey’s prior, the posterior only holds the information in the likelihood.
Of course, a switch between any two priors would lead to a different posterior, a clear influence on the result.
Your main objection seems to be that there is information (besides what is contained in the instrumental record), and that therefore the prior should be informative. That is a valid argument against Lewis’ work.
Lewis would counter that some work uses priors that contain information that is not based on anything much.
Richard,
Yes, I know the result is the posterior. The posterior, however, clearly depends on the choice of prior (see below). I’ll even repeat what James Annan said
It’s almost as if you’re suggesting that if the width of a rectangle is 1, then area is fully defined by the length. I hope you realise that that doesn’t really make any sense.
Yes, I meant it influences the result relative to another possible prior.
It’s not so much an objection as simply a point worth evaluating. A strength of Bayesian analysis is the ability to incorporate prior knowledge. It would be interesting to better understand the form of the Jeffreys’ prior used in Nic Lewis’s analysis since it appears that it – compared to other possible priors – gives more weight to ECS values that are probably physically implausible. Also, it’s my understanding that some would regard using reference priors that do not incorporate any prior knowledge as not really being a Bayesian analysis.
His contains no information, apparently, so that would seem to describe his, by definition.
@-Tol
“For any prior but the Jeffrey’s prior, the posterior combines the information in the prior and the information in the likelihood. For the Jeffrey’s prior, the posterior only holds the information in the likelihood.”
This is the claim that strikes those of us ignorant of the finer details of the mathematical formalism of Bayes analysis as incoherent.
If all other PDFs adopted as priors modify the posterior result by including the ‘information’ embodied in the prior PDF, then how can a zero prior NOT also influence the posterior with its unphysical erroneous ‘information’ in an analogous manner?
Some colleagues and myself used a Jeffreys’ Prior once to estimate failure rate for a novel fan design, where the samples ran with 0 failures. So the likelihood was (1-exp(-lambda*t))^n, a function with maximum at infinity.
However, we presented the results as a worst case analysis, since it seemed the Prior biased the fan failure rate to the low side.
That makes me look sideways at Lewis’s assumptions
Oops! Got the form of the likelihood wrong but there was no ML solution for lambda in this case. Must look up my old notes.
> Gelman’s thoughts about weakly informative priors may help.
Three days ago, i.e. on December 5, 2015 at 7:55 pm, RT’s best buddy chimed in:
Click to access jeffreys.pdf
That it’s difficult to choose among noninformative priors seems strange, considering RT’s remark that they have no implication whatsoever. Does it mean that “contains no information” doesn’t imply “is trivial”?
If RT could respond to Wasserman’s remark that “noninformative priors are the perpetual motion machines of statistics,” that would be nice too.
Rcihard wrote “A non-informative prior does not have any information.”
According to Jeffreys himself, the Jeffrey’s prior is not an uninformative prior, it is a minimally informative prior (see e.g. the BUGS book, p83).
> it does not remain quite possible.
Of course it does, AT, just as it remains quite possible that the Chicago Cubs wins another championship or that the Moon is made of cheese. Incidentally, this can be modelled in Bayesian terms:
http://rabett.blogspot.com/2013/02/on-priors-bayesians-and-frequentists.html
Here’s the link to Radford’s presentation:
http://www.faqs.org/faqs/ai-faq/neural-nets/part3/section-7.html
Richard, Bayes rule says:
p(theta|data) = p(data|theta)p(theta)/p(data).
Now p(data) is just a number that depends on the particular set of observations you have, and if you choose a uniform prior p(theta) = 1, then we can write the posterior as
p(theta|data) = k x p(data|theta),
i.e. the posterior is just the likelihood times a constant. So explain to me, why this transfers less information from the likelihood to the posterior than the Jeffreys’ prior, when for the uniform prior, the posterior effectively is the likelihood.
Hint: The problem lies in how you define information, it isn’t as simple as Richard’s posts suggest.
BTW I think Richard’s posts are having the effect of casting more doubt on Lewis’ analysis than it really deserves by being more positive about the Jeffreys’ prior than is justified. I personally think that Lewis’ approach is a good one, and I am glad that someone has taken the time to investigate it. I also think that we should take it very seriously as a meaningful estimate of ECS. However, we should consider its “non-informative” assumption mean that it is likely to underestimate ECS to some extent as the prior information that it does not consider tends to argue for somewhat higher values. So as long as we don’t treat it as being in some way a privileged “best” estimate of ECS, but as just one of a set of estimates, then there is no problem AFAICS.
Willard,
I think I was interpreting the word “quite” more strongely than you are 🙂
Dikran,
I agree. It some very interesting work and should be taken seriously. However, it shouldn’t – as you say – be given some kind of priviledged position because of some statistical terminology that makes it sound unbiased, compared to other estimates.
@dikran
Note that theta and data swap place in p(theta|data) = k x p(data|theta). This sometimes implies that theta in the posterior is measured in a different dimension than theta in the likelihood. You therefore need to apply the Jacobian.
Richard,
Maybe you could respond to the core point of Dikran’s comment which was to do with how you define information.
ATTP:
More light would be helpful. At the same time, I’m aware that my limited understanding of the topic is a serious impediment, and may make a clarification moot – my use of Bayes theorem and prior and posterior distributions began and ended as a teenager being schooled for exams.
Richard:
All distributions must contain information, in the general sense. The Jefferys prior, as far as I understand it (as per Dan Riley’s earlier comment), will by necessity encode some information about the statistical model from which it’s derived.
Richard Tol:
I’m not sure that’s right. When I asked him why he was using a non-informative prior when we already had some prior independent knowledge about CS, his reply was :
In Lewis 2013, he references Bernardo and Smith 1994; Kass and Wasserman 1996 in support of the above.
This is an interesting point. If you perceive Nic Lewis’s analysis as semi-independent of other methods (and I’m not sure this is strictly true, given that some modelling results are required) then you could argue that it provides a semi-independent result that could be combined with other methods to provide some kind of overall assessment of our understanding. I think this is a reasonable way to look at this and is certainly one way in which I do.
On the other hand, you can quite justifiably see Bayesian analysis of data from the instrumental temperature record as an attempt to quantify climate sensitivity using not only this information, but also other information that allows one to constrain the prior. Again, this seems reasonable but, is no longer quite as independent given that the prior is based on other information.
To me, these are both valid ways in which to consider this. Given my current understanding, I don’t think either provides much of an indication that ECS is likely to fall outside something like 1.5C – 4.5C. Okay, maybe there is some indication that the top end should be 4C, but I think there’s also some indication that the bottom end should be 2C, so maybe the range is 2C – 4C, rather than 1.5C – 4.5C. Again, as James Annan’s paper suggests, a proper analysis that tries to combine the various methods so as to better constrain climate sensitivity would be quite useful, and I’m not sure any such thing has ever really been done.
I should probably add, that in the comment above I was thinking of likely ranges (66%, or greater, chance) not absolute ranges (100% chance).
Experience has that a blog discussion is not the best place for a tutorial in math.
My favourite textbooks on Bayes are by Press, O’Hagan and Berger. Others like Gelman & Carlin.
Richard,
What experience would that be? I thought the idea that the area of a rectangle depends on both the width and the length, was pretty straightforward.
A good book on climate physics is Ray Pierrehumbert’ Principles of planetary climate.
dikran wrote:
Nic Lewis writing for the GWPF lobby group:
Clearly, there *is* a problem.
Richard Toll write “Experience has that a blog discussion is not the best place for a tutorial in math.”
That doesn’t mean that you should avoid giving the appropriate caveats on what you write in order to make sure that you don’t mislead those who are not sufficiently familiar with the area to see them for themselves.
BBD,
You’d think the author of the study would discourage others from drawing overly strong conclusions from their research……oh, hold on?
Dikran,
Feature, not a bug?
> Kass and Wasserman 1996
Wasserman being the Larry Wasserman I cited twice above, and who thinks that “noninformative priors are the perpetual motion machines of statistics.” The article can be found here:
Click to access KassWasserman-JASA-1996.pdf
Speaking of articles, there’s one by Frank Tuyl, Richard Gerlach & Kerrie Mengersen (1998) in which we can read:
http://www.tandfonline.com/doi/pdf/10.1198/000313008X267839
Here’s a link to Geisser 1984:
http://www.jstor.org/stable/2683393
Incidentally, Tuyl, Gerlach & Mengersen has been cited in Lele & Dennis 2009:
Click to access Lele%20Dennis%2009.pdf
Barring more information from Nic, I am beginning to wonder if the vast majority of mathematicians and statisticians might also not “properly” understand Bayesian estimation along with the vast majority of climate scientists?
The situation with regard to frequentist statistics isn’t much different ;o)
@dikran
Agreed. But interpreting P(T|CS) as P(CS|T) — the definition of non-information — requires that T and CS are measured in the same space. If not then you need to account for the transformation of space — mathematicians call this the Jacobian, statisticians Jeffrey’s prior.
One way to think about this is the difference between P(T|CS=1) and P(T|CS=2). Is this like the difference between P(T|CS=5) and P(T|CS=6) or more like the difference between P(T|CS=5) and P(T|CS=10)?
Richard, I note your continued absence of caveats.
“One way to think about this is the difference between P(T|CS=1) and P(T|CS=2). Is this like the difference between P(T|CS=5) and P(T|CS=6) or more like the difference between P(T|CS=5) and P(T|CS=10)?”
No, that would be more a statement about the form of the likelihood rather than the prior (hint the probabilities you have given are all likelihoods, not priors or posteriors as it is CS that we are trying to infer).
If you are using a truly non-informative prior you are a frequentist. OTOH, the entire point of Bayesean analysis is to use prior knowledge.
Eli, I’m not sure I agree with the first one. Frequentists use prior information all the time, for instance in choosing the appropriate significance level in a hypothesis test, this (implicitly) considers the prior relative plausibilities of H0 and H1. The problem is that, because it is implicit, it is often neglected, hence the nice XKCD cartoon
The error made here is not adjusting the significance level to incorporate the prior probability of the sun going nova being very very small.
The nice thing about Bayesianism is that sometimes the only prior knowledge you have is that you don’t know something. The difficulty comes from the things you don’t know you don’t know.
Perhaps I should have stated that the quote that starts with “Some difficulties [with Bernardo’s reference priors thing] are that when there are nuisance parameters[:]” was taken from Cox 2006. It also mentions nuisance parameters, which is discussed in a paper with a quite strong proof:
Click to access Clarke1994Jeffreys.pdf
The result seems to apply to cases where there are no nuisance parameters.
While “reference prior” and “least favorable prior” might sound less sexy than “objective Bayes” and “non-informative prior,” it seems to go hand in hand with the lukewarm project of finding the lowest bound of justified disingenuousness.
Richard, you write “But interpreting P(T|CS) as P(CS|T) — the definition of non-information — requires that T and CS are measured in the same space.”
This looks to me to be a misunderstanding of the Jeffreys prior. T and CS have the same definitions and meanings in P(T|CS), P(CS|T) and P(T) throughout within a given analysis, so it is not at all clear what you mean by “requires that T and CS are measured in the same space”. The idea of the Jeffreys prior is that if two experimenters analyse the same set of observations, but use different parameterisations, say one uses degrees Kelvin per doubling for CS and the other uses degrees Farenheit per quadrupling, they will draw exactly the same conclusions (i.e. the posteriors will be the same once the difference in units is taken into account). However, within each analysis the “spaces” over which T and CS are measured in P(T|CS) and P(CS|T) are exactly the same.
Can you give a reference to support your statement (including a page reference)?
Not that I would argue that P(T|CS) = P(CS|T) is the definition of a non-informative prior, but it is a very apt definition of a prior that has “no effect on the posterior”
OT
GWPF “peer-reviews” paid-for academic studies: Greenpeace email sting. That’s gotta hurt.
@dikran
No. This is not an innocuous swap. Just read Press.
@dikran
Or read Wikipedia’s entry on Jeffrey’s Prior. They give the definition, the definition of information, and show the relationship between Jeffrey’s Prior and an Uniform Prior, including the Jacobian that transforms the latter into the former.
Richard, I can see nothing on that Wikipedia page that supports: “But interpreting P(T|CS) as P(CS|T) — the definition of non-information — requires that T and CS are measured in the same space.”
You could argue that there always exists some transformation of the space such that the Jeffreys prior happens to coincide with the uniform prior, but that is not at all the same as “T and CS are measured in the same space”. They key point is that with the Jeffreys prior all parameterisations give the same posterior, but the exact form of the prior will be different in each case as it depends on the parameterisation. Only in the case that p(CS) = 1 will P(T|CS) be the same as P(CS|T), so it obviously isn’t the Jeffreys definition of “non-information”.
Note that the page talks of an alternate parameterisation (so CS is measured in terms of psi in one analysis and theta in the other), but in both cases the analysis is internally consistent, so one uses p(T|psi), p(psi|T) and p(psi) and the other uses p(T|theta), p(theta|T) and p(theta). As you don’t discuss alternate parameterisation of the same problem in two different analyses, but two parameterisations “spaces” within the same analysis, I suspect you may have misunderstood.
Perhaps a specific question would be easier to answer. Where exactly on the Wikipedia page does it “show the relationship between Jeffrey’s Prior and an Uniform Prior”?
TE,
Given that the opposite has occurred ( lapse rate has decreased instead of increasing ),
and processes act to distribute imbalances, this is improbable. But it remains quite possible.
no, it [“There are ways other than surface temperture rise that can restore radiative balance”] does not remain quite possible.
Well, I calculated the RF of a given atmospheric profile and a series of others with 2x CO2 and various degrees of warming aloft, but not at the surface. Here is the result:
Such a result would not seem at all probable, but it is certainly possible.
dikranmarsupial:
That would be the case if all the estimates are of equal value, but surely that’s not generally true. The appropriateness of the method, the quality of the method’s implementation, and the quality and quantity of the data will all have a bearing on the value.
For example, some have argued here that Nic Lewis’ estimate(s) use a prior peaking at a physically-implausible zero C; if that were true, it would bias the PDF downward and therefore, I think most would agree, devalue the estimate. Similarly, James Annan, Nic Lewis, and Steve Jewson have strongly criticised the use in some papers of a uniform prior for sensitivity (with implausibly high ranges), biasing the PDF upwards, again IMO devalueing the estimate.
NL has categorised his criticisms of the various instrumental ECS estimates in slide 13 of his Ringberg presentation. I think this is a useful excercise. Perhaps Nic’s anaylsis is correct and the Aldrin, Ring, Otto, Schwarz and Lewis estimates are better. Perhaps some of the criticisms here are correct, and his estimate is actually one of the worst. In any case, I don’t think all estimates deserve an equal seat at the table, just by virtue of being a member of the set of estimates.
(Hopefully this embeds)
willard, thanks for the link to the Gelman article, I like the idea of the weakly informative prior, and think I was trying to get at something similar in my discussion with Kenneth Fritsch at CA. I guess the criticism would be that it doesn’t have the helpful statistical properties of the reference prior, and at the same time it (purposefully) isn’t an attempt at a full representation of one’s prior beliefs, in which case, what exactly does the posterior represent ?
I think I may have two comments in moderation (one is the duplicate of the other, sorry). There’s no message to say comment in moderation, so I’m not sure.
Ran across an article from earlier this year that quotes John Nielsen-Gammon:
Make your own punchlines.
Related to previous OT – Richard Tol gets a nod – http://energydesk.greenpeace.org/2015/12/08/exposed-academics-for-hire/
@dikran
The Jeffrey’s Prior for the mean of the Normal distribution is uniform. This is because the density has the sum of variable and parameter.
The Jeffrey’s Prior for the standard deviation of the Normal distribution is uniform in its natural logarithm. This is because the density has the ratio of variable and parameter.
Poisson: uniform in square-root.
Bernouilli: uniform in sine-squared.
oneuniverse,
Except he seems to be suggesting that he thinks there are no problems with his own, which is not – IMO – a good sign.
@dikran
Put differently, assuming that the temperature is normally distributed, the Jeffrey’s Prior of the equilibrium climate sensitivity would be uniform if the expected temperature were proportional to the equilibrium climate sensitivity.
oneuniverse
James Annan has criticised Nic Lewis’ use of ‘objective’ priors. Are you aware of this?
Richard Tol, wrote “The Jeffrey’s Prior for the mean of the Normal distribution is uniform.”
That does not “show the relationship between Jeffrey’s Prior and an Uniform Prior”, that is just giving an example of a problem for which the Jeffrey’s prior happens to be uniform for the traditional parameterisation of the problem. Likewise the other examples are merely cases where the problem can be reparameterised to a form where the Jeffreys prior happens to be a uniform prior.
However that appear to justify your original claim ” “But interpreting P(T|CS) as P(CS|T) — the definition of non-information — requires that T and CS are measured in the same space.””
For the Jeffreys prior to be the uniform prior [such that P(T|CS) is P(CS|T)] does not require that T and CS are “measured in the same space” it requires that the square root of the Fisher information matrix is a constant and “measured in the same space” is not what that means.
@dikran
Recall that this discussion started with you claiming that a non-informative prior is uniform, and me countering that it is uniform after the appropriate transformation.
I wrote “So as long as we don’t treat it as being in some way a privileged “best” estimate of ECS, but as just one of a set of estimates, then there is no problem AFAICS.”
That would be the case if all the estimates are of equal value,
No, when you have more than one way of estimating something, you draw your conclusions by weighing each of the estimates according to their plausibility, i.e. you consider the caveats and advantages of each method. Using a Jeffreys prior does not mean that Lewis’ method is the privileged “best” estimate of ECS, but it does have its good points as well as its problems. Lewis’ analysis does not automatically outweigh estimates from GCMs (which have a better basis in physics) or paleoclimate estimates (which have the advantage of actually showing the equilibrium change). The best thing to do is to critically examine all of them and draw conclusions (which is what e.g. the IPCC did AFAICS).
Richard,
No, I actually think it started with you saying this which appears only to be true if the words you used have meanings in statistics that aren’t the same as what most would understand those words to mean in reality – an apt illustration of this whole discussion, ironically enough.
Richard Tol “@dikran Recall that this discussion started with you claiming that a non-informative prior is uniform, ”
No, you started by claiming
“By construction, the Jeffrey’s prior is non-informative in the sense that the prior has no effect on the posterior — or in the sense that the posterior is fully determined by the likelihood.”
T which I responded to point out that is incorrect, it is the uniform prior that “has no effect on the posterior”. However I did not claim that meant the uniform prior is uninformative, just that it had no effect on the posterior. In fact I clearly stated it wasn’t necessarily uninformative “Yes, in this case a uniform prior (even though doesn’t provide invariance and it is informative for a scale parameter)”. I also stated the true sense in which the Jeffrey’s prior is uninformative:
The Jeffrey’s prior is uninformative in the sense that it maximises the divergence between the prior and the posterior (i.e. it mazimises the information provided by the data and minimises the effect of the prior in a particular technical sense). Note Jeffreys’ himself didn’t consider the prior as uninformative, but minimally informative. I’d agree, as it encodes the prior knowledge that the analysis should be invariant to some re-parameterization (e.g. a change of units).”
which is not “has no effect on the posterior”. This is important as the idea that ““the implications of using a Jeffrey’s prior are exactly zero.” is not actually true (as pointed out by several of the papers cited by other contributors to the thread).
Pekka Pirilä (my respects) said:
The fundamental strength of the Bayesian approach is that it recognizes explicitly the limitations of statistics and statistics based inference.
The prior comes from outside of the method, i.e. from the subject science in case of science. No conclusions can be drawn without some knowledge or assumptions about the prior.
Seems this discussion began implicitly and haphazardly using the nonsensical prior employed by NL. I’m still hoping the discussion will eventually turn to its rightful starting point, i.e., the analysis with the strongest credible claim to certainty, i.e., Hansen’s paleo analysis.
James Annan has posted a sensitive comment on his blog. No photos.
So as to not disrespect the present discussion, (philosophy begins in wander), I should say the pursuit of ECS reality with or without Bayesian statistics should begin with a survey of the strongest ECS estimates and their relative merits/demerits.
…echoing a comment a few comments prior to mine and recalling the agreement of a recently passed resident expert.
Or, as Pekka suggested, break the problem down into more fundamental elements that are more tractable to statistical methods.
Re: Annan’s post…can anyone tell me whether or not Gelman “properly” understands Bayesian estimation?
Joshua,
Richard Tol thinks he’s a sound statistician.
oneuniverse
“For example, some have argued here that Nic Lewis’ estimate(s) use a prior peaking at a physically-implausible zero C; if that were true, it would bias the PDF downward and therefore, I think most would agree, devalue the estimate.”
1) It is not exactly true. For the main results of my studies, I have only used uniform priors. I used them for observed quantities for which the uncertainty was assumed to be Gaussian or t-distributed, not for ECS or other parameters being estimated. What I did show, in Lewis (2014), was that if one directly inferred ECS from assumed likelihood functions using Bayes theorem, the method I used was equivalent, in that case, to using a (Jeffreys’) prior for ECS that peaked at zero. In Lewis (2013), on the other hand, the equivalent prior for ECS did not peak at zero.
2) it is not true that using a Jeffreys’ prior peaking at zero biases the estimate downward and hence devalue the estimate. Perhaps you would be happier if the prior for ECS ceased to rise at, say, 0.25 K or 0.5 K, and then fell to zero at 0 K? But what is the point of doing so? It makes absolutely zero difference to the ECS central estimate or range.
Nic,
I’ve seen this now, but I still don’t know what it means to have probability averaged over the aerosol forcing and the significance of doing so.
Well, sure, I can see that if you don’t change the prior until you get to below 0.5C, that may make no difference, but what if you change the form of the prior so that it’s falling even when get to 1C? As has been pointed out a number of times, an interesting aspect of your work is that it does appear to largely rule out ECS values below 1C, so changing the prior only below 1C is unlikely to have any effect.
Would be interesting to get your views on James Annan and Andrew Gelman’s posts. I guess their key points (which I largely agree with) is that there is nothing particularly special about a Jeffreys prior.
Pingback: Dogma? | …and Then There's Physics
Bill
AR5 covers this. Here’s the money shot
> I don’t think all estimates deserve an equal seat at the table, just by virtue of being a member of the set of estimates.
This reinforces James’ impression that:
http://www.climatedialogue.org/climate-sensitivity-and-transient-climate-response/#comment-901
Applying XKCD’s interpretation of statistical significance on the ECS studies’ graph leads us to the Galileo paradox: Nic’s preferences are barely significant.
***
Since we’re into thought telling, I do think that Nic’s cuckoo tactics cannot be justified by some kind of moral imperative to fight for the freedom of Western society:
I’m guessing that from willard’s latest link, I can presume that John Fassullo as one representative member of the “vast majority” of climate scientists, doesn’t “properly” understand Bayesian estimation:
Which is rather interesting, since Fassullo was a student of one Peter J. Webster.
Willard
“This reinforces James’ impression that:
Nic Lewis appears to be arguing primarily on the basis that all work on climate sensitivity is wrong, except his own, and one other team who gets similar results.”
It seems that James can’t count – or maybe he used a highly informative prior when estimating how many other teams’ work I rated satisfactory! What I wrote in the guest blog that he was commenting on was.
“Which instrumental warming studies are satisfactory?
After setting aside all those instrumental-period-warming based studies where I find substantive faults, only three remain: Aldrin et al (2012), Lewis (2013) [solid line Box 12.1 Figure 1 range using improved diagnostic only] and Otto et al (2013). These all constrain ECS well, with best estimates of 1.5–2.0°C. Ring et al (2012), cited in AR5 but not shown in Box 12.1 Figure 1 as it provided no uncertainty ranges, also appears satisfactory.”
That is three other teams, not one. Moreover, the Otto et al (2013) team includes most of the key IPCC AR5 lead authors involved with chapters relevant to estimating climate sensitivity – fourteen of them in all. (I was also a co-author, but I was not in the driving seat, and it would be wrong to regard that study as my own work.)
> It seems that James can’t count […]
Neither can you, Nic:
http://www.bishop-hill.net/blog/2015/10/5/puffed-rice.html?currentPage=3
In your list, there was not three “other teams,” only two, if by “other teams” we mean teams that don’t include you. (The same kind of distanciation can be observed in your GWPF pamphlet: wanna see?) Either you can’t count or you count well enough to do some double accounting.
You still fail to tell how many other CS studies there were at the time of this count.
dikranmarsupial:
That’s in agreement with what I said, and that’s what Nic Lewis did too.
BBD:
Yes. As I wrote at the CA thread, my layman’s preference, FWIW, is for a best-effort expert prior, particularly when trying to come up with an improved estimate of CS (as opposed to NL’s aim of a single-study poolable result). A weak informative prior would perhaps go some way to reducing the effect of double-counting should the result be used as a prior distribution in subsequent studies. Perhaps the fact that an expert prior is easier for me to understand is biasing my choice. I also added (echoing Steve Mosher) that it’d be good to see the results of a variety of priors (like Annan and Hargreaves 2009).
Having said that, NL’s prior from Lewis 2013 is, I hazard, not to far from a middle-of-the road choice of expert prior. Pekka wrote in the above-linked thread at CA: ” It’s not totally accidental that my preferred subjective prior has some similarity with that [NL’s], but that’s as much as I’m ready to accept: Some similarity with reasonable subjective priors justified by physics based arguments.”. So the results may not change much.
ATTP:
The main question is, surely, to what extent are his criticisms correct ?
Nic, if you read this, would you mind outlining what you see as the limitations of your estimates of climate sensitivity, maybe particularly in reference to the use of ‘objective Bayesian’ techniques? Are there any that have occurred to you since you gave the Ringberg presentation?
If the difference between the ‘objective’ prior you’ve derived and a reasonable ‘subjective’ expert prior is not large, would it not make sense to accommodate the arguably less-controversial expert one?
So, if you take the temperature change since 1979 divided by the radiative forcing since 1979 and multiply by the IPCC 2x CO2 forcing amount ( 3.7W/m^2 ) you get the value ( 1.59 K / W/m^2 ) depicted on the chart:
So sorry – I missed out the slash in the closing blockquote tag of the first paragraph. Please correct if possible, mod? Thank you
vtg,
Thanks for the link.
I was just looking at this Annan post from a few years ago. The resulting discussion is the type of thing I’ve been looking for, i.e., summary AFTER analysis and weeding. I view IPCC reports as an occasionally useful, mostly cumbersome, politically slanted encyclopedia of climate change knowledge. I do much better studying JH papers and the papers referenced therein. Of course, I don’t do much of that… 😦
@-Turbulent Eddie
“So, if you take the temperature change since 1979 divided by the… you get the value ( 1.59 K / W/m^2 )”
Wow.
If you had used the RSS satellite temperature data you could have made it even smaller.
Determining the absolute lowest bound for sensitivity would at least prevent anyone foolishly thinking it could be very close to zero.
.
Nic,
Firstly, I think this is a slightly bizarre thing to have said in the first place. Going around finding fault in the work of others so as to promote your own is a little like a marketing exercise, not a part of sound scientific practice. Also, I don’t think you get to minimise your role in a paper on which you are an author. Unless, I’ve miscounted, only one of the studies in which you find no substantive faults is one on which you are not an author.
Okay, you’ve also counted Ring et al., so maybe it is 2, but that doesn’t really change my point. There are caveats with your work too and suggesting that these do not exist is poor.
TE,
your post, as you’re well aware, is misleading.
You understand the difference between TCR and ECS, and you deliberately show a (very crude) estimate of one on a plot of the other.
Thank you for demonstrating so well how to dishonestly misrepresent data.
Carry on.
Nic, pay attention. Look at the reactions from Annan and Fassullo, ATTP and numerous others. Your credibility is suffering because of the way you are over-promoting your results. A wise man would pause and reconsider at this point.
VTG, yes concepts of TCR and ECS are different, but they also beg the question, what do you think will change so much? Are you not interested in how much different the idealized ECS is from what we actually observe?
Bill,
Genuine question :
In what way is AR5 WG1 politically slanted?
TE,
plotting TCR on an ECS graph, knowing it to be misleading, is dishonest.
Refusing to acknowledge that is unbecoming.
Carry on.
TE,
VTG’s point is that you plotted a TCR estimate (from a very short time interval) on a graph of ECS. I realise that you are a master cherry-picker, but even you can do better than that.
Just out of interest, do you drive by looking in your rear-view mirror only?
Just out of interest, do you drive by looking in your rear-view mirror only?
Do you believe you have powers of prescience that others lack?
TE
Usual grubby behaviour followed by usual attempt to deny grubby behaviour when caught out. You are actively damaging yourself with every post like this.
No, I don’t believe that, but I don’t see how that answers my question.
Cmon TE, man up.
Admit the deceit, gain respect, move on
Yeah. That’s gonna happen.
vtg,
IPCC has a spotless track record of being on the wrong side of slr estimates (for example). It’s not random. How do they do that?
I think you could find causality in the will of governments that guide IPCC editorial policy.
They choose the wrong side of caution and its called being conservative.
I’m very much under the influence of people like Peter Wadhams, James Hansen, and…
Jeremy Jackson, who has a funny, sad and scathing comment near the beginning of this lecture about the probablity of the IPCC being conservative about “everything”. Somewhere else in the lecture he blames them for the 20 years we’ve wasted.
If you like information delivered with humor and an edge and you’ve never checked him out you owe yourself.
He’s a famous ocean ecologist, a great scientist. He sits on an advisory board for NOAA. His specialty is not sea level, and he shares the vulnerability of most great experts of overconfidence in areas outside their specialty, so, take this lecture FWIW, and maybe check out some of his Ocean Apocalyse lectures.
> Going around finding fault in the work of others so as to promote your own is a little like a marketing exercise, not a part of sound scientific practice.
Speaking of marketing exercise, note that Nic basically repeated here what we can read in his GWPF pamphlet:
Click to access A-Sensitive-Matter-Foreword-inc.pdf
Notice the distanciation: there are several studies, i.e. four (out of how many we may never know), two of which he is an author. Yet the fact that he’s the author of half of these “several” studies has not been mentioned the first time Nic talked to me (i.e. at Bishop’s). I had to go read the pamphlet. Could say that this GWPF pamphlet is the footnote where the “good news” was hidden?
Instead, Nic pussyfoots about James’ “one.”
Admit the deceit, gain respect, move on
Well, I did post text before the image indicating precisely what the line indicated and again on the graphic, so I don’t believe that’s an accurate assessment.
My aim, however, is to help you ( and the AT guy ) with your self deception.
You would not find the data involved ( global average surface temperature anomalies and GHG estimates ) particularly controversial, so I’m not sure why you’d like to deny the relationship between the two in regards to the given estimates. Even if you believed there were some process which would lead to an acceleration of temperature rise, I would think you’d be interested in how recent observations stood in relation.
TE,
What you did was the equivalent of adding a car’s speed in kmh to a plot showing other cars in mph.
It was misleading.
You didn’t do it in ignorance, you knew it was misleading.
It was dishonest.
Bill,
Thanks for the explanation. I’d err more on the side of IPCC being scientifically rather than politically conservative personally, but I’d hasten to add that I’m not actually qualified to judge it either way.
vtg,
I think there’s that too (scientific conservatism). I hear both from my mentors. Wadhams says governments want to stay in power, so they ask for studies that will support policies that will be popular. Hansen says scientists are reticent because extreme or provocative studies are not good for job longevity. There seems to be lots of ways the system resists boat rockers. One wonders if there isn’t some fundamental law of the cosmos or human nature that favors failure in our situation.
TE,
Well, you’re certainly not going to do that if you continue to promote denialist claptrap. Seriously, you can do better, but for some reason you appear to not want to.
Well, you’re certainly not going to do that if you continue to promote denialist claptrap. Seriously, you can do better, but for some reason you appear to not want to.
Is it me in denial, or are you denying accepted data?
TE,
No, I don’t think I am. You do understand the difference between some data, and the significance of that data? Things are rarely, if ever, as simple as you appear to want to think they are.
If we could return to Nic’s stuff, that’d be great.
From now on, dear TE, please refrain from peddling.
@-Willard
“If we could return to Nic’s stuff, that’d be great.
From now on, dear TE, please refrain from peddling.”
It helps to think of TE as one of the minor characters in a Shakespeare play, a ‘rude mechanical’ who echos the tragedy of the major characters, but as farce.
Nic, thank you for the explanation.
ATTP:
Finding flaws in the work of others is sound scientific practice.
Publishing work that avoids those same flaws is also sound scientific practice.
Nic Lewis discovered problems in existing studies before he published his own. Do you have any evidence that he did this in order to somehow promote his own as-yet unpublished (and probably unwritten) work? Do you have any evidence that any of his criticisms are motivated by anything other than a desire for good science ?
Source: http://wondermark.com/1k62/
That sea lion is really out of order, intruding upon a private conversation and trespassing on private property. Did you have a point, by the way?
I like this from the sea-lion: “You’ve made a statement in public for all to hear. Are you unable to defend the statements you make? Or simply unwilling to have a reasoned discussion?”
I may be wrong, but I’m fairly certain such an attitude is part of healthy public debate (if a little towards the aggressive end of the spectrum – it’s not one I would generally adopt).
I have no doubt that ATTP can defend his statement, however given that I see the situation very differently , I don’t feel I’m being unreasonable by asking about what evidence he bases his opinion on.
@-oneuniverse
“Do you have any evidence that any of his criticisms are motivated by anything other than a desire for good science ?”
Accurately discerning the desires and motivation of another person is extremely difficult.
It is easier to observe the external effect on others and the general discourse that results from the criticisms, work and responses.
@-“Finding flaws in the work of others is sound scientific practice.”
And then there are sea lions….
NL -” You really don’t properly understand Bayesian estimation. But you are not alone – nor do the vast majority of climate scientists.”
Here, oneuniverse:
Click to access A-Sensitive-Matter-Foreword-inc.pdf
Thank you for your concerns.
oneuniverse –
==> <i"Do you have any evidence that he did this in order to somehow promote his own as-yet unpublished (and probably unwritten) work? Do you have any evidence that any of his criticisms are motivated by anything other than a desire for good science ?"
We have considerable evidence of Nic adopting an entirely unnecessary tribalistic stance from which he approaches discussion of the science. All that much more interesting since he disparages the work of others on the basis of his assessment that they are “activists.”
Joshua:
I think I may have read all of Nic Lewis climate science-related work in the peer-reviewed literature, and I don’t remember any assertion of his that someone’s work is wrong because they’re an “activist”. I assume you’re referring to something else he’s written or said. Could you point me to what you’re referring to? My guess is that if he said someone’s wrong, he’d have provided a substantive reason, even if he did describe them as an activist.
Sometimes, allegories can be quite telling:
Lord Lawson’s think tank modulz buff doesn’t always quote leftists, but when he does, it’s for Sound Science ™.
oneuniverse,
There’s a subtlety here, though. It’s one thing to do a piece analysis in the best way you think you can, and another to present what you claim to be an assessment of all the different analyses and claiming that the only one with no substantive problems is your own. That sounds like marketing.
I know you know there are at least two other independent analyses (Aldrin ea and Ring ea) that he’s included with his, so I don’t know why you continue to say that he claims that his is the only one.
Anyway, his criticisms are technical and specific, and can be checked. As I wrote earlier, the main point of interest for me is whether his criticisms are correct.
.. and indeed, whether the criticisms of his work are correct.
oneunvierse,
Because I’m exaggerating a little to make the point (that you appear to still not really get).
Indeed, but the issue is more that he will publicly state that he has found substantial flaws with othe people’s work, but none with his own. It’s not so much that the criticism of the other work isn’t necessarily valid, it’s the somewhat unlikely suggestion that his own is perfect.
> the main point of interest for me is whether his criticisms are correct.
Ask for evidence, then say it’s not your main point of interest.
***
> his criticisms are technical and specific
There’s nothing very specific about Nic’s “objective” ringtone, and the technical side has been shown wrong. There’s nothing “technical and specific” about Nic’s “best observational” ringtone. So I don’t know why oneuniverse continues to claim that Nic’s criticisms are “specific and technical.”
oneuniverse –
==> “I think I may have read all of Nic Lewis climate science-related work in the peer-reviewed literature, and I don’t remember any assertion of his that someone’s work is wrong because they’re an “activist”.”
First, it wasn’t in his peer-reviewed work.
Next, I shouldn’t have brought it up since I don’t have the link at my fingertips, and I’m not 100% sure that my memory is correct, but I don’t think it’s worth spending the time it would take to use The Google to search for it
….but my recollection is that the comment when Nic disparaged someone’s science on the basis of his determination that they (I think it was Cowtan and/or Way) are “activists” did come with substantive criticism, but it was a useless “something extra” that polluted that substantive criticism, and was yet another in the long list of examples of smart,knowledgeable people employing an obvious logical fallacy from the stance of being a tribalist in the climate wars.
It was also particularly notable since Nic affiliates with an activist organization, many of the members of which who complain when their work is disparaged because they’re activists.
Sameosameo.
I will also note that you changed my wording. I said that Nic disparaged someone’s work on the basis of his determination that they are activists and you rephrased to say that he asserted that their work is wrong because they are an activist.
I don’t think that the difference in our wordings is trivial. You mischaracterized what I said into (probably) what you heard. I find that kind of change-over to be rather interesting and meaningful because it reflects the difficulty of carrying out good faith discussion even when the intent to have one is there. Additionally, it shines a spotlight on the rhetorical ploy that Nic used, where someone can present a connotation that someone else’s work is wrong because they’re an activist but maintain a plausible deniability by saying that he/she never said that the work is wrong because they’re activists.
.
Well, actually, it was much easier than I thought:
http://rankexploits.com/musings/2015/yes-some-things-are-obvious/#comment-134430
and this
http://www.climate-lab-book.ac.uk/comparing-cmip5-observations/#comment-436578
Which, of course, fits hand in glove with this kind of tribalism:
ATTP:
Now that does sound like marketing. I do get the point, by the way, I just don’t agree, clearly.
I think this is maybe another exaggeration? Not finding substantive shortcomings (to the point of making the results of the study unreliable) is not the same as suggesting that the study is perfect, not by a long way.
My point is, Nic Lewis has made a set of claims of a scientific nature. You can say it’s a marketing excercise if that’s how you see it, I can disagree. The validity of scientific claims are checked in the scientific domain, not by some analysis of motives and associations (GWPF, Greenpeace, whatever). It’s like people claiming that the IPCC is a politically-tainted organisation, or that climate scientists have been affected by noble cause corruption. What valid scientific conclusions can one draw about their work from that? I would say zero. So it’s not a very interesting avenue for me.
willard:
I’m discussing the evidence with ATTP, willard, and no, as I’ve explained, it’s not my main interest (I suspect it’s not ATTP’s either). Is that difficult for you to grasp ?
I did so because I think they are. ATTP has agreed with me, I don’t think it’s a controversial statement. You could start with Appendix A of the GWPF document “A Sensitive Matter” that you linked to, if you’re interested (did you actually read it?). For full details, you’ll probably want to contact Nic Lewis.
oneuniverse –
I agree with you there – except the uninteresting part, as I think it’s interesting to see how often smart, knowledgeable people employ fallacious reasoning.
But here’s my question, then. Given what you said about criticizing the IPCC for being politicall-tainted, what conclusions do you draw from Nic’s comments about the activism taint to the work of folks like Cowtan and Way? Why would he add that “something extra” if no scientific conclusions can be drawn from that information?
oneuniverse –
Also this…
But that’s not all he said. He didn’t make a set of scientific claims and leave it at that. The examples I gave of his “something extra” are not the complete set of the “marketing’ that he mixes with his claims of a scientific nature. If requested, I could use The Google to find more, but I’m quite sure there are better uses of my time, and perhaps there’s no point: If you will acknowledge that his participation is by no means limited to specific scientific claims, and includes a “marketing” component, then I don’t need to add more examples. If you won’t, then I suspect that more of what I consider examples wouldn’t convince you anyway.
oneuniverse
There is evidence that so-called ‘observationally-based’ estimates of S are likely to be biased low. This extends from EBMs (Knutti & Rugenstein 2015) to NL’s use of ‘objective’ priors. There’s clear evidence that NL has over-stated the importance of his results and at the same time accused the IPCC of ‘burying’ the ‘good news’. That is almost a conspiracy theory in my book. As a consequence, *my* confidence in NL is rock bottom. Why yours is so high is a mystery.
oneuniverse,
I’ll try one more time, but this is one of those rather tedious discussions that appears to go in circles. I’ve just sat through a very interesting seminar. The speaker was presenting a particular idea but also discussed other possible ideas and gave reasons why they may not be plausible. At the end of his talk, he also discussed various caveats with his own work. However, at no stage did he describe anything as substantially flawed and at no stage did he imply that there weren’t caveats with his own work. The reason that what he presented seemed plausible was because of how he described his own work, not how he described the work of others.
> I did so because I think they are. ATTP has agreed with me, I don’t think it’s a controversial statement.
I don’t think AT agreed with you about the many claims and things Nic said or did that did not amount to technical and specific comments. If all you want to say is that specific and technical comments are specific and technical, then I agree.
Nic made a set of claims that are neither specific nor technical, which we can read among other places in his pamphlet for the Global Warming Policy Foundation, at Judy’s, at our Beloved Bishop’s, and at the Auditor’s.
I don’t think it’s a controversial statement.
***
> My point is, Nic Lewis has made a set of claims of a scientific nature.
That’s different than to say that Nic’s “criticisms are technical and specific.” Nic’s main points are not that specific, and are only technical in the sense that the empiricism (“observational”) and Bayesianism (“objective”) are philosophical minutiae that go beyond scientific matters.
***
> I’m discussing the evidence with ATTP,
Actually, that’s not even true. You’re simply echoing the usual ringtone auditors hammer to deflect from their marketing fluff.
Hi Joshua, thanks for the links. I agree, it’s irrelevant to the validity of their scientific work whether Cowtan, Way, Cawley ea are activists, or part of an activist group. Similarly, it’s irrelevant to the validity of his scientific work whether Nic Lewis’ work has been ‘marketed’. Same for the IPCC.
(In Nic’s defense, in his nearly 1000 word post at Lucia’s, about 20 words mentioned SkS activism, and almost all the rest discussed the technical merits and problems of Cawley’s paper, but the point is well taken).
From Nic’s SkS activism comments, I draw no conclusions about validity of Cowtan and Way’s work, and indeed no conclusions about the validity of his own work. I also suspect Nic doesn’t have a high opinion of SkS.
ATTP:
Maybe because he didn’t see them as substantially flawed? You’re surely not saying that no studies should be described as such. For example, the tropospheric temperature trends paper of Douglass ea 2007 was widely criticised as being substantially flawed, to the point of being inadmissable as evidence – nothing wrong with that.
As I’ve pointed out, it’s incorrect (or an exaggeration) to say that NL doesn’t discuss caveats in his own work – I’ve quoted some above, and you’ll find more in his papers. Perhaps he doesn’t discuss them enough to your satisfaction, and that’s fair enough – I think you probably have a point there.
Anyway, as you say, I think we’ve run this into the ground. Thank you for the discussion and hospitality. Just to move it back to the science, Lewis and Crok wrote this in their GWPF article, which seems worth checking (re: conservation of energy) :
Sorry – I have another comment in moderation.
oneuniverse,
A report written for the GWPF? Also, the title of the section from which you quote is
which is almost certainly not true if the intention is to actually estimate the ECS.
oneuniverse,
Not one that I can find.
oneuniverse –
I acknowledge your acknowledgement.
You said earlier you didn’t find such stuff interesting. While I agree that someone’s “marketing” doesn’t speak to the validity of their science, I do find it interesting because it speaks very directly to the context in which discussion about climate change is taking place.
And it does speak to Nic’s reasoning more generally, when he so easily and apparently without insight engages in obviously fallacious reasoning. Although it is far from speaking directly to the validity of the science, it does suggest to me that he is not particularly good at controlling for his own motivated reasoning. That, IMO, is information that, as someone who can’t assess the technical arguments myself, can use to help evaluate probabilities.
Certainly, it isn’t only Nic that lacks that skill. But I think it’s important that Nic is held in high esteem by so many “skeptics” when they routinely judge the work of “mainstream” scientists negatively because of what they determine to be a measure of “marketing.” It’s also instructive that Nic, himself, as an activist, thinks that it is valid to disparage the scientific value of someone’s work on the basis of them being, in his eyes, and “activist.”
All part of the beautiful unintended irony so commonly found in the “climate-o-sphere.”
oneuniverse –
I’ll pose this question again, in hopes that you’ll answer it:
==> “Why would he add that “something extra” if no scientific conclusions can be drawn from that information?
Goodness me, oneuniverse, you ignored my comment.
==> ” oneuniverse, you ignored my comment.
I’ll step off now, and as a noble and self-sacrificing gesture, allow the more technical discussion to unfold without my distractions….
…plus, I got stuff to do. 🙂
In his email to Andy Gelman, Nic writes:
http://andrewgelman.com/2015/12/10/28302
Are these claims strictly scientific?
> [I]n his nearly 1000 word post at Lucia’s, about 20 words mentioned […]
And then there’s minimization.
BBD, I just saw your comment.
ATTP, thank you, the comment appeared after a short delay, sorry.
I’m just heading out, will be able to reply in a few hours.
oneuniverse writes
Nic’s claims cannot be checked in the scientific domain unless he releases his code, explains what approach he took to the literature in crafting his assumptions, and demonstrates that his results are robust enough to survive the use of a plausible prior. Until then, we have hands shuffling behind a velvet curtain and then, voila! Which leaves us trying to guess what happened behind the curtain, based on social and psychological indicators, such as the company he keeps. He appears sympathetic to, and is regard well by, people whose work clearly fails in the scientific domain. This guilt by association currently carries more weight than it needs to because of the lack of transparency in his own work, but he could ditch the velvet curtain whenever he chose to, and we could then assess the work on its own merits.
Joshua:
I’m guessing the fact that they’re SkS contributors was a warning to Nic Lewis that their work may suffer from the motivated reasoning and shoddiness often found at SkS (IMO). However, not all SkS contributors are of equal skill. I think it’s a caveat to him that the work needs to be carefully evaluated before its accepted. That’s not fallacious reasoning, by the way.
Unless he didn’t engage in fallacious reasoning. My reasoned guess above is supported by what he wrote a few comments later in that Ed Hawkins post :
Leto, the code and data for Nic Lewis’ published climate science papers can be found at his WordPress site.
BBD, I don’t really have much to add. Yes, you’re confidence in NL is low, I gathered. I’m content to see how the debate plays out in the peer-reviewed literature.
Last comment, sorry :
ATTP:
I guess one can just wait for it to be discussed in the PRL. I think the instrumental data provides the best estimates of effective climate sensitivity. EqCS is more debatable, my guess is that as time passes, the instrumental data will also provide the best estimate of of EqCS.
oneunivers –
==> “That’s not fallacious reasoning, by the way.”
I disagree. First, IMO, judging scientific work on the identity of who produced it, rather than on it’s own merits, is a classic fallacy – particularly if you have the brains and scientific chops to evaluate it at a technical level. Guilt-by-association is also fallacious.
I can’t evaluate Nic’s work on it’s own merits, and I try hard not to judge his work on the fact that he associates with people who produce poorly reasoned analysis (such as we can commonly find from Richard Tol or stuff at WUWT). But I do use the lack of control for biases that he displays to give me information to assess the probabilities of whether his scientific analysis is biased by his tribal identification. His displayed poor control for his biases is in no way a conclusive means for me to evaluate his science, but it’s relevant information.
But even further, his comments are “something extra.” They serve absolutely no analytical purpose. Even beyond that he succumbs to a fallacious method for judging work, he could do that and be silent about it. That he announces his reliance on fallacious reasoning shows that he goes further, to put forth an identity-oriented marker – to stake out his affinity with an “us” and against a “them.”
There is no other purpose to those statements. They serve absolutely no analytical function. If he wants to present his scientific analysis, and offer a critique of others, all he has to do is stay on topic. That he strays off topic, and even more, introduces his analytical critique with an irrelevant preface (see willard’s comment about miniization w/r/t your word count comparison which is a falsely simplistic metric to judge the relative importance of his ad hom introduction compared to the rest of his critique), shows quite clearly that his focus is not merely on the science itself. Again, those comments are completely irrelevant to the topic at hand. Why, then, does he make them?
==> ” However, not all SkS contributors are of equal skill”
Not all WUWT contributors are of equal skill. Should I scrutinize Nic’s work differently – because he contributes to the same website as someone like Tim Ball – than I would if he only presented his work in peer-reviewed journals, for example?
==> “Unless he didn’t engage in fallacious reasoning.”
Well, I guess that you and I just disagree about that. For example,
If that logic works for you, then I’m not sure there’s much more for us to discuss….but why let that stop me?. I see no way that in that context, differing levels of scrutinization isn’t a reflection if (1) disparaging someone’s work on the basis of who they are and not the merits of the work itself and (2) a “motivation” in the sense of “motivated reasoning.”
Should I have a lower bar for technical validity (as would be be product of a lower level of scrutiny) for something that people I generally agree with write as compared to people I don’t generally agree with? Of course, we all do that to one degree or another. IMO, that people do it is a trivial observation. What’s important is whether people try to correct for that tendency. Nic, apparently, dismisses the negative implications of engaging in that fallacious form of reasoning as he proudly asserts that differing levels of scrutiny should be based on who wrote something (or even worse, who they associate with) and not the merits of the work itself.
Perhaps not for you, but for me, that’s information that increases the probability that he doesn’t control for his own biases in his scientific work.
Oneuniverse… Thanks for the link to Nic’s code. I’ll take back what I said about the transparency – I was commenting on the basis of a lack of a response to earlier comments asking for the code.
It remains to be seen how clear the code is, and I know it is very difficult to trawl through someone else’s code unless it has specifically been written to make it easy – but until someone with the necessary knowledge has had a go at reproducing the work and exploring its robustness, I agree it is premature to dismiss it on the basis of the company he keeps.
oneuniverse – as I wrote in the comments to the “Sensitivity to cumulative emissions” post:
A robust model would show little sensitivity to the choices Nic made. A scientist interested in learning and sharing the knowledge gained would include results using alternate choices to show that the model is (or is not) robust to these (oftentimes arbitrary) decisions.
Nic’s ‘weasel words’ divert attention from the fact these choices may in fact be all that’s needed to explain the results. Or, if one has a more suspicious nature, that these choices were made specifically to bias the results.
This discussion is entirely moot because atmospheric warming occurs via heat released from the ocean in episodic events. Observations to date cannot be relied upon to back out climate sensitivity as a linear variable, and because the evolution of climate sensitivity is nonlinear over time (though may be monotonic under certain boundary-limited conditions). It is not a single number but likely to be a vector – perhaps nonlinear if ice sheet collapse, methane outgassing has a rapid enough effect. Palaeo backed by model evidence is the best we have at the moment.
All the stuff that points to low ECS based on obs can be discounted because it is based on an emergent and nonlinear signal.
Climate change should be viewed as a hydrodynamic process, not a radiatively-additive process (though it contributes, it is not the whole story).
Someone will challenge me for references. It’s all across the literature and not in one place, though we do have some just submitted and other stuff close to being put in. Other people have similar views but it isn’t mainstream.
Roger Jones: It’s all across the literature and not in one place
All in one place
Climate models, based on physical laws that describe the structure and dynamics of the atmosphere and ocean, as well as processes on land, have been developed to simulate climate. Models help us understand climate sensitivity because we can change processes in the model one by one and study their interactions. But if models were our only tool, climate sensitivity would always have large uncertainty. Models are imperfect, and we will never be sure that they include all important processes. Fortunately, Earth’s history provides a remarkably rich record of how our planet responded to climate forcings in the past. Paleoclimate records yield, by far, our most accurate assessment of climate sensitivity and climate feedbacks.
EqCS is defined as a single number and within certain bounds is arguably a single number in reality. If you’re not defining it that way then your objection is a semantic one.
Roger btw,
:Looking through your posts, You also wrote
you don’t have the process right. The radiative forcing happens in the atmosphere and that’s where the heat starts, but it goes straight into the ocean. The warming of the atmosphere does not happen in situ, but comes from a warming ocean. That’s why the energetics of the hydrological cycle should be seen as a thermodynamic phenomenon, not a radiative one.
To add to MT’s point, the warming is greater over the land, but it is generated from ocean-atmosphere interactions and not an atmosphere warming in place.
I’ve been wondering about this. What is the logic or the proof that the atmosphere is not heated directly by the greenhouse feedback radiation?
bill,
it does happen in the atmosphere but the heat does not stay there. Most is absorbed by the ocean and re-released. Either that, or the atmosphere warms in situ, which it shows no sigh of doing (and this is where I depart from the mainstream in the interpretation of what happens using statistical tools). The climate system is being viewed as a classical system a la Newtonian physics instead of a complex system, a la Lorenzian attractors. It approximates to a Newtonian system, but not on the time scales we need to distinguish to properly characterise climate risk for decision making.
And your comment on semantics. It has to be worked out where in a reasoning process that semantics may reside. If the EqCS is defined as one number, what one can say about it for the evolution of climate risk is will be limited if that risk is defined by a sufficiently different process. If it’s about an imaginary number and what it means for policy – that is semantics.
And the Hansen and Sato paper is good, but there’s more to it as well. Ghil (recent papers), Ozawa et al 2003
Except that noone who associates positively with the GWPF is in a position to criticise another organisation for motivated reasoning and shoddiness.
Roger, seems like the distinction you make is like the division between kinetics and thermodynamics in chemistry. Neither one completely describes the phenomenon of a reaction. You only have to look at Hansen’s new paper on sea level and storms to see non-linearity taken seriously. The potential abruptness of ice sheet collapse and the attendant rise in sea level that result from a major feedback in the southern ocean. The curvy, non-monotonic path that global temperature is likely to take in the next century or two as a result of a shutdown in the AMOC. That news is older than Storms of my Grandchildren.
Roger,: it does happen in the atmosphere but the heat does not stay there. Most is absorbed by the ocean and re-released. Either that, or the atmosphere warms in situ, which it shows no sigh of doing (and this is where I depart from the mainstream in the interpretation of what happens using statistical tools).
I actually agree with you here, without knowing the physics. If the atmosphere was good at absorbing the same wavelength (long wavelength?) as is involved in greenhouse warming, then there wouldn’t be a special case for the trace gases. Seems like this is probably well known.
oneuniverse wrote “I think it’s a caveat to him that the work needs to be carefully evaluated before its accepted. That’s not fallacious reasoning, by the way.”
It is fallacious, all work needs to be carefully evaluated before it is accepted, regardless of its source. Otherwise the obvious corollary is that there are sources that you feel don’t need to be carefully evaluated before the work is accepted, and if they are sources that agree with your position, that is a recipe for confirmation bias.
Getting something published in a journal doesn’t mean it is immediately trustworthy. It means it has made it through the basic sanity check that peer-review can be expected to provide most of the time. Acceptance of a scientific idea comes much later, when it has been “carefully evaluated” by the scientific community, and the evidence of that acceptance is generally provided by things like citations in work where the ideas it contains are used. Thus all papers need careful evaluation, especially the ones you agree with, that is what real scientific skepticism is about.
Note I have ignored your insulting behaviour and responded to you in a civil manner.
OTOH, non-linearity is difficult for modellers, and as a result they have done poorly with sea ice and land ice, which has resulted in major failures for the mainstream IPCC; consternation and frustration for Wadhams and Hansen.
Roger Jones
+1
Roger, regarding semantics: ECS is useful as defined. The modern record of atm. global mean temperature is probably a bad way to derive it. That is not the fault of the definition, it is the fault of the people doing it that way.
If you want to study the likely pathway by which temps will evolve, then call that something different.
bill
Not sure – and Roger Jones will hopefully correct me if I am barking up the wrong tree – but what he is saying about atmospheric warming being driven by heat release from the oceans reminded me of Compo & Sardeshmukh (2009), which you might find interesting.
Bill,
#69028. The nonlinearity in temps we have been analysing is in the models (and in the observations) – the output is just not being analysed appropriately. The behaviour we are seeing is on decadal timescales and is independent of the larger, discontinuities associated with ice sheets, etc. Nonlinearity is multi-scalar for a number of reasons.
And I obviously didn’t explain the semantics thing well enough. ECS is useful, but if people are going to use observations to back it out of the Delta F – Delta N)/Delta T relationship and imply that the result has some implication for policy (and should be preferred over parameter-based uncertainty for instance) – then they are playing semantics. None of the measures for sensitivity are equivalent and it is important to use them for what they can adequately represent. You are implying a new measurement is needed but that would not solve this issue – best to live with it and be transparent and true to the underlying assumptions. ECS is related to the magnitude of warming in the 21st century, but not the 20th century, if the climate models are any guide. Given we have only one Earth to sample observations from, they are the best guide we have.
Many researchers in the next little while are going to wake up that transient sensitivity measurements are actually not very useful in decision making. They are hung up on making probabilistic predictions of mean change, but those predictions are not really relevant when you look at the types of decisions that need to be made.
Scientists often think that what is useful for them will be useful for the user. Sometimes this is true and sometimes it is not (have fallen into this trap often enough to be very wary).
BBD,
they are on the right track. Another good reference for us to use – much appreciated.
dikranmarsupial:
Ideally, yes. (In practice, not all work is carefully evaluated before it gets incorporated into the canon since resources are limited and humans are fallible, but science, due its self-correcting nature, tends to works anyway). I should’ve added the word “particularly” before “carefully” to correctly express my meaning : if the work is coming from a less trusted source, one will naturally want to take care to verify the quality. I agree with you about also wanting to be extra careful in checking the papers that one agrees with.
dikranmarsupial:
I assume you mean the criticism of SkS. That’s the nature of criticism I guess, some may feel hurt or insulted – that wasn’t the aim of it, though. I did actually refrain from writing that comment until Joshua asked a second time for my answer, as I know there are some SkS contributors who read this blog, and I thought that the answer to his question was fairly obvious. FWIW Kevin Cowtan’s work seems to be excellent.
ATTP:
Actually, they are. How much attention gets paid to them, and how seriously they’re taken, is a different matter. Similarly for someone who associates positively with SkS, Watts Up With That, Greenpeace, The Young Conservatives. People will form their opinions. In the end, as dikran wrote, “all work needs to be carefully evaluated before it is accepted, regardless of its source”. Ideally, I would add.
I haven’t read any of the GWPF’s publications apart from the “Sensitive Matter” document, but might now read Indur Goklany’s “Carbon Dioxide: The Good News”, compare it against the published literature and see how it fares. I’ll then have a better idea about the quality of the GWPF’s work.
Roger, I came across that excerpt from Hansen today and pounced on the chance to broadcast it. In the process I became unnecessarily disagreeable. On a topic on which you and I fundamentally agree. So I will take BBD’s point and back out.
Long live paleo! Long live our Holocene home!
Regards.
Well, yes, obviously they can choose to do so. I meant that doing so is ironic given that most of what they promote is shoddy and heavily influence by motivated reasoning. Glass houses and stones, springs to mind.
BBD, thanks for the ref — I will peruse and savor at my leisure. 🙂
Yes, check out that legendary GWPF ‘peer review’… 😉
When people say the oceans release heat it always bothers me. Obviously oceans realize a huge amount of heat as a huge amount of SW goes into them. LW either does not go into the oceans, or the mechanisms by which it does are poorly explained. The oceans are very very different beast. There is no obvious equivalent to land heat content piling up on the downwind side of continents, nor is there an equivalent to wind-driven upwelling on one side of a continent. At one month not long ago NOAA listed their land component of Land and Ocean at 44th warmest, and it rapidly advanced to warmest… when the anomalous winds stopped blowing. And yet, on net, OHC almost always goes up. So to me this is all about the Eastern Pacific, which cooled 1985 to 2013, and I suspect it has now entered a warming trend.
JCH,
the mechanisms by which it does are poorly explained.
That’s it. It’s pretty simple, really. Most of the heat comes through the tropics. It gets absorbed into the ocean via winds (simple heat transfer aided by winds). The air over land cannot heat independently of the ocean because of the massive mixing in weather systems. There is an already existing amplification effect from ocean to land – all warming does is to accelerate this.
Nor does it make sense to look at land-ocean differences in terms of energy balance. There is more going on. I’ve said too much … but the way the climate system is being conceptualised at the moment is missing key aspects.
Climate change and variability cannot proceed independently. Think about it.
oneuniverse wrote “I assume you mean the criticism of SkS. That’s the nature of criticism I guess, some may feel hurt or insulted – that wasn’t the aim of it, though.”
No, criticism means pointing out specific errors and providing evidence that is at sufficient to allow the criticism to be addressed (at least in principle). You did neither. It is this sort of partisan behaviour that prevents communication of the science, and if you want to know why so few are able to remain civil while discussing climate, your own behaviour is an example of what tends to cauase the breakdown.
Note that while I am an SkS contributor, and the author of the paper joshua m (and the author solely responsible for the error), you will note that I demonstrated my “motivated thinking” on this thread by being explicitly complementary of Nic Lewis’s work, while pointing out the required (as I see them) caveats.
As to SkS being sloppy, if you stick within the (quite reasonable) comments policy there, then I’m sure your suggestions of corrections to any factual errors there would be most welcome. They certainly are on any of the articles I wrote for that site.
oneuniverse wrote “Actually, they are. How much attention gets paid to them, and how seriously they’re taken, is a different matter. Similarly for someone who associates positively with SkS, Watts Up With That, Greenpeace, The Young Conservatives. People will form their opinions. In the end, as dikran wrote, “all work needs to be carefully evaluated before it is accepted, regardless of its source”. Ideally, I would add.”
Actually that is not really what I meant. I meant carefully evaluated on its own merits, the identity of the source is irrelevant, what matters in the consistency of the scientific argument and the support from the evidence.
FWIW, about six months ago a similar exercise was set off by Cliff Asness and Aaron Brown. Now Eli, were he so inclined, could provide a list of several interesting responses, but perhaps a link to one of them (which links to the others) would be a good place to start
View at Medium.com
Mark Buchanan summed the situation up. Brad deLong also had something
————————-
Anyway, for clarification, let’s use an analogy. Imagine there’s a black box with a red light fixed to it. The light flashes every second or so on average, but in a highly irregular and unpredictable way. Some people argue that the flashing is getting more frequent with time, and showing larger fluctuations from its average behavior. Others say, no way, that’s an illusion, it’s always been irregular and these apparent changes are only insignificant and temporary fluctuations.
Two sets of people set to work to figure out if the pattern of flashing really is changing, and to predict how much we should expect it to change in the near future, if at all. The two teams go about their work in very different ways. Team A decides to work just with the mathematical pattern of flashing recorded over a not-too-distant interval of the past — say, one week. The other team, Team B, also uses that information, but decides to supplement it with other recordings of the flashing pattern from further in the past, some going back months, even years. Team B goes further too, using X-ray, MRI and ultrasound imaging of the box to work out a detailed, but certainly incomplete, picture of what goes on inside the box — gears and electronics and other stuff — to produce the flashing. They do experiments outside of the box to tease apart these mechanisms, and to get insight into how different mechanisms might interact within the box.
As the system turns out to be highly complex, Team B also starts to build replicas of the box, as well as large-scale computer models designed to simulate the interplay of all the mechanisms inside of the box. They test and refine both the replicas and the simulations over time using real data from the box. The members of Team B, knowing how easy it is for people to confuse themselves, and to believe they understand more than they really do, also splits itself into a number of sub-teams which compete against each other on standard data sets so they can get objective measures of improvement of these simulations over time. Who can run a simulation, based on plausible mechanisms, which can reproduce what the box did between 10 and 12 week ago? How does a model, trained on that interval, do if applied to other intervals later on? In this way, Team B slowly builds up a capacity for understanding what goes on in the box, and for predicting how it will likely behave next.
Now, suppose Team A and Team B make predictions for what they think is most likely to happen to the flashing pattern in the near future — say over the next 5 weeks. Both would acknowledge that the task is difficult given the complexity of the system. But which team do you think is more likely to make the better prediction? I think most people would naturally choose Team B, as they’re using a much richer set of information and data about the box and it’s behavior than Team A. They’re taking into account lots of things that Team A is not. Usually, the more information one brings to bear on a problem, the better one does on that problem. Indeed, most of the theories developed by Team A based on the short time series alone can be immediately shown to be highly unlikely by comparison with other data studied by Team B.
———————–
So now Nic can discuss the black box with the light on top problem.
On the December 10, 2015 at 3:10 pm, oneuniverse:
On the December 10, 2015 at 11:04 pm, oneuniverse:
The first comment was made before responding to Joshua’s December 10, 2015 at 2:57 pm comment. The second comment was made after having responded to Joshua’s December 10, 2015 at 2:57 pm comment.
Trying to immunize Nic’s performances with a true Scotsman after having strawmanned AT is a rare thing of beauty.
dikranmarsupial:
What, because I stated my opinion (you’re right, it fell short of a proper criticism) that the work at SkS often suffers from motivated reasoning and shoddiness?
On this same thread :
– ATTP has stated NL lacks understanding of basic physics and the scientific method
– BBD has stated that NL is collaborating to push misinformation – an accusation of dishonesty.
– BBD has stated that in “certain quarters, the objective is not to understand physical reality, it is to obtain a specific type of result” – in reference to Richard Tol or maybe the GWPF, again an accusation of dishonesty.
– Willard has suggested that NL is like a cuckoo, as I understood it, an impostor-scientist trying to eliminate the geniune scientists/papers to make room for himself, or something like that – another accusation of dishonesty (or maybe just a cry for help from Willard).
– ATTP has suggested that NL’s scientific critiques are primarily motivated by a desire for self-promotion rather than scientific inquiry.
– Kevin O’Neil has accused NL of using weasel words to divert attention from important matters, and, possibly, that he’s purposefully made choices to bias his result, which would be another accusation of dishonesy.
Multiple accusations of dishonesty aren’t normally considered part of civil debate. If you’re defense is that they’re merely speaking what they see as the truth, well so did I. Apart from Kevin O’Neil’s, the above comments were made before my SkS one, so I don’t think I caused the breakdown in civility. Of course, if you’re referring to something else I said, maybe you could point out what it was that caused the “breakdown”. (Please note, this is not a “mummy they did it first” defence – I stand by what I said, and I would have said it even if no-one had made the comments above – it was a response to a specific question by Joshua.)
If you’re concerned about civility in the climate debate, note that your organisation SkS hasn’t helped in the past, for example by giving out out a “Climate B.S.” award to Spencer and Braswell for publishing “On the Misdiagnosis of Surface Temperature Feedbacks from Variations in Earth’s Radiant Energy Balance” in Remote Sensing, or having sections like “Lindzen’s Illusions” and “Christy Crocks”. (I see the latter two have gone, or rather, the banners have gone – the Illusions stuff is still there. I haven’t looked in at SkS for a while – has it improved a bit?).
If it helps you feel better, my low opinion of SkS was not based on your work, and as I said earlier, I found your comments here generally very relevant and helpful, so thank you.
Hmmm, this wasn’t what I was going for. My point was that there are many people here with expertise. If someone is going to dismiss others because they regard themselve as somehow more expert in something than others, then others can simply reciprocate in kind. My point was that appeals to authority are weak.
oneuniverse
What else do you call selective blanking of inconvenient evidence from your argument? Would you prefer denial? We could agree on that and I would retract the implication of dishonesty at once.
ATTP, okay, but saying “You don’t understand X” isn’t an appeal to authority, which would be an argument along the lines of “I’m right because I understand X (and you don’t)” – an argument that Nic didn’t make. The first is a statement (which may be true or false), the second is a logical fallacy.
> or something like that
From the cuckoo’s mouth:
http://andrewgelman.com/2015/12/10/28302
Notice how Nic presents himself to Andrew Gelman: as a climate scientist.
Notice the parenthesis.
***
Perhaps oneuniverse would prefer James’ formulation:
http://www.climatedialogue.org/climate-sensitivity-and-transient-climate-response/#comment-901
Notice Nic’s response above:
(Seems that Nic can’t count either, or that if he does he’s double accounting.)
Compare and contrast with how he dismissed Robert Way’s work.
In fact, apply what oneuniverse himself said regarding such behavior before his sealioning turned against him.
BBD, what I wrote to dikran isn’t predicated on whether the accusations of dishonesty are correct or not. I was disputing the fact that I’d caused a breakdown in civility by stating my opinion of SkS, given the preceding discussion about Nic Lewis and the GWPF.
Except “you don’t understand X” can be followed by “you don’t understand Y”. That was my point. People appealing to their expertise through stating that others don’t understand something, is effectively an appeal to authority. If Nic feels comfortable stating that I don’t understand Bayesian analaysis, then a suitable response is that he doesn’t understand physics. I’d rather we did neither.
Except why are you in a position to do this? Are you Nic Lewis or associated with the GWPF?
> but saying “You don’t understand X” isn’t an appeal to authority […]
Sentences are not fallacies, arguments are. That sentence seems to imply some mind probing, which hints at an ad hominem. This specific judgement presupposes that its utterer can authoritatively evaluate AT’s understanding. This is where the appeal to authority lies.
Hope this helps.
***
Furthermore, notice the ad hom pattern:
– AT doesn’t understand Bayesian stuff
– James can’t count
– Robert is an activist
Evaluate these claims using oneuniverse’s concern about tone and civility. In fact, compare and contrast with Nic’s email to Andrew:
http://andrewgelman.com/2015/12/10/28302
oneuniverse –
==> “ATTP, okay, but saying “You don’t understand X” isn’t an appeal to authority, which would be an argument along the lines of “I’m right because I understand X (and you don’t)” – an argument that Nic didn’t make. The first is a statement (which may be true or false), the second is a logical fallacy.”
I gotta disagree with you there. Consider two options:
(1) You don’t understand X, and I can tell that you don’t understand X because you said Y.
(2) I disagree with you about Y.
The first is certainly an implied appeal to self-authority. The second gets to the main point, a disagreement about something, without any such implication. The first adds “something extra” that is completely extraneous to the main point, and in fact which more than likely confuses fact with opinion.
Let’s apply my thinking to this context: Suppose I were to say: “You don’t understand what an appeal to self-authority is.”
Is that not an appeal to my own authority to determine that you don’t understand the subject at hand, merely on the basis of you disagreeing with me?
Perhaps I could extend it: “I don’t discount what you say, but I scrutinize what you say because you hang out with people that don’t understand what comprises an appeal to self-authority.”
I’m not quite following, sorry : in a position to do what? Dispute that I caused a breakdown in civility? State my opinion about SkS?
No to both. I can’t tell what this has to do with your previous question.
ATTP, Joshua, unless Nic is claiming that he’s right on the basis of his (or someone else’s) say-so, I don’t think it’s an appeal to authority. It sounds like we have different definitions. Anyway, time for me to get some sleep, so I’ll say goodnight.
oneuniverse – did you notice the quotes around “weasel words”? Did you notice I referred to the previous post? Did you bother to check the context? I’ll take that into consideration as I throw your comment away – as making little sense. I’ll also note you never addressed the substance of the comment. To wit: the sensitivity of a model to arbitrary data decisions should be tested and revealed by the author. I’ve read hundreds of science papers that do just that. Nic Lewis did not, even though some would argue that all of his arbitrary selections fell to the low side.
You actually seem to take the GWPF seriously. How can you? Do you really need any more than this to discount them as anything but a policy/advocacy organization?
Is the GWPF avin’ a larf?
oneuniverse: You seem to think GWPF needs to be taken seriously… What do you think about William Happer recommending money being sent to him be laundered through Donor’s Trust with the intent of hiding who paid him? What do you think about is direct comparison of GWPF style Pal review versus proper peer review?
Here’s the email chain discussing exactly how he did this in the past, and how he wants to continue doing it now;
https://www.documentcloud.org/documents/2642410-Email-Chain-Happer-O-Keefe-and-Donors-Trust.html
I’ll listen to the GWPF the same day I look forward to a call from a telemarketer. There’s no difference between paid advertising and paid advertising near as I can tell. Can you elucidate as to what the difference is between the GWPF and a telemarketer? Both often try to hide their intentions and why they are calling… So its very confusing to me.
oneuniverse “If it helps you feel better, my low opinion of SkS was not based on your work”
Rubbish, you agreed that my being a contributor to SkS justified caution regarding the paper I wrote (which did happen to have an error in it, but that is beside the point), so your criticism was directed at me as well. It is a shame that you don’t have the grace to admit that you were wrong (as I did).
oneuniverse wrote “If you’re concerned about civility in the climate debate, note that your organisation SkS hasn’t helped in the past, for example by giving out out a “Climate B.S.” award to Spencer and Braswell for publishing “On the Misdiagnosis of Surface Temperature Feedbacks from Variations in Earth’s Radiant Energy Balance” in Remote Sensing, or having sections like “Lindzen’s Illusions” and “Christy Crocks”. (I see the latter two have gone, or rather, the banners have gone – the Illusions stuff is still there. I haven’t looked in at SkS for a while – has it improved a bit?).”
So three examples, two of which have been taken down, doesn’t support your case too well. Now I am not claiming that SkS is perfect (what is?), but one of the reasons I am a contributor there is that it sticks to the science much more than most climate blogs, many of which (and I am not going to name names) have insulting titles on a very regular basis and are perfectly happy to let insulting content-free comments stand. In my view makes your criticism unfair, as well as being an ad-hominem.
Joshua,
I’m not quite following your point. I had two main points. One was simply that dismissing what someone has said because they don’t understand X, is an appeal to authority (implicitly, maybe). The other was that if you choose to do so when in a discussion with others who have relevant (but different) expertise, then it can degenerate into counterclaims, which seems a poor way to have a discussion.
oneuniverse,
I was asking because you seemed to think that you could criticise SkS because others here had criticised Nic and the GWPF. I was wondering why you felt justified in doing so if you’re neither Nic Lewis, nor associated with the GWPF.
dikranmarsupial:
Yes, I don’t see any problem with an attitude like that, I think it’s normal, and any extra scrutiny your paper gains as a result will only help to improve it . I feel the same way about anything from the GWPF. My recommendation to anyone who wants to rely on any of either organasations’ work would be to check it thoroughly before proceeding.
All three happened and still exist on the website, so they do support my case. Just because SkS subsequently took down the front page banners of the two of them, after they’d been up for over a year, doesn’t somehow erase that. The articles for both the Crocks and Illusions were still there as of this morning (including the wording) so they haven’t been actually been taken down, just the advertisement for them on the front page. Note, I’m not saying that SkS should or should not be civil, I’m saying that kind of thing is a negative contribution to a civil debate, which is a concern that you raised above, not me.
ATTP:
I didn’t criticise SkS because others had criticised Nic and the GWPF. As I wrote earlier, “I stand by what I said, and I would have said it even if no-one had made the comments above – it was a response to a specific question by Joshua.” Joshua’s question was about Nic’s comment about SkS and activism; I couldn’t give the answer that I think to be true without voicing the criticism/opinion that I did.
Back on the topic of the post:
So, I’ve read the comments with interest. It seems that the discussion of the use of Jeffrey’s Prior and whether it is informative or not misses the point. The whole reason for the Prior is to reflect our prior knowledge of the system. If we use a Prior that not only does not reflect that knowledge, but which contradicts that knowledge (e.g. assigning zero the highest probability when we know with near 100% confidence that it ain’t zero), that isn’t merely an uninformative prior–it’s a stupid prior. Also, saying Jeffrey’s Prior is uninformative is not the same as saying it is objective–the choice of distribution determines the form of the prior. And different distributions yield very different priors.
If you are uncomfortable using a “smart” prior, just use likelihood, as there are well developed ways of carrying out inference from likelihood. Just remember that it isn’t a probability distribution.
oneuniverse,
This is turning into one of those rather tedious exchanges that is simply going in circles. My point is about consistency. If someone wants to dismiss other people’s views because of their apparent lack of expertise, then the same can be done to them. Noone can be expert at all aspects of a complex issue. If someone wants to dismiss other people’s work because of an association with an organisation, then the same can be done to them. So, if Nic Lewis wants to be treated as an expert, then maybe he should reciprocate in discussions. Bear in mind, that Nic’s actual expertise in this area is relatively low, even if his expertise in the details of statistical analysis is not. One the other hand, if he doesn’t care, then carry on. If Nic wants his works to be taken on its merits, then maybe he shouldn’t dismiss other people’s work because of an association. That’s really all. Treat others how you would like to be treated yourself. Simple.
ATTP, indeed. especially the last two sentences.
snarkrates “The whole reason for the Prior is to reflect our prior knowledge of the system.” I think it was Lewis’ intention that his estimate was an indication of what this particular set of observations imply in isolation, and the “minimally informative” prior was intended to accomplish that. I think the intention is reasonable (provided that caveat is provided when it is being discussed), but I don’t think the “minimally informative” prior really achieves that because it specifies the prior knowledge that we don’t know anything about ECS except that it is a scale parameter, i.e. it is an active statement of ignorance, rather than a passive non-inclusion of what we know a-priori. The point about likelihood seems reasonable to me.
“it’s a stupid prior.” I wouldn’t go that far, the peak at zero isn’t that important from a practical perspective as it is (apparently) ruled out by the likelihood, more problematic is the relative weights of values less than one and greater than one (which is roughly the direct effect of a doubling of CO2, and marks the boundary between net negative and net positive feedback). I don’t think that negative feedback (in the sense of reducing the direct effect of CO2) is anything like as plausible as net positive feedback, given what we know. While the peak at zero has no effect on the posterior range for ECS, I suspect the general shape of the prior does.
Statisticians quite often use assumptions that are known to be false a-priori, but the resulting analysis is often still sensible. For instance the usual test for the fairness of a coin (generally only used in teaching stats) takes the null hypothesis that the coin is exactly unbiased, i.e. the probability of a head is exactly the probability of a tail, which we know a-priori isn’t going to be exactly true because the coin is asymmetric. Likewise the null hypothesis for the test of a trend in GMST is that the trend is exactly zero, but that is almost certainly not true as well. The Laplace prior is a reasonable thing to do, provided the necessary caveats are given, and the assumptions involved are acknowledged and discussed.
Anders –
==> “I’m not quite following your point. I had two main points. One was simply that dismissing what someone has said because they don’t understand X, is an appeal to authority (implicitly, maybe). The other was that if you choose to do so when in a discussion with others who have relevant (but different) expertise, then it can degenerate into counterclaims, which seems a poor way to have a discussion.”
??? I was in agreement. Nic’s whole: No one else here “properly” understands Bayesian estimation except I. is a naked appeal to self-authority. It has no useful function in a discussion focused on building mutual understanding. It is like his guilt-by-association in that sense, or his disparagement of technical analysis on the basis of who wrote something or who they associate with.
I did say, “I’m not quite following” 🙂
I see where I went wrong. I thought you’d quoted something from me and then said
You were quoting a response to me. In my defense, I hadn’t yet had my coffee 🙂
My position is all GMST datasets are underestimating the GMST. Is there an assumption the GMST is known?
JCH,
Victor Venema seems to agree that we shouldn’t rule out that the observations might be running cold.
Dikranmarsupial,
I agree that if you have data that place a strong constraint on the quantity, the characteristics of the Prior are not that important. However, that is not the case here. No single source of data places strong constraints on ECS–particularly at the high end. Now the prior does make the high-end go away–so did Annan’s Cauchy prior. However, it is not a good choice in this case because:
1) the data do not dominate the resulting posterior
2) it contains information we know to be false–which violates the whole point of the Prior
If you want to see what a given dataset says in and of itself, it makes more sense to use likelihood. This does not take care of the high-end tail, which is why Nic did not do this, I suspect. So, I suspect that it was more an exercise in making the inconvenient aspects of the data go away than it looking at what the data actually say.
My experience using JP is that it is really only viable when your data are strong.
> Note, I’m not saying that SkS should or should not be civil, I’m saying that kind of thing is a negative contribution to a civil debate, which is a concern that you raised above, not me.
Hence it was a tu quoque. (It’s not the same as reciprocation, but the explanation will have to wait.) Whatever SkS does doesn’t imply anything regarding Dikran, whom only has direct control over what he himself writes. The tu quoque also acted as a deflection from the most interesting case Joshua presented, in a comment oneuniverse did not seem to take into consideration in his word-numbering argument:
http://www.climate-lab-book.ac.uk/comparing-cmip5-observations/#comment-436578
We’re far from the 20 words out of a “nearly 1000 word post” that oneuniverse mentioned earlier. The above comment clearly contradicts oneuniverse’s and Dikran’s agreed stance on how scientiststs should proceed. The only thing that remains is for oneuniverse to agree with Dikran that Nic’s argument above is fallacious.
The alternative is of course that onuniverse agrees with Nic’s above argument. If he does, he must at the very least acknowledge that Nic’s responses to James were based on double accounting. (If he could acknowledge that the whole counting business is a red herring, that would be nice too.) From Nic’s double accounting and twisted logic about confidence inspiration, it follows that Nic’s choice of a non-informative (or minimally informative, as he himself conceded in his email to Andrew Gelman) does not inspire confidence in the reliability and impartiality of his modulz.
snarkrates
“If you are uncomfortable using a “smart” prior, just use likelihood, as there are well developed ways of carrying out inference from likelihood.”
I agree: in particular likelihood ratio methods where there are no nuisance parameters, and profile likelihood methods where there are.
“If you want to see what a given dataset says in and of itself, it makes more sense to use likelihood.”
In my experience, when estimating climate sensitivity (profile) likelihood ratio methods give very similar inference to objective Bayesian methods using Jeffreys’ prior.
“This does not take care of the high-end tail, which is why Nic did not do this, I suspect”
False on both counts, like many of the things written about me in this thread.
If you read Lewis (2015), you will see that I did use profile likelihood ratio methods, as well as objective Bayesian methods. And I have found profile likelihood ratio methods actually tend to give slightly lower upper bounds.for ECS than using Jeffreys’ prior, probably because of their asymptotic normality assumption.
oneuniverse –
One more comment and then I’ll leave it at that (how many times have I made that claim and not lived up to it?)
Yes, if we start with different definitions, then this discussion won’t go anywhere. And more than likely, we’ll pick and choose our definitions like someone referencing only one of many definitions for a word in a dictionary as if that proves their usage authoritatively.
I’m less focused on some authoritative determination on these issues (which I think is like chasing tails, as the authority of a determination is most likely a product of biases and predispositions – I’m a descriptivist not a prescriptivist) than I am on considering how and where communication breaks down.
I’ll repeat that the point of focus for me is the functionality of Nic’s rhetoric w/r/t advancing communication. I’ll repeat that IMO, Nic’s “extra” as I’ve highlighted serves no positive function, unless point-scoring, identity-aggression, identity-defense, affirming group affiliation are your goals.
Let me take another of your comments to underline my point:
The fact that it’s “normal” (in the sense of commonplace – obviously, “normal” attitudes, in the sense of being some kind of baseline set of attitudes, also include those which aren’t like those Nic displayed) doesn’t mean that it serves a positive function. You say that ‘extra” scrutiny will improve someone’s work. Well, by the same sort of logic then, less than extra scrutiny – of the sort that Nic’s implying that he doesn’t apply to work not produce by SKSers such as his own work or those of people whose views he agrees with – would lead to sub-optimal results,
Indeed, the basic logic of “it’s normal, it will improve your work” is not different, IMO, than the strange logic of Nic’s where he says that he doesn’t discount something if he determines that it requires extra scrutiny because of who the author associates with. How does thinking that it requires extra scrutiny not come about if there hasn’t been, previously, a discounting?
And again, it isn’t really that someone applies such an approach that bothers me – because we all do it – but the defense of such thinking as you’ve offered, and as Nic justifies in an apparent belief that it’s just fine if he does it.
Sameosameo gets us nowhere. It doesn’t, actually, improve work or improve the salon of scientific exchange.
Nic,
I[f] you think there is something here that is simply not true, you are welcome to point it out, here or privately if you prefer. Not liking what’s been said, however, doesn’t qualify. I would add that my preference is that if you do want to claim that some things that have been said are not true, that you actually do point out where, as it is never my intention to allow things that are untrue to be said here.
Nic,
Your 17-83% range for ECS from Lewis (2015) is something like 1.1 – 2.4C. The IPCC likely range is 1.5-4.5C. So, your range overlaps with theirs, but largely cuts off the upper half of their range. As you well know, many other experts regard the ECS as likely being above 2C. It’s also the case that a number of the climate models with emergent constraints that compare well with observations tend to have ECS values above 3C. I know I’ve said this to you before, but if you continue to promote that the ECS is more likely below 2C than above, and that it might even have a reasonable chance of being below about 1.6C, you – or someone else – is going to have to find some way to explain this. Appealling to statistical techniques does not provide a suitable explanation for why physically-motived models (and other estimates too) suggest that the ECS is likely higher than your analysis suggests. The goal is to understand reality, not simply properly apply some kind of statistical technique.
I think Nic can include his double accounting into the “many of the things” that are presumably false.
Actually, I hadn’t seen this comment that Joshua highlights here. Nic makes a big deal out of a minor error in Cawley et al. (where they multiply instead of dividing). Unfortunately, these silly mistakes do happen. Making a big deal out of the fact that it was made and not noticed is rather silly – IMO. He also says:
which is odd in the first place, but also wrong as I think Kevin Cowtan is indeed a physicist (his first degree was Theoretical Physics).
> although they are academics none of […] is a physicist
Nic is not an academic. Contrary to what he tried to sell to Andrew Gelman, neither is he a climate scientist. Nic’s first and foremost an accountant, a fact that might not have the same impact as the usual “background in mathematics” understatement. While this may explain the naturalness of his double accounting, it doesn’t imply anything regarding the quality of his overall work. To presume otherwise would be fallacious.
Thanks for the link, Willard.
Climate sensitivity may be higher than we thought according to a new paper by Dr. Kate Marvel, Dr. Gavin A. Schmidt et al. on why climate sensitivity from transient obs are biased low:
Implications for climate sensitivity from the response to individual forcings
Abstract
Climate sensitivity to doubled CO2 is a widely used metric for the large-scale response to external forcing. Climate models predict a wide range for two commonly used definitions: the transient climate response (TCR: the warming after 70 years of CO2 concentrations that rise at 1% per year), and the equilibrium climate sensitivity (ECS: the equilibrium temperature change following a doubling of CO2 concentrations). Many observational data sets have been used to constrain these values, including temperature trends over the recent past1, 2, 3, 4, 5, 6, inferences from palaeoclimate7, 8 and process-based constraints from the modern satellite era9, 10. However, as the IPCC recently reported11, different classes of observational constraints produce somewhat incongruent ranges. Here we show that climate sensitivity estimates derived from recent observations must account for the efficacy of each forcing active during the historical period. When we use single-forcing experiments to estimate these efficacies and calculate climate sensitivity from the observed twentieth-century warming, our estimates of both TCR and ECS are revised upwards compared to previous studies, improving the consistency with independent constraints.
http://www.nature.com/nclimate/journal/vaop/ncurrent/full/nclimate2888.html#close
Pingback: 2015 in review | …and Then There's Physics
Pingback: 2015 in review | …and Then There's Physics
Pingback: What a surprise …. not | …and Then There's Physics
Pingback: Water vapour and climate | …and Then There's Physics
Pingback: ECS too low? | …and Then There's Physics
Pingback: Reconciling climate sensitivity estimates – part III, or IV? | …and Then There's Physics
Pingback: Combining different ECS estimates | …and Then There's Physics
Bayesian deconstruction of climate sensitivity estimates using simple models: implicit priors and the confusion of the inverse
James D. Annan and Julia C. Hargreaves (21 Apr 2020)
https://esd.copernicus.org/articles/11/347/2020/
“There is also a strand of Bayesianism which asserts more broadly that in any given experimental context there is a single preferred prior, typically one which maximises the influence of the likelihood in some well-defined manner. The Jeffreys prior is one common approach within this “objective Bayesian” framework. However, it has the disadvantage that it assigns zero probability to events that the observations are uninformative about. This “see no evil” approach does have mathematical benefits but it is hard to accept as a robust method if the results of the analysis are intended to be of practical use. In the real world, our inability to (currently) observe something cannot rationally be considered sufficient reason to rule it out. We do not consider objective Bayesian approaches further.”
🙂