Discussion > Hypothesis testing in climatology
Shub
This discussion thread is intended to be about testing the science. Feel free it ignore it if you prefer ignorance.
Entropic man
This is a discussion thread about science but you are 'discussing' it with people right? The same people whom you consider stupid?
To me, the fact that you would come here and invest time in laying out your ideas to a target audience, trying to convince them and yet despise the same audience produces cognitive dissonance.
Your so-called science consists of simplifications, assumptions and circular logic that *you* are not willing to examine. Hypothesis testing in climatology is *impossible* because there is no control. This is a basic fact of experimental science. In the strict sense, you cannot test hypotheses when only observation is possible.
Where is your acknowledgement of this basic problem?
shub
Compare this:
....the first is his rudeness about Trenberth. I automatically doubt anything written in the language of propaganda.
to
If you follow the denier line and say all the observations are wrong, then you have no evidence at all!
shub
Shub
Unpack please.
Both statements are valid. Please be more specific about your complaint.
Entropic man
Think about what you wrote, yourself. Think about your words and language.
shub
Shub
"Hypothesis testing in climatology is *impossible* because there is no control. This is a basic fact of experimental science. In the strict sense, you cannot test hypotheses when only observation is possible."
Thank you. By your logic, science can only be done by laboratory experiment, and anything else is not science. How do you suggest scientists study anything too large, distant or long term to fit into their laboratories? How do you overcome the lack of formal controls? I suggest you look to astronomy as a successful science limited largely to observation and study their methods.
Among those who study peopoganda your argument is known as "impossible standards". You demand that all evidence meet a standard impossible to achieve in the real world, then use that impossibility as an excuse to reject it.
Regarding my comments.
Resort to rudeness is not evidence against an argument, it is an attempt to disguise the weakness of your own case. It tends to be a propaganda technique, and a good sign that the author has a bias and therefore should not be taken at face value.
In my opinion a number of commenters here meet definitions 2 and 5 in the first group and 2,3,7 and 8 in the second group.
Entropic man
Shub, Entropic Man started this thread with a statement about what he thinks I think. When I stated what I think I think, he invited me to go elsewhere. I think that sums up the little universe he keeps on his desk top.
Borrowing from a recent a article by Paul Voosen about Dan Kahan (via Judith Curry):
"There are more things that are plausible for what’s going on than are true," Kahan pleaded. "The only way to make progress is to use empirical methods to rip from the sea of the plausible the thing that actually matters. Otherwise we drown in storytelling."
EM has his own particular story and he's going to tell us. Fair enough, as long as he doesn't expect people to walk away believing it as presented. He often states that it is up to critics to tell a better story, ignoring the fact that the 'true' story might be 'No story. Move along'.
To be taken seriously, it is the scientist's job to investigate as many other plausible stories as possible for themself, before bringing it to peers or critics. When an obvious plausible story is waved away or ignored completely then the audience rapidly loses any serious interest.
michael hart
Michael hart
An excellent suggestion. To the extent that this can be done in the laboratory the physics of the greenhouse effect etc is well established. Now, how do you suggest the same empirical methods should be applied on a planetary scale?
Entropic man
Michael hart
It is not a rhetorical question.
Questions such as the degree to which the observed changes in temperature are due to CO2 or natural variation can only be settled in the field, not the laboratory.
On this thread I am less interested in the answer, and more interested in how to go about getting the answer.
You may have plausible stories, but they have to be tested like any other hypothesis. Like it or not, a laboratory is not the place to do it. Rather than just objecting, I would welcome constructive ideas on how to go about it.
Entropic man
This discussion thread is intended to be about testing the science.
No, you started off by cherry picking data and some spurious rubbish about the pause ending because Giss data had a minor increase.
Considering the known faults with Giss and its smearing of data over the poles where it has no readings it was a farce.
Best ignored as per EM's advice.
Pity any pupils he taught this rubbish to.
Breath of Fresh Air
"Thank you. By your logic, science can only be done by laboratory experiment, and anything else is not science."
Say what? I said hypothesis testing cannot be done without experiment in the strict sense. Is that the same as saying 'science can only be done by laboratory experiment'? What is 'hypothesis testing' to you and is it the same as 'science'?
Right from the beginning of this thread, you have shown no sign that you understand this important distinction. Before I explain it to you pedantically, would you please be explicit and lay out the differences between 'hypothesis testing' and 'science' for me, in your own words?
shub
BoFA
Entropic man did a similar thing with sea levels a year or so ago why the rate of rise fell and then recovered to a rate above the average before "settling down" again. Harrabin would be proud of him ;-).
sandyS
Science is an attempt to increase understanding of reality by rational investigation, generally known as the scientific method.
Hypothesis testing is part of the scientific method.
Just to remind you, this generally starts with observation of reality.
From those observations one develops possible explanations, hypotheses.
These are used to predict the outcome of experiments or new observations.
The results of the experiments or new observations are then compared with the predictions of the hypothesis, which is then accepted, modified or scrapped as appropriate.
Entropic man
Entropic man
So from what you say there are explanations of how CO2 sometimes heats the lower atmosphere, sometimes the sea surface and sometimes deep oceans where it ca't be measured, amongst other things. But, more importantly when it's going to change back to atmospheric warming and back to deep ocean warming after that. Unless the current situation of deep ocean warming is unique.
sandyS
entropic, this is getting bad
"New observations" are not the same as "experiments".
'New observations' may coincidentally be the same as the predictions your hypothesis makes. Such an occurrence cannot be distinguished from confirmation of predictions. The null hypothesis cannot be rejected.
The title of your post is 'hypothesis testing', not 'hypothesis modification' or 'scrapping'.
I am not trying to beat up on the argument. When it comes to accepting this simple and basic limitation, supporters of the orthodoxy usually resort to an emotional "we have only one Earth" distraction. Climate scientists are not stupid - this is why they use computer models.
It is true that experiments, though not equivalent to the rigorous two-system set-up, are still possible in observational systems. All that is needed is long enough a period of observation divided into experimental and control time periods. As of now, we do not have all the elements required. The observations do not stretch long enough back in time.
What observations we do have -proxy data, however, shows that temperature change of similar magnitude occurred in the control period as is observed in the experimental period. Problem #1
Within the homogenous period of high-quality observation - instrumental data, there are two periods of differing human hydrocarbon combustion. This is the period before WWII, and the one after, which encompass the Western post-war industrial productivity boom. The observed temperature rate of rise is roughly the same between the two periods. Problem #3
Again, within the homogenous high-quality instrumental period, there are periods before and after predictions were made, pre-2001 and -post. The observations do not match the predictions. Problem #3.
The UNFCCC was constituted before the attribution argument was conclusively set forth by the IPCC. Problem #4.
The climate models encapsulate current understanding of the complex climate system and incorporate the cycle of observation-comparison-modification that you speak of. Their output does not match observations. Problem #5.
Formal hypothesis testing is not possible at short time-scales with the climate system. What 'testing' is possible is currently underway.
shub
EM wrote:
"95% confidence limits. In the context of Hadcrut4, they define their global temperature averages with an uncertainty quoted as +/- 0.1C (+/-2Standard Deviations). 95% of these averages should be within 0.1C of the actual average temperature."
That's fair enough, note the average temperatures vary considerably, ie 1983 - 1985 is a difference of about 0.7 Deg C.
"I applied this to the trend lines. If the trend is valid 95% of the quoted averages should be within 0.1C of the trend line."
No they shouldn't, you _are_ just making this up.
Sure look at the graph you posted....
http://www.woodfortrees.org/plot/hadcrut4gl/from:1970/to:2015/every/every/plot/hadcrut4gl/from:1970/to:1998/every/trend/plot/hadcrut4gl/from:1998/to:2013/every/trend
In ~1983 the trend was just over 0, the hadcrut4 value was ~0.4 higher.
I thought you were supposed to be a physics teacher?
Nial
EM - whenever you talk about "95% confidence limits" (ie jumping directly from standard deviation to confidence interval) isn't there an implication that the relevant distributions are normal? Assuming normality without good reason usually leads to statistical nonsense.
Martin A
Martin A
Most temperature data are a good fit to a normal distribution. Some, such as climate sensitivity, are fat-tailed distributions.
Nial
The 95% confidence limits reflect the internal variation in the climate, weather etc. If internal variation was all that influenced the measurements the graph would be flat with most of the measurements within 2SD of the same level.
A few random excursions beyond those limits are to be expected even in the absence of any external variation. A prolonged excursion will be due to an external variable such as a volcano or a change in aerosol pollution.
If the short term variations are being superimposed on a long term change in conditions you get a sloping trend line.
The actual temperature data contains elements of all three. There is internal variation, larger variations due to external variation, and an underlying upward trend. The fun is in identifying, quantifying and explaining them.
Entropic man
Relevant to those who wish to draw straight lines through data, as opposed to look at it, Judith Curry higlights a "Superb post by Matt Briggs on the bogus use of statistics on temperature series". And I agree.
Titled "Netherlands Temperature Controversy: Or, Yet Again, How Not To Do Time Series", Briggs reminds us of the need to 'touch-base' with reality and how we view the data.
michael hart
Shub
We have a surprising level of agreement. You probably summed it up in the term "formal". It is not possible to test many climate phenomena by the formal scientific method. You have to use the less formal methods common in field science.
I do not regard this as invalidating scientific method in our context , though some do.
Let's see if I can address your five problems.
1) It is hard to find comparison control periods.Periods of similar temperature conditions such as the late Pliocene do not have the rates of change.seen in recent times. Perhaps the closest area the early cooling periods of other recent interglacial.
2) The rise in CO2 began around 1880 (Law Dome). You are comparing two periods of rising CO2, both with a rising temperature. A better recent control would be 1850-1880, a period of constant CO2 and constant temperature.
3) and 5) The earliest serious model, Hansen's 1d model, ignored everything except CO2 and ocean heat uptake. It came close to current warming.
More recent models incorporate more variables and some of them bring increased uncertainty. Aerosols are probably the clearest example. Most of the CMIP5 model runs overestimated warming because they underestimated aerosol negative forcing. The runs with most aerosols gave most accurate projections. When the models match observed aerosol conditions, they match observed temperatures.
That is a reason to regard models as a viable substitute for controlled trials. Like any experimental design they need work, but not rejection out of hand.
4) You'll need to explain that a bit more.
"Formal hypothesis testing is not possible at short time-scales with the climate system. What 'testing' is possible is currently underway."
Yes. For professionals models are their tools.
For an amateur less sophisticated methods have to suffice. In my own informal way that is what I do. For example, somewhere around here Radical Rodent predicts cooling, The graph I drew is an informal way of comparing his expectation with mine.
Entropic man
Michael hart
The post itself reflects the all or nothing approach typical of sceptics using the "impossible standards" tactic.
What I found more interesting was the exchange between David Appell and Bob Kurland in the cooments. Read it. It would save us rehashing the same arguments here.
Entropic man
Entropic, you are commenting saying 'these are my general observations' , the point of which is lost on me.
I stated that
[1] hypothesis testing is not possible in climate science.
[2] consequently, the discipline is an observational one prone to confirmation bias
I am interested in knowing whether you agree or not.
It is an all-or-nothing when it comes to hypothesis testing because hypothesis testing is not possible. What makes you think this is a 'tactic'? You started this thread. It is an impossible standard because it is.
This is a crucial distinction because there are sciences where hypothesis testing is possible.
shub
In response to my point EM wrote:
"Nial
The 95% confidence limits reflect the internal variation in the climate....." and loads more waffle that didn't address the point.....
"I applied this to the trend lines. If the trend is valid 95% of the quoted averages should be within 0.1C of the trend line."
This is completely incorrect. The trend can be valid and have >>5% of the samples outside 0.1C of the trend line.
Again, look at the distribution of temperature values wrt the trend from 1970 - 1998 in the graph you posted....
http://www.woodfortrees.org/plot/hadcrut4gl/from:1970/to:2015/every/every/plot/hadcrut4gl/from:1970/to:1998/every/trend/plot/hadcrut4gl/from:1998/to:2013/every/trend
You cannot take the current deviation of ~0.1 C from the trend as 'proof the pause has stopped'.
Nial
EM Wrote:
> 1) It is hard to find comparison control periods.
No it isn't. 1910 - 1940.
http://www.woodfortrees.org/plot/hadcrut4gl/from:1910/to:1940/every/trend/plot/hadcrut4gl/from:1970/to:2000/every/trend/plot/hadcrut4gl/from:1840/to:2015
What makes the warming from 1970 - 2000 'dangerous and unprecedented' when there's a very recent almost identical precedent?
Why is CO2 to blame for the second period of warming but not the first?
A simple question, but still no answer.....
Does this not blow the 'hypothesis' out of the water?
Nial
How come entropic man writes all this stuff here and takes up people's time and goes somewhere else and calls the same people 'stupid'?