Climate change

[Daniel1960]

How is your "future projection" any different than a "model predictions"?

 

Oh wait, models are bad and your projections are good...

 

So, are you saying that you would base your projections on models over data?

[kenberg]

I may start a thread devoted to the general question of "How do we decide what and whom to believe in areas where we are not at all experts or even particularly well informed?"

 

Let me take a crack at data versus models, starting with my weight. I can tell you wnat my weight is, or at least what the scale says that it is. I have a rough idea of what it has been over the past year and since I get weighed in the doctor's office I could get some recorded data there. I could say whether over the last several years my weight has gone up or down. This involves data. If I am going to project into the future, I think I need a model. What would happen if I cut out my evening glass of wine? I really like eating dried fruit and nuts, but they have calories so I might consider that. I exercise but I could do more. Exercise really involves some modeling. I have known people who argue that exercise can't help much because it just doesn't burn enough calories unless you maybe bike fifty miles a day or some such. But I think that exercise helps regulate an appetite. . I find that I am less likely, not more likely, to want to eat a Whopper after a nice hike. My tastes change. So we model. Primitively in the above example but still we model.

 

So it is not a matter of data or a model, if we are to project into the future and look to the likely results of action or inaction, we model. We must.

 

The distinction can be very useful when we look at criticism of a view. Do we doubt the accuracy of the data, or do we accept the data as accurate but question the realism of a model that projects future results from the data?

 

Anyway, data, if carefully gathered, can say something of where we are and where we have been. I doubt that data, on its own, can say anything of where we are headed. Even just some simple linear extrapolation is a model.

[hrothgar]

So, are you saying that you would base your projections on models over data?

 

No, I am saying that you are too stupid to understand that a "projection" is the same thing as a "model" and that both are based on data.

[Daniel1960]

No, I am saying that you are too stupid to understand that a "projection" is the same thing as a "model" and that both are based on data.

 

I think a little introspective action is needed here, as they are not the same. If you were to investigate further, you might find out why. Then again, you refer to others as being stupid, who appear to know more than yourself.

[hrothgar]

I think a little introspective action is needed here, as they are not the same. If you were to investigate further, you might find out why. Then again, you refer to others as being stupid, who appear to know more than yourself.

 

Webster's defined the word "Projection" as "an estimate of future possibilities based on a current trend". (Webster's offers a number of other definitions, how this is clearly the most appropriate)

 

http://www.merriam-webster.com/dictionary/projection

 

Wikipedia explains "Statistical model" as "A statistical model embodies a set of assumptions concerning the generation of the observed data, and similar data from a larger population. A model represents, often in considerably idealized form, the data-generating process. The model assumptions describe a set of probability distributions, some of which are assumed to adequately approximate the distribution from which a particular data set is sampled."

 

http://en.wikipedia.org/wiki/Statistical_model

 

Projection is nothing more than an applied use of a model and any kind of projection presupposes some kind of model.

[Daniel1960]

Webster's defined the word "Projection" as "an estimate of future possibilities based on a current trend". (Webster's offers a number of other definitions, how this is clearly the most appropriate)

 

http://www.merriam-webster.com/dictionary/projection

 

Wikipedia defines the expression "Statistical model" as "A statistical model embodies a set of assumptions concerning the generation of the observed data, and similar data from a larger population. A model represents, often in considerably idealized form, the data-generating process. The model assumptions describe a set of probability distributions, some of which are assumed to adequately approximate the distribution from which a particular data set is sampled."

 

Projection is nothing more than an applied use of a model and any kind of projection presupposes a some kind of model.

 

Reread your definition of a projection. A current trend is based on accumulated data, not a model.

 

Not that I agree with your definition. A projection could be based on either data, like your current trend or a model, based on deired inputs. Models typically include assumptions as to what the future may entail. This may or may not resemble the past data.

 

Here is an example showing the difference between projections based on models and those based on data:

 

http://www.skepticalscience.com/pics/Akasofu_Prediction.png

 

A model can be made to give any desired result, and is only as good as the selected input. Some models have behaved quite well, others, not so. Economists have a rough time with their forecasts, and the economy is less chaotic than the climate.

 

http://www.cbsnews.com/news/why-are-economic-forecasts-wrong-so-often/

[hrothgar]

 

Here is an example showing the difference between projections based on models and those based on data:

 

http://www.skepticalscience.com/pics/Akasofu_Prediction.png

 

 

Sorry if this comes across as an appeal to authority, but

 

1. I get paid hundreds of thousands of dollars a year to do statistical modeling

2. Before this, I was product manager for MATLAB's statistics system

3. Before this, I was fifth in my class at MIT

4. Before this, I did a Masters in Mathematical Economics and taught undergraduate statistics

5. Before this, I graduated Wesleyan University, where I was the top student in both the Economics and the Government departments.

 

I am mentioning this because I am trying to establish that I know a lot about both "modeling" and "data" and how these expressions are used in both professional and academic settings. So, with all due respect, when I say that I find your claims both idiosyncratic and laughable, it probably means something.

 

Let's take a look at the chart that you choose to illustrate your point:

 

I assume that the lower of the two dotted red lines is meant to illustrate a projection based on "data".

 

Guess what? This projection also presupposes a model. In this case, it would appear as if the model is a sinusoidal oscillation around a linear trend.

 

This is a model. (It's almost certainly a simpler model than the one used by the IPCC, but it is a model none the less and anyone with a background in stats or economics would agree with me)

 

I have been playing around a bunch in R today. I've been making using of a function called "lm" which can be used to generate trend lines like the one you see in that model. Want to guess what "lm" stands for?

 

(Here's a hint. The first word is "linear")

[Daniel1960]

Sorry if this comes across as an appeal to authority, but

 

1. I get paid hundreds of thousands of dollars a year to do statistical modeling

2. Before this, I was product manager for MATLAB's statistics system

3. Before this, I was fifth in my class at MIT

4. Before this, I did a Masters in Mathematical Economics and taught undergraduate statistics

5. Before this, I graduated Wesleyan University, where I was the top student in both the Economics and the Government departments.

 

I am mentioning this because I am trying to establish that I know a lot about both "modeling" and "data" and how these expressions are used in both professional and academic settings. So, with all due respect, when I say that I find your claims both idiosyncratic and laughable, it probably means something.

 

Let's take a look at the chart that you choose to illustrate your point:

 

I assume that the lower of the two dotted red lines is meant to illustrate a projection based on "data".

 

Guess what? This projection also presupposes a model. In this case, it would appear as if the model is a sinusoidal oscillation around a linear trend.

 

This is a model. (It's almost certainly a simpler model than the one used by the IPCC, but it is a model none the less and anyone with a background in stats or economics would agree with me)

 

I have been playing around a bunch in R today. I've been making using of a function called "lm" which can be used to generate trend lines like the one you see in that model. Want to guess what "lm" stands for?

 

(Here's a hint. The first word is "linear")

 

This is all very nice I graduated over 30 years ago from the University of Michigan. While I cannot compare to your statistical resume, I did take graduate course in statisctics, and use them regularly in my scientific work (chemistry).

 

I think you are grasping when you refer to every trend as a model. Back when we had to do all our calculations by hand, that may have been the case. Today, modelers have the luxury of computers, which can perform multiple, complicated calculations in the fraction of a second, often changing several parameters simultaneously. Nowadays, the term "model" refers to much more than a simple linear trend.

[Daniel1960]

Yes, this paper, published in October, 2014, that you recommended earlier in this thread, illustrates the extremes of possible sea level rise by the year 2100:

 

 

Although it is possible that sea level rise could be 48 cm on the low end or 180 cm on the high end, these extremes "are just less probable than others." And indeed, that is true; I agree completely.

 

What I have trouble understanding, then, is why you put forward an even more extreme position on sea level rise:

 

 

 

Representing reality, we have some recent news: Arctic Ice Reaches a Low Winter Maximum

 

Using a projection (see hrothgar's definitiona) based on the current trend for sea level rise (3.2 mm/yr from satellite calculations, less by tidal gauges), the projected sea level rise by 2100 would be ~26 cm. The projections in the linked paper, assumes a thermal expansion equivalent to 32 cm of rise (likely range 25-39 cm). Interestingly, the reference paper they used for their assumptions, calculated a sea level rise of 13 cm by 2100 using CMIP5 models, if warming is kept below 2C, and 28 cm, under worst-case warming scenarios. Jevrejeva, et. al. used the high-end warming only in their calculations. Indeed, this paper shows not a range of potential sea level rise, rather it shows the range of upper limits to sea level rise. You are comparing a likely value of sea level rise based on current trends to an upper limit to sea level rise based on modelled assumptions. By the way, the lower end of their maximum sea level rise is 29 cm.

 

http://onlinelibrary.wiley.com/doi/10.1029/2012GL052947/pdf

 

Again, the model is only as good as the data inputted or assumption made.

[hrothgar]

 

I think you are grasping when you refer to every trend as a model. Back when we had to do all our calculations by hand, that may have been the case. Today, modelers have the luxury of computers, which can perform multiple, complicated calculations in the fraction of a second, often changing several parameters simultaneously. Nowadays, the term "model" refers to much more than a simple linear trend.

 

 

I readily admit that the increase in computing power has enabled people to develop more complicated models.

However, expression model still includes naive techniques like fitting a trend line to the data.

 

Going back to the chart that you original proposed.

 

A trend line is one possible way to describe the relationship between the dependent and the independent variable

A trend line + a sinusoidal component is another

 

People need a way to differentiate between these two sets of assumptions.

They use the word model to formally describe the relationship.

 

As far as I can tell, you use the word "projection" to describe models that you like and "model" to describe models that you don't like.

I don't find this particularly compelling.

[hrothgar]

 

Nowadays, the term "model" refers to much more than a simple linear trend.

 

 

BTW, just so its clear that I am not making ***** up, here is a useful little quote from the R documentation describing how one specifies formula's for linear models.

 

 

The models fit by, e.g., the lm and glm functions are specified in a compact symbolic form. The ~ operator is basic in the formation of such models. An expression of the form y ~ model is interpreted as a specification that the response y is modelled by a linear predictor specified symbolically by model. Such a model consists of a series of terms separated by + operators. The terms themselves consist of variable and factor names separated by : operators. Such a term is interpreted as the interaction of all the variables and factors appearing in the term.

 

In addition to + and :, a number of other operators are useful in model formulae. The * operator denotes factor crossing: a*b interpreted as a+b+a:b. The ^ operator indicates crossing to the specified degree. For example (a+b+c)^2 is identical to (a+b+c)*(a+b+c) which in turn expands to a formula containing the main effects for a, b and c together with their second-order interactions. The %in% operator indicates that the terms on its left are nested within those on the right. For example a + b %in% a expands to the formula a + a:b. The - operator removes the specified terms, so that (a+b+c)^2 - a:b is identical to a + b + c + b:c + a:c. It can also used to remove the intercept term: when fitting a linear model y ~ x - 1 specifies a line through the origin. A model with no intercept can be also specified as y ~ x + 0 or y ~ 0 + x.

[Daniel1960]

I readily admit that the increase in computing power has enabled people to develop more complicated models.

However, expression model still includes naive techniques like fitting a trend line to the data.

 

Going back to the chart that you original proposed.

 

A trend line is one possible way to describe the relationship between the dependent and the independent variable

A trend line + a sinusoidal component is another

 

People need a way to differentiate between these two sets of assumptions.

They use the word model to formally describe the relationship.

 

As far as I can tell, you use the word "projection" to describe models that you like and "model" to describe models that you don't like.

I don't find this particularly compelling.

No, a projection is a predictor of a future scenario. It may be based on current trends, past data, or mathematical models. A projection is not a model itself. Maybe I am biased against modelers, because I have witnessed too many times when a (computer) model was less reliable than scientific research. Does that satisfy your definition?

[hrothgar]

Maybe I am biased against modelers ... Does that satisfy your definition?

 

I would agree with the first six words

[Daniel1960]

I would agree with the first six words

Would you admit that you are biased against researchers, in favor of modelers?

[hrothgar]

Would you admit that you are biased against researchers, in favor of modelers?

 

I don't consider the distinction to be salient

[kenberg]

I have an honest question here. We can imagine two approaches to predicting from data. I will formulate them ine extreme ways to be clear about the difference. One way is to say "Just ive me the numbers. I don't care if the numbers represent temperatures in the Antartic or the price of McDonalds stock. I will analyze the numbers and predict the future." Another approach is to care very much about what the numbers represent and to bring all of the accumulated knowledge of cause and effect to bear in trying to predict what happens next.

 

Each approach has it pitfalls. The first ignores influences that might well be relevant, the second brings in beliefs about causality that might if might not be correct. Some sort of combination seems best. The problem is too complicated to expect direct verification such as with relativity in 1919 (and even that needed further work).

 

Anyway, I am wondering, after we soften the extreme formulation to better match with reality, if this is the distinction being argued over. Or, if that is not the distinction, what is?

 

Side note: I forget where I read this but supposedly before the 1919 observations were made a journalist asked Einstein what would happen if his predictions were not borne out. He replied "In that case I would feel very sorry for the Almighty, because the theory is correct".

[hrothgar]

I have an honest question here. We can imagine two approaches to predicting from data. I will formulate them ine extreme ways to be clear about the difference. One way is to say "Just ive me the numbers. I don't care if the numbers represent temperatures in the Antartic or the price of McDonalds stock. I will analyze the numbers and predict the future." Another approach is to care very much about what the numbers represent and to bring all of the accumulated knowledge of cause and effect to bear in trying to predict what happens next.

 

Each approach has it pitfalls. The first ignores influences that might well be relevant, the second brings in beliefs about causality that might if might not be correct. Some sort of combination seems best. The problem is too complicated to expect direct verification such as with relativity in 1919 (and even that needed further work).

 

Within the statistics and machine learning community, people distinguish between parametric and non-parametric models.

 

Regression analysis is a classic example of a parametric model. The techniques pre-suppose a parametric model that describes the relationship between the dependent and the independent variables. There are also any number of "black box" modeling techniques (neural networks, smoothing splines, boosted regression trees, etc.) These techniques do not presuppose any kind of parametric relationship, rather the try to identify patterns found within the data. There are also hybrid techniques such as GAMs (General Additive Models). There are advantages and disadvantages to each of the methodologies. As a rule, people tend to prefer using parametric techniques, especially when you are doing extrapolation rather than interpolation.

 

What I find amusing about this all is that Daniel is simultaneous decrying the complexity of the IPCC models while stressing the need to adopt purely data driven techniques which require complex, non-parametric modeling. Going back to Daniels own example, he holds up a linear trend with a sinusoidal oscillation as his prototypical data driven technique when this is textbook example of parametric modeling.

 

The degree of cogitative dissonance is astounding...

[Daniel1960]

I have an honest question here. We can imagine two approaches to predicting from data. I will formulate them ine extreme ways to be clear about the difference. One way is to say "Just ive me the numbers. I don't care if the numbers represent temperatures in the Antartic or the price of McDonalds stock. I will analyze the numbers and predict the future." Another approach is to care very much about what the numbers represent and to bring all of the accumulated knowledge of cause and effect to bear in trying to predict what happens next.

 

Each approach has it pitfalls. The first ignores influences that might well be relevant, the second brings in beliefs about causality that might if might not be correct. Some sort of combination seems best. The problem is too complicated to expect direct verification such as with relativity in 1919 (and even that needed further work).

 

Anyway, I am wondering, after we soften the extreme formulation to better match with reality, if this is the distinction being argued over. Or, if that is not the distinction, what is?

 

Side note: I forget where I read this but supposedly before the 1919 observations were made a journalist asked Einstein what would happen if his predictions were not borne out. He replied "In that case I would feel very sorry for the Almighty, because the theory is correct".

 

A little bit of both is needed. The dependence on either varies with the situation. Newton's first law of physics states, "an object in motion stays in motion with the same speed and in the same direction unless acted upon by an unbalanced force." Hence, the prevaling force (trend from the data) will remain in effect until another force acts upon it to cause a change. In a complex system (such as the Earth's climate), several competing forces are acting simultaneously. What we are witnessing is the summation of these forces. Some have tried to consolidate these forces into a simple model, based on a few parameters, on the assumption that the other parameters are either too small to contribute or cancel each other. From a pure numbers standpoint, all these parameters are incorporated into the accumulated the data, and the observed trend is the result.

 

History (and physics) tells us that nothing will continue indefinitely. There are self-imposed limits. The temperature of the Earth cannot rise forever. Past climates have experienced wide ranges, sometimes ending suddenly. In some cases, we have identified large external forcings as the most likely case. Other times, we simply do not know. Predicting large future changes in such a system is difficult, if not impossible, due to our limited knowledge. Conversely, stating that certain forces are likely to alter the trends slightly, in one direction or the other, based on sound scientific evidence, is reasonable. Our planet has absorbed all the internal and external forces, and returned the current climate situation. Some claim that our planet is precariously perched on a delicate balance of nature, and that the slightest disturbance will throw us over some magical tipping point. Paleoclimatology shows us that this is not the case. Newton's third law states, "for every action, there is an equal and opposite reaction." Natural feedbacks tend to act in an opposite direction to the inital force. While this will not necessarily return to the original scenario, the concept of a given force leading to larger and more numerous positive feedbacks, contradicts most of physics. Otherwise the search for the perpetual motion mahine would have returned several products, by now.

 

This returns use to Newton's first law. The climate is likey to continue on its current trend, unless acted upon by an unbalanced force.

[kenberg]

Regression analysis is a classic example of a parametric model. The techniques pre-suppose a parametric model that describes the relationship between the dependent and the independent variables. There are also any number of "black box" modeling techniques (neural networks, smoothing splines, boosted regression trees, etc.) These techniques do not presuppose any kind of parametric relationship, rather the try to identify patterns found within the data. There are also hybrid techniques such as GAMs (General Additive Models). There are advantages and disadvantages to each of the methodologies. As a rule, people tend to prefer using parametric techniques, especially when you are doing extrapolation rather than interpolation.

 

Thanks for this. I want to pursue it a bit.

 

First I should come out of the closet. I am pre-disposed to believe that human activity is a problem for the planet. Immediately maybe, eventually certainly. There are seven billion or so of us and if we all start putting 20K miles a year on an SUV and put air conditioning in our dog houses I don't think that it will work out well.

 

As to the statistics. Despite my mathematical training I am unfamiliar with modern statistical techniques. Not totally ignorant of them but we are not on close terms. I expect I could learn some of this but more relevantly I think that if I look into it some I will be able to pretty well tell whom I can trust and whom I cannot. Even that would take work, but doable I think.

 

A question: Does causation fit into your scheme above? Or perhaps I should ask where and how it fits. Regression models, on their own, will not disclose causation, right?

 

My thinking is that we almost certainly have to do something, and I expect that if I got into this more I would drop the "almost". It's important to be effective, and for this it is important to be as sure as we can about cause and effect.

[hrothgar]

 

A question: Does causation fit into your scheme above? Or perhaps I should ask where and how it fits. Regression models, on their own, will not disclose causation, right?

 

 

The short, simple answer is "no". Regression techniques are used to describe correlation, not causation.

 

The longer, more complicated answer is "not really". Regression techniques work by generating a surface that minimized the sum of the squared errors between predicted and actual. Traditional regression techniques associate all of the variance with the Y axis. However, there are other techniques like orthogonal regression that associate variance with both the X and Y axes. Arguable, when you make these sorts of assumptions, you are getting close to ascribing causation.

[Daniel1960]

Kenberg,

A recent paper discusses some of the issues with the global climate models.

 

http://arstechnica.com/science/2015/01/are-climate-models-biased/

 

With some being better than others.

 

http://phys.org/news/2014-07-vindicates-climate-accused.html

[mike777]

A report in my local paper today says by 2024 the sea level may rise 10.6 inches off our coast.

 

The Coastal Resources Commission said there are 2 reasons:

1) ancient geological forces are causing part of the NA continent to subside.

2) Shifts in the speed and position of the Gulf Stream are pushing the seas higher in our region.

[kenberg]

Kenberg,

A recent paper discusses some of the issues with the global climate models.

 

http://arstechnica.c...-models-biased/

 

With some being better than others.

 

http://phys.org/news...te-accused.html

 

Thank you. I took a quick look at both and I will, I think, get back for a longer look. Climate change is one of many areas where I reluctantly accept a civic obligation to at least try to be informed. It's obviously both difficult and important.

[y66]

There are seven billion or so of us and if we all start putting 20K miles a year on an SUV and put air conditioning in our dog houses I don't think that it will work out well.

If all of the approximately 4 billion people between the ages of 16 and 65 now put 20K miles a year on an SUV in 2015 and every year thereafter, and their SUVS get 20 MPG now, increasing 2 percent a year in efficiency thereafter, and the driving population grows 0.8 percent a year, the resulting cumulative C02 emissions from SUVs alone will hit 565 gigatonnes sometime in 2030. If you want to quibble with the driving age estimate and the mileage estimates, fine, make it 2030 plus or minus 3 and don't forget to add something for the dogs.

[y66]

What do conservative policy intellectuals think about climate change? by Ben Adler at grist.org.