Jump to content

Talk:Climate change/Archive 10

Page contents not supported in other languages.
From Wikipedia, the free encyclopedia
Archive 5←Archive 8Archive 9Archive 10Archive 11Archive 12→Archive 15

Image:Cost-of-storms-by-decade.gif

Is displaying extrapolation original research?

Recently added figure, revised on the right as per requests below. Note that neither figure includes historical data from 2000-2005. Original caption: "The cost of extreme weather is increasing and could reach a trillion dollars by 2050. source data: IPCC, 2001."
Better revision. Caption: The cost of extreme weather is rising rapidly and could reach 350 billion U.S. 2001 dollars per year by 2025. source data: IPCC, 2001. Some of the cost increase is due to added exposure such as building on the coast." The logistic sigmoid fit from the underlying cause did not fit the higher resolution data as well as the equation shown on the graph. James P. S. 00:16, 24 December 2005 (UTC)

While I think it is a good thing for wikipedia to talk about the changing financial cost of extreme weather, this figure seems to be engaging in a fairly strong case of original research. The data cited with the figure is all historical, and apparently provides the circles on the plot as decadal averages. This part is fine. It is the choice to draw a smooth curve through the data (cubic polynomial) and then imply this predicts losses significantly into the future that bothers me. The fact that the 1990s had sharply higher losses is important, but using this to predict > 10 times higher losses in 2050 is fairly extreme. In reality the economic losses from severe weather depend on a complex array of social and environmental factors and we really ought to have a source for the predictions as well as the observations.

Please read WP:NOR. Wikipedia does not create new scientific analyses, but rather relies on those that have been published. I'm fairly confident no one has published your model of the increase because it does not take into account the broad uncertainties in the social, economic, and environmental factors that will affect the future of that curve. I could easily imagine it being just as likely to saturate as continue to accelerate. We cannot publish future predictions of economic loss due severe storms without an external justification for the numbers we choose. Dragons flight 21:14, 19 December 2005 (UTC)

That extrapolation graph has been published in other fora. What kind of an authority would satisfy you here? James P. S. 21:52, 19 December 2005 (UTC)
Well, to begin with, extrapolation beyond the data is misleading without a confidence interval. The other issue is the use of the curve that was used - does it generate a significantly better fit than a linear (or exponential) curve? Even if the fit is better, you have to take into account the reduction in DF. More important is the underlying model - you can safely enough use "whatever fits best" when interpolating - it's just a phenomenological description. But if you are extrapolating, and extrapolation that far beyond the data, you need to use some sort of mechanistic model. For the purpose of Wikipedia there's the whole NOR thing as well, of course. Guettarda 21:21, 19 December 2005 (UTC)
I agree that confidence intervals would be great and I encourage you to create and upload your own graph with them. In the mean time, I believe the 1 DF third-degree polynomial R2=0.98 beats the 2 DF second-degree polynomial where R2 was 0.6, and the linear fit by even more, if I remember correctly. Any extrapolation is better than nothing. James P. S. 21:52, 19 December 2005 (UTC)
Dragons flight is right - we should not present predictions that have not been made by good external sources. Rd232 talk 21:24, 19 December 2005 (UTC)
I'm a new user, give me a break. If you grant that I am "good" then does the fact that I made that graph a year ago count as "external" or are you claiming that by creating a user account I am no longer able to contribute because I am "internal"? James P. S. 21:52, 19 December 2005 (UTC)
"Good external sources" means a reputable (preferably peer-reviewed) publication. Rd232 talk 22:01, 19 December 2005 (UTC)
I'm sorry if this dispute is making wikipedia seem an unfriendly place - sadly you picked one of the most contentious articles there is to start making edits. The question is not about whether you are "good" or "external" - it is simpler. Has this graph been published elsewhere? If not then it counts as original research. It is also highly questionable to extrapolate in this way for reasons already outlined. Please don't take this personally but try and understand that to make this article credible when this topic is very controversial requires that it is based on well referenced publications.--NHSavage 22:08, 19 December 2005 (UTC)
Fair enough: the graph was published on Slashdot in June. Thank you. James P. S. 22:33, 19 December 2005 (UTC)
I would point out in my defense that Dragon's flight has been selectivly reversing graph x-axes. Please see: User_talk:Dragons_flight/Images --James S. 19:58, 28 December 2005 (UTC)

Nrcprm2026 please, please, please:1) sit down and have a nice cup of tea or your favourite calming drink 2) read this page on wikipedia which explains what counts as original research especially WP:NOR#What_counts_as_a_reputable_publication.3F and then 3) think "does my plot count as original research?"--NHSavage 22:46, 19 December 2005 (UTC)

Please note the reviewer's comments on the original Slashdot publication. Do those not count as de facto peer review? If not, why not? James P. S. 22:50, 19 December 2005 (UTC)
I've been following this discussion and will weigh in here, on Slashdot as de facto peer review. Slashdot, as said by William M. Connolley, is essentially self-publication. So are the respondant's comments. Peer review ia a much more formal process and referees are selected by editors, with consideration of their credit and reputation. --Aude 22:58, 19 December 2005 (UTC)
Slashdot includes a reputation credit system called "karma" James P. S. 19:59, 24 December 2005 (UTC)
...which does not, by any objective measure, constitute Peer review. Anastrophe 20:03, 24 December 2005 (UTC)
I am not certain that there can be any "objective" measure of peer review quality which would not include Slashdot's karma system as at least partially qualifying for the term. However, that doesn't matter because I have promised to get third-party climate blog reviews of the graph below. Please do not add or remove the graph from any non-talk pages until I have returned with those outside comments, as also promised below. James P. S. 21:29, 24 December 2005 (UTC)

The original research is a problem. James, your having posted it onto slashdot it not enough. You ought to understand that, but if not: slashdot is essentially self-publication. It doesn't really differ from posting to a newsgroup. You don't have to show, "has this graph been seen before", but "has this graph been published in a reputable forum", where forum is preferrably a scientific journal (or reputable blog such as RealClimate...). William M. Connolley 22:44, 19 December 2005 (UTC)

As for the "original research" claim, again, I disagree. Extrapolation is not research. I spent hours revising the graph per the extensive requests above, and if you don't like it because the resulting revision shows a steeper rate of increase than the original, then why don't you make sure that all those people who have provided specific suggestions for improvement here which I acted on agree with you before deleting it again, please? James P. S. 15:45, 21 December 2005 (UTC)
While I appreciate the effort you put into it, it's still "original" work. Extrapolation is OR - it's a conscious decision to pick one model over another - which attributes mechanistic meaning to the model. It does not provide rationale behind model selection. You have not discussed why one model is superior to the other, and you have not convinced your colleagues that your model selection was reasonable (which is what happens with a peer-reviewed pub). So it's still unpublished, it's still OR, and it choice of the models amounts to expressing an opinion - an opinion which is not supported in the original place of publication. Guettarda 16:11, 21 December 2005 (UTC)
Extrapolation is just another way of displaying historical data. Just as there are good and bad photographs, there are good and bad extrapolations. Does anyone have any reasons why the new extrapolation which was based on the comments above counts as "original research" or should otherwise be witheld? It wasn't I who asked for the changes which, when implemented, produce a much more alarming graph. James P. S. 00:40, 22 December 2005 (UTC)
Extrapolation is not just another way of displaying historical data. Your plot would be much less troublesome if you restricted yourself to historical data. But you don't - you have a certain, very naive model (cubic interpolation) and use that to make predictions outside your data range. And if the model produces absurd data, you just fudge it by moving to a log domain. What is your justification for that? How do you know the forward extrapolation is not just as absurd and needs fudging? Indeed, it looks like by 2050 severe weather will create a lot more damage than the GPD. How can you have any trust in your model? You have tried polynomials of degree 1-3. What about polynomials of higher degrees? With a polynomial of degree 5, I can get a perfect fit! What about an exponential function (in fact, I guess by moving to a log domain you may have made a pseudo-exponential)? What about the sum of several periodic functions? You do not deal with any of these questions. You have no good reason for choosing this particular model. Having a decent fit on such a small data set does not validate a model. As Guettarda wrote, you need a real model, i.e. something that explains what's going on. In this case, you have a number of effects that are overlapping. Increased storm intensity and frequency is one. But it is currently believed that these are primatily due to medium term cyclic changes in the climate (so you would get one periodical term). A warming-related component may underly this - so you get one linear term (assuming a lot of linear relationships for which I have no justification - this is just for illustration). Normal economic growth is exponential, so you get a small exponential term. A lot of wealthy snowbirds building expensive property in hurricane areas...I have no idea about how to model this. And how do these effects combine? Multiplication seems reasonable. That gives you a fairly ugly function with a lot of variables...you can see that your data is insufficient to solve it. BTW, I did not suggest to "improve" the graph by going to the log domain. I pointed out that the previous result was obviously wrong. Reducing the level of obvious error is unlikely to provide a real fit, though.--Stephan Schulz 08:00, 22 December 2005 (UTC)
The log domain is a proper domain for nonnegative data, but the logistic sigmoid is probably better. Any function growing from zero with noise added to its strictly increasing function will tend to the lognormal distribution. I tried a lot of models to see which one had the greater R2, and I stopped at 0.98, even though I should have known to keep going because the underlying cause more closely fits the logistic sigmoid curve. You can click on the new image to get source code with which you can try your own models. I agree with your suggestion to try to fit a sigmoid. The only problem is that I'm not sure how to do confidence bands on a logistic sigmoid fit. Do you know how to do that? James P. S. 20:26, 22 December 2005 (UTC)
Sorry, I do not seem to make myself clear. The problem is that there is an infinite number of possible fitting function. There even is an infinite number of possible models that can be fitted with a better correlation factor than 0.98. In fact, there is an infinite number of models with correlation factor 1 - e.g. any polynomial with a degree greater than the number of data points. I can give you a fitting function with correlation factor 1 for any result you want to see in 2010. The chance that a randomly picked model is correct is zero. Your confidence interval is useless, because it works under the assumption that the model is correct and the data has a random, normal-distributed error. But we don't know if the model is correct. Finding a resonable model is hard and would be original research. --Stephan Schulz 23:25, 22 December 2005 (UTC)
Are there any reasons that the best known model of the underlying cause of added atmospheric energy would not be the most reasonable? The process taking place here is the review of past research. James P. S. 23:37, 22 December 2005 (UTC)

I still think we should just have the historical data, if we can't find predictions published in scientific journals or comparable. Rd232 talk 00:46, 24 December 2005 (UTC)

removed as original research. Vsmith 01:22, 24 December 2005 (UTC)
Again, an extrapolation is simply a way of displaying data; just as you might choose a different line style or axis scaling, the decision to place an extrapolation on a graph is a matter of presentation, not research. The original data is displayed unaltered. There has been no additional research performed. Many people, including myself, have reviewed the existing research and provided comments. The revisions to the graph have been based solely on those comments. If anyone were to claim that this research is "original" then they would be denying the value of the contributions above. This process of review is the means by which the most appropriate model for the extrapolation was found. James P. S. 02:56, 24 December 2005 (UTC)
this is still non-responsive. where did you acquire the data set from? or are we to understand that you have deduced the values from looking at the existing graph? there's no accuracy to the graph. the data set you refer to is in 'whole' billions of dollars, which is absurdely imprecise. furthermore, i can't believe i have to even bring this up - but how about putting up your first, second, and third revisions, rather than the first and third? because if you don't see that there's a problem here, then i'm floored. your first graph suggested costs exceeding 400 billion by 2025, your second graph was suggesting in excess of 4 trillion in costs - which you claimed was an even more accurate representation of the data - now your third graph more 'conservatively' suggests 350 billion. and yet the old maxim "lies, damned lies, and statistics" falls on deaf ears? good god. Anastrophe 03:47, 24 December 2005 (UTC)
The data set came from the graph; you can get a larger version by substituting "large" for "small" in the URL to check my digitization on my userpage. There are only two significant digits for all of the figures except the equation parameters (4 sig. digits) on the chart. There would be no significant change in the extrapolation or anything else if the data were more precise. The decadal averages, while smoothing out noise, more closely fit an steeper exponential curve, which is why I promised you that I would use the entire data set instead of just the averages, which I did. Note that the confidence interval bands are wide enough on both of the recent revisions that they overlap very substantially. The exponential fits grow much faster than the original cubic fit, but so do the error bands. I am very satisfied with the current revision and think it is probably the best that can be done without very detailed socioeconomic, political, and climate modeling combined.
please explain the discrepancy of more than an order of magnitude between your formerly accurate revision two, and your now more accurate revision three. and why we should believe that revision four and so on won't deviate as or more wildly. this is just guessing. it's meaningless. i predict that storm damage for 2013 will be seventy five cents. prove me wrong. Anastrophe 09:50, 24 December 2005 (UTC)
I never said that the second revision from the log domain was more accurate; only that it responded to three of the objections raised here. At that time I made it quite clear that I wanted to try the logistic sigmoid upon which the cause is based, and when I did, the simple exponential (which is closely related to the logistic sigmoid) was by far the better fit. The greater number of data points addresses questions about the significance far better than any use of the decadal averages ever could. James P. S. 19:59, 24 December 2005 (UTC)
James P. S. just said "I never said that the second revision from the log domain was more accurate". false. "The R2 is still 0.98, but more accurate and significant at the p=0.01 level. The prediction is now far more dire: $4 trilion in the 2020s. There are now 95% confidence interval bands as requested. James P. S. 09:05, 23 December 2005 (UTC)".
In fact I did say that, and I apologise for my mistake. I should have said only that it was more significant. There is no way to say whether one or another extrapolation is more accurate, except in retrospect. The third revision is far more significant than the other two because it has so many more data points for its two-parameter model. James P. S. 20:59, 24 December 2005 (UTC)
You are plucking numbers from thin air. each revision has been claimed to be 'better' than the previous one, while oscillating wildly in the conclusions suggested. why is this matter even being entertained? i found four other articles with this graph jammed into them. this is bordering on vandalism. Anastrophe 20:10, 24 December 2005 (UTC)
this is no longer bordering on vandalism, it is in fact vandalism. stop jamming your graph into a half dozen articles on wikipedia. Anastrophe 20:32, 24 December 2005 (UTC)
You are seriously accusing the addition of this graph as vandalism? I was going to ask a third opinion, but now I think it would be better to go to RfC on the question you raised as to whether the graph constitutes vandalism. Do you agree? James P. S. 20:59, 24 December 2005 (UTC)
And might I remind you that the data for 1999-2005 has not yet been incorporated into the extrapolation? I wonder what you think the 2005 data, likely to exceed the $100 billion mark at the top of the 95% confidence interval of the revised extrapolation, will do to the resulting curve? James P. S. 08:23, 24 December 2005 (UTC)
well, why aren't you adding that value to the graph? ...
I would if I could. Does anyone know where to find more recent data? James P. S. 19:59, 24 December 2005 (UTC)
...i mean 'likely to exceed $100B' is just as accurate as the existing extrapolations you're plucking from thin air, so go for it. since i think that storm damage in 2013 will be seventy five cents, please add that to your dataset, and see what it does to the curve. we'll all have a jolly old time pretending this means anything at all. Anastrophe 09:50, 24 December 2005 (UTC)
What odds are you willing to bet on your 2013 prediction?  ;-) I predict that you are not an actuary. James P. S. 19:59, 24 December 2005 (UTC)
I predict that by 2013, a new technology will have been devised, whereby solar-powered devices will extract CO2 from the atmosphere, converting it to harmless diamonds and O2. this will have a catastrophically negative effect upon diamond markets, for which I'll create graphs showing how this extreme correction to global warming is causing extreme unemployment for diamond merchants, and that we should create a special welfare system for them, along with diamond credits, and possibly legal action against those promulgating this new technology. prove me wrong with a new graph. Anastrophe 20:16, 24 December 2005 (UTC)
I hope you are right, but until such technology exists, it can not me measured, it does not cause any variation in the observed cost data, and so it can not be extrapolated. James P. S. 21:21, 24 December 2005 (UTC)
An extrapolation is not simply a way of displaying data. It is predicting beyond the data. A prediction like this is either original research or a simple pipe-dream, unless it has been published in a peer reviewed source it is out. Out from this article and any article. So, where has this extrapolation been published? Vsmith 04:16, 24 December 2005 (UTC)
I disagree. Extrapolation bars and curves are commonly referred to as "chart decorations" by graphic designers. They do suggest a prediction, which is what they are supposed to do, but they strictly reflect only the underlying data upon which they are based. The practice of printing extrapolations with historical data is commonplace in news, almanac, encyclopedia, and many other kinds of publishing, and is never considered original research. The graph is no more original research than if I were to reflect in the text of the article that, for example, "If present trends continue, the cost of extreme weather events could exceed 350 billion inflation-adjusted U.S. dollars per year by 2025." Don't you agree that is better said with a graph which displays the historical statement directly than with that sentence? James P. S. 08:23, 24 December 2005 (UTC)
Chart decoration do add more oomph - seems more official or eyecandy or whatever. Still falls under the category: How to lie with statistics. Vsmith 14:46, 24 December 2005 (UTC)
your last graph suggested that if present trends continue, the cost of extreme weather events could exceed one trillion inflation-adjusted U.S. dollars per year by 2025. why the 650 billion dollar discrepancy? a billion here, a billion there, who cares? Anastrophe 09:50, 24 December 2005 (UTC)
Please observe the 95% confidence intervals. Both graphs fall well within each other's confidence interval. James P. S. 19:59, 24 December 2005 (UTC)

No substantial periodic component

How do you get from the relationship "increased atmospheric energy -> more extreme weather events -> higher costs" to your interpolation? But there are a couple of other problems. The rising cost of weather-related events has a number of reasons. More extreme weather events are one reason. But many of those are Atlantic hurricanes (which also tend to be very expensive), and the current belief is that this is primarily due to a cyclic change in hurricane frequency. It's very likely that an increase in atmospheric energy will further affect this, but there is no reasonable model available yet. And the main reason for the increase in cost is that many American retirees move to Florida and build expensive buildings in hurricane-prone areas. So damages would rise very much even if hurricanes stay exactly the same. --Stephan Schulz 23:50, 22 December 2005 (UTC)

Do you have any evidence that hurricane frequency is cyclic? There is plenty of evidence that increased storm strength is occuring on-shore, too. Do you have any evidence that these events are not related to atmospheric energy increases? I'm not arguing that increased building has a lot to do with the increased cost. Is there any reason not to extrapolate that increased building? James P. S. 00:08, 23 December 2005 (UTC)
See e.g. here. All things being equal, putting more energy into the atmosphere will lead to stronger storms, of course. But we do not yet know if all things are equal enough. E.g. we expect the polar regions to warm up faster. So we may be reducing the temperature gradient that drives (some) storms. I doubt it, but, as far as I know, this is one of the open questions. Another is if the recent increase in hurricane activity is already bolstered by global warming. I don't say this is not so, I do say this is a thing we don't yet know. --Stephan Schulz 01:02, 23 December 2005 (UTC)
Thanks for the link. I didn't see anything supporting cyclicity of hurricanes apart from a recent upswing noted near the word "cycle," but I did note the following excerpts:
scientists believe that global warming will result in more intense hurricanes, as increasing sea surface temperatures provide energy for storm intensification. An MIT study published recently in Nature provides the first data analysis indicating that tropical storms are indeed becoming more powerful over time.... Higher ocean temperatures may also influence the tracks of hurricanes, increasing the likelihood of hurricanes tracking through the Caribbean or making landfall on the U.S. east coast.... A scientific analysis indicates human-induced climate change likely increased the severity of the 2003 European heat wave that killed about thousands of people. The same study predicts that as the climate change progresses, similar heat events will become normal rather than exceptional....
James P. S. 04:38, 23 December 2005 (UTC)
I was refering to Although the average number of hurricanes between 1995 and 2005 is probably unprecedented, we have not seen a long-term increase in hurricane frequency during the 20th century overall. Instead, we have seen periods of high hurricane activity that last for several decades, followed by decades of low activity. The 1920s-30s and 1950s-60s were active periods. In 1995 we entered and are currently in the latest natural phase of high hurricane frequency, which is expected to persist for another decade or two. Note also how this would skew your curve. --Stephan Schulz 07:31, 23 December 2005 (UTC)
I don't see the cycles. I'd love to see the plot of your own fourier analysis. Let me know if you get any strong periodicities. There is certainly a long-term trend. There is a wealth of additional historical data, but note section (13) about periodicities is not very consistent with other descriptions of periodicities. I doubt that there is a substantial periodic component. James P. S. 09:05, 23 December 2005 (UTC)
Well, given that the period seems to be about 50-60 years (and rather irregular), what do you expect to see in a graph from 1950 to today? You may notice the drop in between the last maximum and the current one. "Table 7" is exacly what you expect in this situation (and it has other problems, in that tropical storm detection only became reliable with extensive satellite coverage. --Stephan Schulz 09:21, 23 December 2005 (UTC)

Significance level of early revisions

(replying to initial comment by Dragons flight 21:14, 19 December 2005 (UTC) above)

What R2 value would you consider reasonable? Note that there are five data points and only four degrees of freedom. The strength of the extrapolation based upon the R2=0.98 is therefore very significant. The only original research here is asking R whether the first, second, or third-degree polynomial had the greater R2 from the data. Moreover, there is no question that future data shown must be extrapolation (no time travel allowed.) James P. S. 20:51, 19 December 2005 (UTC)

Actually, R2=0.98 on your remaining degree of freedom is only about 90% significant (i.e. not even two sigma), but that is besides the point.
Would you mind showing your math on that, please? James P. S. 21:52, 19 December 2005 (UTC)
The t statistic = r*Sqrt((Neffective)/(1-r^2)) will approximate a Student's distribution when r is the Pearson correlation statistic such as you have used. In ordinary correlation analysis Neffective = Nmeasured - 2 as neither the mean nor magnitude of the slope are significant in computing r. As you have fit to a cubic polynomial, you have also removed the significance of the next two curvature terms, giving Neffective = Nmeasured - 4 = 1. For r2 = 0.98 => t ~= 7. From a table for two-tailed Student distribution with N = 1, one can find that this gives p < 0.09 or 91% significance. Now there are some fudges in this. For one I haven't checked exactly how well the approximation to a Student distribution holds as N -> 1, though I do know at N ~= 10 it is more than adequate. Also, the derivation of the distribution assumes unrelated data not a comparison between actual data and values fit to that data.
As a check on this result, I just now performed a Monte Carlo analysis wherein I generated 5 points at random, fit a cubic polynomial to them, and then measured the correlation values between the fit points and the random data which had been fit. This technique showed a probability of obtaining r2 >= 0.98 as p ~= 0.18, or 82% significance. It would appear therefore that analysis based on the presumed distribution discussed above actually somewhat overestimated the significance of your result.
Let me get this straight: You are saying that five "random" points (equally spaced apart on the x-axis?) have a 18% chance of more than 98% of their variation explained by a third degree polynomial? What was the range on the y-axis? 1/5 of the x-axis spacing? James P. S. 05:46, 20 December 2005 (UTC)
Yes. I selected a number from 0 to 1 at random in the y direction for each point, which were placed at 1 through 5 as you suggest. Dragons flight 06:43, 20 December 2005 (UTC)
Basically the point here is that it is difficult to derive a statistically significant result when your fitting procedure consumes most of your data sample. Dragons flight 23:36, 19 December 2005 (UTC)
You are obviously an accomplished statistician and graph-maker. Why don't you draw an extrapolation with confidence intervals for this time series? James P. S. 05:46, 20 December 2005 (UTC)

One more comment - you can't select curves on the basis of R2's - they aren't reliable for model selection. You need to look at something else, like AIC. And, again - if you are extrapolating you need confidence intervals and you need an underlying model. Using simple curve-fitting to extrapolate so far beyond the data is not valid inference. Guettarda 23:31, 19 December 2005 (UTC)

The latest revision, based on the formula y=a*(x/b), was the best fit by far out of 3,665 equations as measured by both DF-adjusted R2, TableCurve's F-statistic (very similar to AIC) and total residual error. James P. S. 19:59, 24 December 2005 (UTC)

Complete set of regression models

Entering onto the research itself: you have, apparently, fitted a 3rd order poly. What about 2nd? or 4th? or exp? I'm saying that as a point of interest, but (sorry) I'm going to object however well you answer the last point.... William M. Connolley 22:44, 19 December 2005 (UTC)

I most certainly did try several extrapolations and chose the one with the greatest R2. James P. S. 05:46, 20 December 2005 (UTC)
The third revision tried over 3,660 models with TableCurve. They have a 30-day free eval download for Windows if anyone else would like to propose an alternative. James P. S. 20:51, 24 December 2005 (UTC)

Nonnegative data

Fitting a curve to a number of points and looking at the correlation coefficient is fine if you have a good model of what goes on. It can also work if you have a number of candidate models and a large sample. But in this case, you don't have a reasonable model, and you have just 5 data points. Using your extrapolation backwards in time, we will find that in 1900, the world earned massive amounts of money from hurricances...obviously, there is something wrong with the model. --Stephan Schulz 23:10, 19 December 2005 (UTC)

Of course the correct curve would be in the log domain for nonnegative data. I used the log1p transform function in R without any noticable change in the resulting fit. James P. S. 05:46, 20 December 2005 (UTC)
The second revision used a second-degree polynomial fit in the log domain, the underlying cause more closely fits a logistic sigmoid, and the best fit by far to the yearly data in the third revision is an exponential, all of which are nonnegative. James P. S. 19:59, 24 December 2005 (UTC)

Context of the IPCC source data

To offer you some hope: you've sourced the data from an IPCC graphic. If you can find where that graphic appears in the IPCC report and show how *they* have used it, then you would be on firmer ground. But it would still probably belong in effects of GW. William M. Connolley 22:44, 19 December 2005 (UTC)

I would be happy to just add that historical graphic, if it were at all legible at thumbnail size. They used it as you might expect from the artwork in the background of the slide -- it's a warning about the increase. James P. S. 05:46, 20 December 2005 (UTC)
There's an interesting graph on the IPCC website (from IPCC:TAR, climate change 2001 synthesis report, chapter 8), that is pretty similar to what James (Nrcprm2026) has used as the source data, but it also includes a trend line (two, one for financial losses, one for insured losses. more importantly, it shows how many of these losses are weather related, and the relationship between the now, and the then of each. I think it includes '99 data too.
perhaps most importantly for James, it also show the rate of change over the last 50 years, which is obviously (from the graph), increasing.
I suggest that we upload either that graph or a similar one, and forget the debated graph until better, more accurate, and more up-to-date data is found? --naught101 08:59, 4 January 2006 (UTC)
that's definitely the graph to go with. not original research, published widely, sourceable, and doesn't make wild, unsupportable predictions about the future. current data is cautionary enough. Anastrophe 18:05, 4 January 2006 (UTC)
No way! It doesn't have current data, which would be nearly three times the height of the graph and would completely change the trend line. Nor does it have confidence intervals. I am pretty sure that I will soon be able to procure more current data from the same people who collected the 1950-1998 data. —James S. 18:49, 4 January 2006 (UTC)
Errrmmm... if adding more data to that graph would totally change the trend line, then you have a problem, because it makes the trend line very dependent on the exact period you select. Which makes the extrapolation very suspect. William M. Connolley 19:25, 4 January 2006 (UTC).
that's precisely the reason i've challenged Nrcprm2026 to produce a graph using only the data from 1994 to 1998. if we pretend no reliable info exists before then, and use the remaining data set he's extrapolating from, the trend will be remarkably different. but this is all academic really. the number of times now that the essentially fatal flaws in this extrapolation have been pointed out borders on the laughable. original research. ignores economic and sociological factors that are far more responsible for the nature of the curve (hundreds of expensive three star hotels on the florida beach are far more costly to build, and concommittantly to repair or replace, than the small hotels and motels that used to populate the florida cost a mere forty years ago). ignores massive population growth along florida and gulf coasts. and that's just the U.S.. But this "No way!" response really sums it up for me. it's not that the graph itself is important, it's that Nrcprm2026 wants his graph up, or no graph at all. before the estimates (stress estimates) of 2005 costs were noted in the press, Nrcprm2026 was delighted to stick with just the current data. Now, he's insisting on adding estimates - which he refers to as 'current data' - to his dataset, which is not just bad form, it's bad science. how many dozens of screens of discussion have been rendered, all to entertain this? one person relentlessly pushing his original research, a dozen others pointing out the manifold problems with that research. absurd. Anastrophe 23:38, 4 January 2006 (UTC)

Veracity

that graph should be the poster child for 'lies, damned lies, and statistics'. Anastrophe 23:44, 19 December 2005 (UTC)

Oh come on. I've seen far worse. I propose that we get the historical data up to the present (including 2005) and see if the the linear, 2nd order, and 3rd order polynomial fits don't all more than double their prediction for 2050. If I'm right will you retract that comment, Anastrophe? James P. S. 05:46, 20 December 2005 (UTC)
most certainly not. that graph is absolutely _absurd_, and i use the term advisedly. and "I've seen far worse" is certainly the faintest defense i've heard in a long time. as it stands, i happen to reject most of these 'predictions' for what they are: fiction. there's no science in fictional representations of the future. until such time as we have time-machines, of course.
actually, let me modify "most certainly not". i'll be happy to discuss your 'predictions' in 2050, when they can be scrutinized against factual data. Anastrophe 06:36, 20 December 2005 (UTC)
Quit giving the new guy a hard time, he means well. Dragons flight 06:43, 20 December 2005 (UTC)
i thought we were interested in science. i stand corrected.Anastrophe 07:09, 20 December 2005 (UTC)
Where does "interest in science" conflict with reasonably friendly behaviour against others? Wikipedia is not only a collection (as such it would be quickly vandalized into uselessness), it's a community. BTW, your shift key is broken.--Stephan Schulz 07:35, 20 December 2005 (UTC)
how does a snipe like 'your shift key is broken' enhance that sense of commaradarie? i've seen discussion pages, edit summaries, and _articles_ with far worse than a preference for lower case being their defect. when editing articles, i conform to the queen's english, so to speak. elsewhere, why should it matter, but as grist for a jab such as yours? i would quibble with your contention that wp is *not* useless with the degree of vandalism it experiences, but that's another discussion entirely. the graph presented is preposterous. the projections have no basis in fact, or science. it's amazing this discussion is even taking place. 08:15, 20 December 2005 (UTC)
Well, you're not a newbie, so you are hopefully more robust ;-). Anyways, given that you write to communicate, and that you write every sentence once, while we read it (presumably) many times, it makes sense to spend some time making your writing as legible as possible. And that means following the conventions of the language as far as possible. Your writing style makes your text harder to read. As such, it wastes my time. I apologize if my "jab" was not friendly enough. --Stephan Schulz 08:24, 20 December 2005 (UTC)
"Your writing style makes your text harder to read. As such, it wastes my time." mmm. your disabilities aren't my responsibility. you must have a terrible time with handwritten content. oh well. again, there's no science on display in the graph. i'll be exceptionally amused if it winds up in the article. have you seen this article? http://www.sciam.com/article.cfm?chanID=sa003&articleID=000525AD-1223-1354-922383414B7F0000 it's equally amusing. this 'researcher' ran his climate modelling supercomputed program out to 2100, then declared - i kid you not, it's in the article -"Climate change is going to be even more dramatic than we previously thought". how amazing is that! he can actually predict the future!. i wrote to sciam, but heard nothing back. no retraction. that is incredibly irresponsible, but it really just points to this mentality that has become pervasive - 'the world is going to be destroyed by global warming, just you watch'. ah well. ranting again. good night.Anastrophe 10:06, 20 December 2005 (UTC)
Indeed. I hate reading most handwritten texts. As a consequence, I don't usually read them (hint!). As you may have noticed, we agree about the scientific value of the diagram. Still no reason to call the author a liar - in contrast to some other contributors he is willing to discuss and learn. As for the SciAm article: That seems to be fairly solid science, although it's hard to judge from a pop science article. How you can compare a complex, detailed climate model that has been validated on known data with the 5-plot curve on data that is mostly affected by social and economic developments is beyond me. Just remember the next time you stumble: There is no reason to brace yourself for the fall - that's just useless prediction of the future.--Stephan Schulz 13:32, 20 December 2005 (UTC)
please show precisely where i "call[ed] the author a liar". word for word please. i generally have a strong disdain for having words put in my mouth, then being attacked for them. i think the author is fantasizing. that's different from lying....regarding your metaphor, we're falling? interesting assumption. the dire predictions we are hammered with are based *no discernable harm that has yet occurred*. regarding the sciam article, had the researcher said "may be" rather than "is going to be" he'd have some credibility. Anastrophe 17:43, 20 December 2005 (UTC)

(back left) "that graph should be the poster child for 'lies, damned lies, and statistics'. Anastrophe 23:44, 19 December 2005 (UTC)". To me (and apparently to others) that statement implies that the creator is, in your eyes, a liar. If you did not intend to communicate this opinion, maybe you should be more careful in what you write. It is only slightly harder than using proper capitalization. It also helps all of us to waste less time.--Stephan Schulz 23:21, 20 December 2005 (UTC)

As the author in question, I believe you clearly insinuated that even if, as I claim, using data up to and including 2005 would double the 2050 predictions for the 1st, 2nd, and 3rd order polynomial extrapolations (note that the source data for that chart ends at 1999) you would still consider that graph "the poster child for lies, damned lies, and statistics." Therefore, since you have stated that you would believe that even if adding current data in the extrapolation causes it to increase, then can you give me any reason that you should not be considered the poster child for POV deception through failed rhetoric? James P. S. 19:43, 20 December 2005 (UTC)

good effort there with the attempted rhetorical smokescreen! the burden of proof is on you to prove your extrapolations far into the future have meaning. hey, i have fun with statistics sometimes too - we should have all died in a nuclear holocaust long ago if statistical predictions had held true. your graph is simply more of the same fear-mongering - a horrible horrible fate awaits us due to global warming. is it any surprise the general public, brought up in an era of intense skepticism, challenges the honesty of those promulgating the global warming theory, when they resort to dire predictions, based on guesswork, rather than presenting FACTS? are you not content with the FACT that the economic cost of extreme weather has significantly increased in the last decade? isn't that enough? why make things up about an unknown future to try to convince people? Anastrophe 20:02, 20 December 2005 (UTC)
I propose that Anastrophe and I take this up on each other's talk pages until, I hope, we reach a resolution to be reported here. James P. S. 20:23, 20 December 2005 (UTC)

Inflation adjustment

The revised image has an additional degree of freedom because it is in the nonnegative log domain. The R2 is still 0.98, but more accurate and significant at the p=0.01 level. The prediction is now far more dire: $4 trilion in the 2020s. There are now 95% confidence interval bands as requested. James P. S. 09:05, 23 December 2005 (UTC)

since this graph is predicated on inflation-adjusted costs, please explain to me how you are adjusting for inflation in the future. what year's constant dollars are you calculating against? what projected future rates are you adjusting downward for? or are your proposed costs against constant dollars today? Anastrophe 04:40, 21 December 2005 (UTC)
The source data is in constant (real) U.S. dollars, so the assumption is that inflation from 2000-2020 will be the same on average as it was in 1950-1999. James P. S. 06:32, 21 December 2005 (UTC)
the "average" inflation from 1950-1999??? you do realize that the lower the granularity of your inflation adjustment, the greater the deviation from reality? inflation from 1950-59 averaged .69%; inflation from 1970-79 averaged 11.35%. however, inflation in 1950 was 1.09% - in 1951, it was 7.88%. you do realize that projections for dollar-costs *in the future* explicitly must adjust *downward* against today's constant dollar, unless you are proposing recession for the next twenty years? i suspect your chart exaggerates the costs massively, due to this probable error. Anastrophe 10:00, 21 December 2005 (UTC)
i would still like to know how you are adjusting your values for future inflation. because i suspect you're biasing upward rather than downward for those values. also, what is your *source* data? or are we to understand that you are simply looking at the existing IPCC graph and drawing your lines from that, rather than actual raw data? furthermore, you state that your source for the second graph is exactly the same as the source for your first graph. why are you excluding the more recent available data, which was one of the (many) concerns that have been expressed? Anastrophe 19:30, 21 December 2005 (UTC)
i just looked at the script listed with the graph as the source code. where is the inflation adjustment? unless i'm mistaken, there is none in your code. Anastrophe 19:35, 21 December 2005 (UTC)
you repeatedly refer to the 'source code', but not the source data. the only source data you've referenced is an existing graph, not raw data. nor have you addressed the significant problems that your graph appears to have by extrapolating out a 50 average of inflation, rather than a more appropriate granularity of annual averages. plus, the logical issue of extrapolating past rates of inflation to future values against constant dollars now - which suggests that you are calculating inflation backwards, and artificially inflating the resulting values by double (double the extremely course value of a 50 year average). the formula you use to create the graph is one thing. the actual raw data is entirely another. again, what values are you using to project future inflation rates against? Anastrophe 21:50, 22 December 2005 (UTC)
I have the source data as well as the decadal averages from the IPCC chart. I promise I will use them as soon as I figure out how to extrapolate logistic or gompertz sigmoids with confidence intervals. As for the inflation adjustment, why are you objecting to using the historical inflation adjustment (real dollars) built in to the source data? Do you have any reason to believe that inflation will in the next half-century be significantly more or less what it was in the past half century? James P. S. 22:39, 22 December 2005 (UTC)
i'm not an economist. but i believe that if you use constant dollars based upon today, that means that values for the past are displayed higher relative to today's dollars, and values in the future *are decreased relative to todays dollars*. past inflation means 'older dollars' were worth more than today's dollars, so they are represented on the graph at a higher cost than they cost in actual dollars at the time. against today's dollar, inflation for the future means that 'future dollars' will be worth less than today's dollars. so whatever formula you use for calculating the past inflation must reverse trend the moment you begin talking about the future.
for example, at a constant 2005 dollar value, $100 billion in extreme weather costs in 2004 would be graphed as about $102.6 billion. if you project $100 billion in extreme weather costs in 2006, it will be graphed as about $97.4 billion, if you assume approximately the same rates of inflation going into the future. i suspect your future trends are being calculated as if they were past trends, and displaying dollar values rising into the future, rather than falling. Anastrophe 00:10, 23 December 2005 (UTC)
If the source data was not adjusted for inflation, then the extrapolation would need to be adjusted downward. However, the source data is inflation-adjusted; the IPCC has already applied an adjustment index to their figures. Your example makes no sense to me because if the actual cost were $100B in 2004 and 2005, then the inflation-adjusted real cost would be sloping up from $97.4B in 2004 to $100B in 2005, leading to an extrapolation for 2006 of $102.4B. My understanding is that monitary extrapolation needs to be done in real terms after adjusting for inflation unless the extrapolation includes an exponential term to model the inflationary component of the variation. James P. S. 00:28, 23 December 2005 (UTC)
you're right, i was making it much more complicated than it is. i still think your extrapolations are essentially meaningless however. also, i'd like a pointer to the source DATA, not a graph. i can't speak for anyone else, but that graph doesn't show me any precise values. for example, what value did you assign the very first bar of the graph at 1950? is that $9.5B? $9.7B? $9.726B?
I'm using the detailed yearly now, from the chart. The data set from the graph is on my userpage. James P. S. 00:18, 24 December 2005 (UTC)

Does the graph imply any causes?

I agree that the graphic is inappropriate. Mostly because it (falsely) implies that most of that cost increase is due to weather changes: they aren't. They are very largely due to societal changes, which aren't removed by the inflation adjusting. William M. Connolley 09:41, 21 December 2005 (UTC).

How does it imply anything about the cause of the changes? I strongly disagree that there is any such implication. Moreover, there are plenty of sources in the Effects article showing that storm strength is in fact increasing, independent of societal changes. No matter whether either and/or some other cause(s) are to blame, the fact remains that the cost is increasing rapidly.
This objection is addressed in the new caption. "Some of the increase is due to greater exposure such as building on the coast." James P. S. 09:05, 23 December 2005 (UTC)

My principle objection to this figure is that it, falsely, implies that most of the increase in cost is due to increased storminess.

Why do you believe that the graph implies anything about the causes or the relative proportions of the causes? James P. S. 19:59, 24 December 2005 (UTC)

There is really no good evidence for this. There is evidence to the contrary. Most of the increase in damage is due to societal changes: more people living near coasts. I don't agree with RP on everyting, but [1] is one source. Also RC: [2]. Or Changes globally in tropical and extra-tropical storm intensity and frequency are dominated by inter-decadal to multi-decadal variations, with no significant trends evident over the 20th century. Conflicting analyses make it difficult to draw definitive conclusions about changes in storm activity, especially in the extra-tropics. No systematic changes in the frequency of tornadoes, thunder days, or hail events are evident in the limited areas analysed. from the IPCC [3]. William M. Connolley 09:13, 24 December 2005 (UTC).

From the source cited in the periodicity discussion above:
scientists believe that global warming will result in more intense hurricanes, as increasing sea surface temperatures provide energy for storm intensification. An MIT study published recently in Nature provides the first data analysis indicating that tropical storms are indeed becoming more powerful over time.... Higher ocean temperatures may also influence the tracks of hurricanes, increasing the likelihood of hurricanes tracking through the Caribbean or making landfall on the U.S. east coast.... A scientific analysis indicates human-induced climate change likely increased the severity of the 2003 European heat wave that killed about thousands of people. The same study predicts that as the climate change progresses, similar heat events will become normal rather than exceptional....
Please see also this evidence of a long-term trend.
Again, I'm not arguing that increased building has a lot to do with the increased cost. Is there any reason not to extrapolate that increased building trend? James P. S. 21:47, 24 December 2005 (UTC)

Don't add or remove until comments from 3rd party experts

Please please don't add this graph in elsewhere, e.g. to Mitigation of global warming or Effects of global warming until we've sorted out the objections here (though in my view you're unlikely to get them sorted out). I know you're fond of the graph but you'll just annoy people if we have to follow you around removing it. William M. Connolley 09:51, 24 December 2005 (UTC).

Okay, I will do my best to go get some comments on the graph from at least one reputable climate blog such at RealClimate. In return I would ask that you not remove it from any of the pages where it had been added shortly after the creation of the third revision, and still is now. I will copy the comments from third-party experts back here when I have received them. James P. S. 19:59, 24 December 2005 (UTC)
I have sent multiple requests to climate blog editors as follows:
Dear _____ blog editor(s):
_____ has been nominated in a search for independent third parties to comment on this graph.
This request is being made because of a dispute arising in another forum. Your participation is strictly voluntary. Would you please comment on this graph, and/or post it to your blog where others might comment on it?
Specific questions central to the dispute are:
1. Is the extrapolation reasonable?
2. Does the graph imply anything about the causes of the variation shown?
3. Does the graph have a neutral point of view?
4. Is the graph as a whole merely a review and extrapolation of the source data, or is the graph original research?
When you have at least one comment, please reply by email....
Thank you.
Sincerely,
James Salsman
I await replies and will post them here when I receive them. James P. S. 23:21, 24 December 2005 (UTC)

Gavin from RealClimate was the first to reply:

Date: Sat, 24 Dec 2005 21:31:58 -0500 (EST)
Subject: Re: please comment on graph to help settle dispute
From: contrib@realclimate.org
Thanks for thinking of us. As I prepare to adjudicate, please note that since I don't really know what the dispute is, my ruling may not exactly serve your purpose.... but here goes anyway.
Costs of weather-related damages are indeed soaring, and if you had added in data up to 2005, you would have found the rises continuing as extrapolated in the graph. However, the *cause* of the increasing costs is not evident from the graph. Analyses for purely US damages indicated that the vast majority of the increase is due to greater development in areas that are vulnerable (coasts, Florida, flood plains, etc.), rather than an increase in storm activity or flood frequency. Data from more widespread sources indicates that most of the rise there too is development related, though there is a possibility of a climate effect. Given that coastal development is unlikely to stop any time soon, it is a pretty conservative assumption to guess that the rise in damages will continue. Estimates of the climate change component of that will likely increase in the tropics, but will likely always be a smaller component than the development change. So in that sense, the graph *itself* has a neutral POV. But you would need to examine the context in which it is used to see whether that package is similarly neutral. I'm not sure that using an Excel curve fitting package really counts as original research (technically I guess it is, but it wouldn't be sufficient for a research article).
Hope that statisfies....
Gavin

Given that, I am inclined to change "Some" to "Most" in the caption, as William M. Connolley has requested. --James S. 18:36, 25 December 2005 (UTC)

"Given that coastal development is unlikely to stop any time soon". at least for the US, this assumption is not necessarily supportable. and it's an incomplete assumption as well - development may continue, but government mandates, or simply the will of the marketplace, may result in significantly reduced costs due to implementation of extreme-weather resistant building methods and architecture. The government may cease providing absurdely inexpensive hurricane/flooding insurance to those building in coastal areas. or as in louisiana, people may simply decide that it's not worth it to live at or below sea level in hurricane alley. assumptions about the future have a way of biting you in the ass. Anastrophe 19:06, 25 December 2005 (UTC)
Since such government mandates, marketplace will, and better building methods don't exist yet, they can't be considered in the extrapolation. As for whether the government will stop providing flood insuance subsidies, that hasn't happened yet, and doesn't seem likely.
The part about the pre-1999 data accurately predicting the 1999-2005 period is important. I wish I could find some actual numeric data for at least 1999-2003. Anyone? --James S. 18:17, 26 December 2005 (UTC)

I am expecting two more replies. --James S. 18:36, 25 December 2005 (UTC)

I also posted the question here --James S. 18:39, 28 December 2005 (UTC)

Given that I'm a member of RC too, do I get to be a third party expert as well? If so, I second my own opinion, as well as Gavins. William M. Connolley 20:46, 28 December 2005 (UTC).

Economic projections

The cost of extreme weather is rising rapidly and could reach 350 billion 2001 U.S. dollars per year by 2025. source data: IPCC, 2001. Most of the cost increase is due to added exposure such as building on the coast.

Once again, I have removed the graph — see right. (I notice it's been added to many more articles). Where are the "two more replies"? I don't yet see the consensus here for extrapolating the graph out to 2025. Furthermore, I'm confused as to why the image caption says "the cost of extreme weather is rising rapidly and could reach 350 billion 2001 U.S. dollars...", while when I go to Image:Extreme-weather-cost.gif, it says 250 billion. I think in order to include such a figure, it needs a reputable source for that specific figure. (See Wikipedia Cite sources guidelines.) From looking at external sources such as IPCC and NAS, I don't see any of them trying to extrapolate costs of extreme weather events to so many years out. They don't because there are so many complicating socio-economic factors, that any projected figure is just a guess at best. In economics, projections rarely extend beyond ten years. --Aude 04:30, 29 December 2005 (UTC)

As above, a nominated expert said that the extrapolation was reasonable, doesn't imply causes, has a NPOV, and only borderline OR, and Wm. Connolley, who had raised objections concerning implications of causes, "seconded" that expert's comments. I'm still waiting for replies from the other two I emailed. They must be on vacation.
The 250/350 discrepancy is intentional, and has to do with the vast width of the 95% prediction confidence interval, and the fact that the data for this year, 2005, is going to be at least $20 billion above the existing extrapolation's confidence interval for this year.
Do you have a source for the claim that projections rarely extend beyond ten years? I have U.S. goveernment graphs for health care spending which project to 2075. —James S. 06:33, 29 December 2005 (UTC)
My engineering professors would have flunked me for producing such a wild extrapolation based on the data shown (and one spent much of a semester teaching us all about graphs). Steven Schultz is right - the data simply do not support an exponential rise as being better than higher-order polynomial fits, nor is such a rise intuitive. Finally, the consensus I see is that this graph is original research, and bad research at that. I'm removing it from Effects of global warming. (BTW - I expect huge impacts from Global Warming - just not this one.) Simesa 10:12, 29 December 2005 (UTC)

Still don't like the graph

This graph talks about the cost of extreme weather events from within articles about global warming. There is a clear implication in that that there is some connection! And yet, there isn't: the bulk of the increase in damage comes from societal changes, not from weather changes. Given this alone, the graph should *not* be in any of the climate change type articles. If there is a societal changes article, it might belong there. William M. Connolley 17:15, 29 December 2005 (UTC).

Please stop putting the graph in every single article that's related to the topic of global warming. Wikipedia is written in a Summary Style and each article (e.g. Fossil fuel, Mitigation of global warming) needs to focus just on that specific topic and not tangentially related topics. --Aude 17:39, 29 December 2005 (UTC)
it should probably be pulled from energy development too. Anastrophe 17:52, 29 December 2005 (UTC)

Who believes that increased storm strength due to radiative forcing is not caused by fossil fuel? I am replacing the 2nd revision of the graph, based on the implication that some or all of you actually believes that there is no connection. It is obvious, well documented, emperically verified, and proven in dozens of venues. If there is any remaining evidence against, please post it here before deleting the work of those of us who, with the vast majority, understand radiative forcing.

All of what you say above is wrong. The link is not at all strong: all the evidence is the opposite: that the bulk of the damage increase is due to societal changes, not due to weather changes. Please stop re-adding this graph: its obvious now that opinion is against you. You can't just force it in against opposition. William M. Connolley 22:25, 29 December 2005 (UTC).
Truth is not a democratic process.
Tedious platitudes will get you nowhere. Editing wiki is a matter of consensus. The consensus is (obviously) against you and your graph. Before you annoy people so badly that you are tainted forever with it, pause and consider. William M. Connolley 18:22, 30 December 2005 (UTC)
That the bulk of the damages are due to building is stated in the captions, as you requested. You know very well that storm strength has been increasing, if you have read the ample evidence cited above. You seconded the statement that the extrapolation is reasonable. [Oh no I haven't - William M. Connolley 18:22, 30 December 2005 (UTC)] Are you so sure that the graph should not be included that you are willing to submit to arbitration by the board? I will continue to replace the graph until you are able to convince me otherwise by reason. Otherwise, if you are so sure that the graph is bad, then prepare to arbitrate. —James S. 01:32, 30 December 2005 (UTC)
Placing the graph within the GW pages implies, very strongly, a (incorrect) relation to climate change. Else, why is the graph in the GW articles at all? In my opinion, there is no good reason to include it. William M. Connolley 18:22, 30 December 2005 (UTC).
The graph isn't solely about increased storm strength (or reasons for increased storm strength), as you suggest in your above comments. The issue with the graph is about making unsubstantiated extrapolations, without citing the projected figures from reputable sources. If someone finds (I haven't found it yet) projections of costs of extreme weather in a reputable scholarly journal, by NAS, IPCC or other reputable source, then we can include it. We need to stick with Wikipedia guidelines and policies of NPOV, Cite sources, and No original research. --Aude 22:29, 29 December 2005 (UTC)
The captions of all three and the two latest revisions of the graphs clearly cite sources. The extrapolation has been considered reasonable by the independent third-party expert suggested by William M. Connolley, who seconded their statement. The 95% confidence interval bands are weak, as is the case you are trying to make. Do you have any reasons that you believe the graph does not have a NPOV? The case for OR is borderline at best. If you feel that your arguments are actually cogent, then please prepare a case for arbitration. —James S. 01:32, 30 December 2005 (UTC)
You cite the source IPCC data, but not the extrapolation or projected costs. --Aude 01:58, 30 December 2005 (UTC)

Morover, if you feel compelled to delete the graph, please replace it with one of your own or anyone else's design which you feel more properly represents the most accurate extrapolation. —James S. 22:06, 29 December 2005 (UTC)

Why? William M. Connolley 22:25, 29 December 2005 (UTC).
Good question. Have you created any extrapolation graphs? —James S. 01:32, 30 December 2005 (UTC)

Vote on Image:Cost-of-storms-by-decade.gif

Let's just settle this, once and for all, rather then an ongoing revert war. Let's vote on whether or not to include the graph in the Global warming article. --Aude 17:43, 29 December 2005 (UTC)

Why have you called the vote on the initial revision? I trust that you will have the courtesy to leave the graphs in place while the survey and related processes are taking place; will you? —James S. 01:32, 30 December 2005 (UTC)

This vote also extends to global warming-related articles including Effects of global warming, Fossil fuel, Mitigation of global warming, and others. --Aude 17:47, 29 December 2005 (UTC)

As far as I can tell, current surveys and RfCs are required on each individual article's page. You can not expect a survey for one article to take place on another's talk page. —James S. 01:32, 30 December 2005 (UTC)

Include Image:Cost-of-storms-by-decade.gif

Do not include Image:Cost-of-storms-by-decade.gif

  • Anastrophe 17:51, 29 December 2005 (UTC) - dataset incomplete, ends 1998. more recent data implicitly available. on that basis alone, posting of graph is misleading, since it "predicts" data where actual data should be available.
I agree that more recent data would be great. I believe I have found the source of the IPCC data from an old draft PDF! —James S. 04:07, 30 December 2005 (UTC)
  • bikeable (talk) 22:29, 29 December 2005 (UTC) I'm new to this conversation, but have been watching the revert war, and the graph is troublesome. If it were identical to a graph in a peer-reviewed journal, I would consider it acceptable. As is, given how far out it extrapolates, and the question of which regression model is appropriate for this data, it looks to me like original research. (After all, you could get a much higher R2 by using an n-ordered polynomial for n data points, but that doesn't mean that the curve would be meaningful.)
If the graph were identical to something in a journal, it would be a comyright violation. Do you understand the meaning of the 95% prediction confidence interval? —James S. 01:34, 30 December 2005 (UTC)
If the graph were cited from a journal, it might be fair use; or data in a journal article could be replotted. It would also have the benefit of being peer-reviewed and not original research. And I do understand what the confidence interval means, thank you, and it is highly dependent on, indeed essentially a description of, the regression model chosen. bikeable (talk) 01:56, 30 December 2005 (UTC)
Do you believe that the data has not been replotted from the IPCC source linked to in the caption? Since you understand that the 95% prediction confidence interval describes the model, which was chosen by least sum-of-square error against the historical data, then why did you suggest using a model with a greater number of parameters? Do you know the difference between an R2 and a degrees-of-freedom adjusted R2? —James S. 04:04, 30 December 2005 (UTC)
I have to warn you that I'm not going to get dragged into a debate about this; I've expressed my opinion. My n-polynomial comment was (roughly) an attempt at reductio ad absurdum, almost a joke; and yes, I know that the data is from IPCC, but I would like a peer review of the regression, or, for example, a combined climate/economic model, rather than a simplistic fit to the data, which I think shows almost nothing. bikeable (talk) 06:01, 30 December 2005 (UTC)
is there some particular reason you're registering your 'include' vote in the 'don't include' section??Anastrophe 01:29, 30 December 2005 (UTC)
The vote was called on Image:Cost-of-storms-by-decade.gif, which I replaced days ago. —James S. 01:32, 30 December 2005 (UTC)
My cut & paste mistake, from the previous heading. Everyone here has been following the discussion and knows the debate is about Image:Extreme-weather-cost.gif (the latest revision of Image:Cost-of-storms-by-decade.jpg). --Aude 01:47, 30 December 2005 (UTC)
Why did you attempt to edit the subject of the poll after our votes had already been cast? —James S. 04:04, 30 December 2005 (UTC)
Do you have a specific criticism that can be addressed, or only general argument by appeal to emotion? —James S. 04:04, 30 December 2005 (UTC)
  • As far as I'm concerned, all non-historical data is to be rejected as too controversial (these lengthy and time-consuming discussions would seem to demonstrate that), with the sole exception being predictions made by authoritative sources whose contributions are notable enough to be included in the article. Such sources include IPCC and leading researchers in the field of the economics of global warming. End of story, for me. Rd232 talk 17:11, 30 December 2005 (UTC)
  • Don't include. I've made my position clear earlier. The underlying model is not justified, and will not be acceptable unless published in a real, non-bogus, peer-reviewed venue. The fact that the model and prediction have changed 4 or 5 (or more...I've lost count) times due to feed-back is additional evidence about the fragility of the extrapolation. --Stephan Schulz 18:52, 1 January 2006 (UTC)

Why Saffir-Simpson Scale stops at 5

"When you get up into winds in excess of 155 mph you have enough damage," Simpson said in a 1999 interview with the National Weather Log, a publication of the National Oceanic and Atmospheric Administration. "If that extreme wind sustains itself for as much as six seconds on a building it's going to cause rupturing damages that are serious no matter how well it's engineered. So I think that it's immaterial what will happen with winds stronger than 156 miles per hour. That's the reason why we didn't try to go any higher than that," Simpson said. Simesa 09:48, 30 December 2005 (UTC)

Is it your position that total amount of damage is no longer proportional to windspeed above 155 mph? That is absurd and you know it. —James S. 10:46, 30 December 2005 (UTC)
If it takes 156 mph to bring down a building, a 200mph wind causes no more damage than a 156mph one, does it? There's an asymptotic limit: when everything's destroyed, it can't get any worse. Ergo damage not always proportional to windspeed. Rd232 talk 17:05, 30 December 2005 (UTC)
The average global windspeed must be less than half of the speed of sound. Way too much upside on the storm strength dimension. I agree that there is finite coastline. I am astonished by those who have claimed that there is not clear and convincing evidence that storm strength is increasing at an accelerating pace at sea, on coasts, and inland. —James S. 20:18, 1 January 2006 (UTC)
"The average global windspeed must be less than half of the speed of sound." - what does that have to do with the price of fish?? Rd232 talk 01:15, 2 January 2006 (UTC)
It's the upper bound on average wind speed from statistical gas mechanics. The answer to your question is yes, if it takes 156 mph to bring down a building, a 200mph wind will cause more damage than a 156mph one, because more buildings will be brought down, on average, by any increase. There is no magical upper limit at which wind speed increases will no longer do increasing damage. The fact that you and Simesa would even suggest such a thing shows the superiority of my position, and the bankruptcy of yours. How do you look yourself in a mirror after suggesting such nonsense? —James S. 05:51, 2 January 2006 (UTC)
heh. recommend removing the "This user assumes good faith" userbox on your userpage there, chief. Anastrophe 05:55, 2 January 2006 (UTC)

Vote on Image:Extreme-weather-cost.gif

Subject of this poll, with 2005 data added; source data: IPCC, 2001

this (Extreme-weather-cost.gif) graph is not the subject of this poll. it has been modified since the time the polling began. Anastrophe 07:04, 2 January 2006 (UTC)

Alternative; source data: IPCC, 2001

My cut & paste mistake, with the previous poll, from the previous heading. Nrcprm2026 insists on a new poll. I think it's ridiculous that it has to come to this, but so be it. Let's just settle this, once and for all, rather then an ongoing revert war. Let's vote on whether or not to include the graph Image:Extreme-weather-cost.gif in the Global warming article. --Aude 01:53, 30 December 2005 (UTC)

I note that you attempted to change the subject of the earlier poll after votes had already been cast. Furthermore, I do not believe that truth is decided by majority vote. I reserve the right to submit this matter to arbitration or RfC. —James S. 04:04, 30 December 2005 (UTC)
Please see Wikipedia:Dispute_resolution#Conduct_a_survey — this step should be taken first, before going to mediation or arbitration. -Aude (talk | contribs) 04:19, 30 December 2005 (UTC)

This vote also extends to global warming-related articles including Effects of global warming, Fossil fuel, Mitigation of global warming, and others. --Aude 01:53, 30 December 2005 (UTC)

I object to your attempt to poll questions about articles on other articles' talk pages. I will only recognize the result of polls concerning articles on their individual talk pages, and reserve the right to raise this question in RfC and arbitration. —James S. 04:04, 30 December 2005 (UTC)
I think it's unnecessary for us to hold polls in all the related articles, and us all to follow you around to vote. This talk page and discussion is referenced on the related articles' talk pages. ---Aude (talk | contribs) 04:26, 30 December 2005 (UTC)

Include Image:Extreme-weather-cost.gif

Do not include Image:Extreme-weather-cost.gif

I'd be inclined to agree to a freeze if you would give me more time to find the source data. So far all I have is this list of names: Frich P., Alexander L.V., Della-Marta P., Gleason B., Haylock M., Klein Tank A. and Peterson T. —James S. 08:41, 30 December 2005 (UTC)
  • Simesa 09:56, 30 December 2005 (UTC) - Including data from 2005 would tend to support the projection - but ignore the fact that that much damage occurs only once per city (and that New Orleans is unique - see Loop Current). Bouncing the rubble won't cause losses. Also, higher winds won't necessarily cause more damage (see Simpson quote above). An exponential projection has no physical effects arguing for it. This graph should be frozen out.
  • As far as I'm concerned, all non-historical data is to be rejected as too controversial (these lengthy and time-consuming discussions would seem to demonstrate that), with the sole exception being predictions made by authoritative sources whose contributions are notable enough to be included in the article. Such sources include IPCC and leading researchers in the field of the economics of global warming. End of story, for me. Rd232 talk 17:12, 30 December 2005 (UTC)
  • Based on the sources cited, the 2nd revision Image:Cost-of-storms-by-decade.jpg is a better fit to 2005. Don't shoot the messenger. I can find no credible evidence that the strength of storms is not increasing. I'm still waiting for 1999-2004. —James S. 11:29, 1 January 2006 (UTC)
This is not about whether the strength of storms is increasing. This is about your graphical predictions of future events. Rd232 talk 01:17, 2 January 2006 (UTC)

The graph represented above has been modified from the one this vote was for/against. further, the new graph is even more wildly inaccurate, as no adjustment of the previous figures for inflation has been done.Anastrophe 06:00, 2 January 2006 (UTC)

As requested, I have adjusted the 2005 preliminary figure of $200 billion to inflation-adjusted 2001 U.S. dollars ($178 billion, assuming the majority are paid in the latter half of the year, which seems reasonable.) —James S. 06:50, 2 January 2006 (UTC)

Allegations of POV pushing by Nrcprm2026

On the talk page for Future energy development, Nrcprm2026 stated the following in response to me: "Because you have acted without proper justification, I am replacing the 2nd revision of the graph, because the fact that people like you exist mean that a more serious warning needs to be published". This is a clear admission of POV pushing, rather than an attempt to extrapolate the data in good faith. On this basis alone, the graph should be "banned" from WP, in my opinion.Anastrophe 18:14, 30 December 2005 (UTC)

I admit I'm pretty sick of Nrc pushing his graph. Its now obvious that he is the only one that wants it; his resorting to threats of RFC/Arb is pointless. William M. Connolley 18:17, 30 December 2005 (UTC).

It seems to me like anyone who would want to suppress the relationship between the use of fossil fuel and the increased cost of storms from storm strength is intellectually bankrupt because they must then deny conservation of energy when they try to explain what happens to the excess atmospheric energy caused by global warming. I deny that I have been creating POV graphs, but the choice between the top two most accurate graph won't be settled until I find more recent data. —James S. 07:09, 1 January 2006 (UTC)

This is just a stupid comment. There are a huge number of complex interelated factors when going from increased energy content to the cost of storm damamge. Climatologists are doing leading edge research to try and predict how much storms will increase in the next century. Social factors like where people build, what type of houses they build (I can never get my head around why there seem to be so few brick houses in Florida...), building costs and population increase all impact on the costs of a storm event of a fixed size. You cannot claim to explain all of these factors with a regression to a few years of data. It is just exceptionally simplistic. If you can give a proof of why all these factors can be modelled by a exponential increase in time please consider publishing it in a scientific journal.--NHSavage 10:15, 1 January 2006 (UTC)
A huge number of factors?
  1. sunlight
  2. opacity in visible light
  3. albedo
  4. black box thermal radition
  5. opacity in infrared
  6. radiative forcing
  7. the fact that a huge proportion of human wealth is less than 20 ft. above sea level
What are the limiting factors that might dampen the cost growth and storm strength curves?
  • finite coastline
  • maximum average global windspeed must be less than half of the speed of sound.
Anything else? —James S. 11:10, 1 January 2006 (UTC)
James, have you noticed that you are emphasizing a ~9 fold increase in the real cost of extreme weather while the number of hurricanes (at least hitting the US) did not change significantly over the 1960-1998 window of observation? Maybe the storms did get a little stronger, but given there is little change in the number of storms (or major storms) during this interval, isn't it logical to assume that other factors (such as population growth and coastal development) are also playing an important role in governing the trend? Dragons flight 12:25, 1 January 2006 (UTC)
No, I am suggesting a 20-fold increase from 2005 ($200 billion) to 2025 ($4 trillion.) Based on the preliminary 2005 total worldwide cost of $200 billion. Population growth and coastal development plays a larger part than the increasing strength of storms. —James S. 20:42, 1 January 2006 (UTC)
so, i take it you have re-adjusted for inflation the previous data against which you are now comparing it to 2005 costs?Anastrophe 21:36, 1 January 2006 (UTC)
PS. Storm intensity is primarily governed by available temperature gradients (i.e. ocean relative to upper troposphere), the change in which would be a second order effect relative to climate change. The first order effect, simply making everything warmer, has a relatively neglible impact on storm intensity. Which is not to say that global warming won't affect storm intensity; it will, but you wouldn't be able to get good estimate of those effects by simply appealing to conservation of energy. Dragons flight 12:25, 1 January 2006 (UTC)
So when more water evaporates and it rains more because it's warmer, then that is a "secondary effect" which is somehow outside the perview of nonlinear regression? Why? —James S. 20:42, 1 January 2006 (UTC)
There is a obvious factor that might dampen the cost growth - if houses get damaged by storms often enough people are likely to a) build few houses in storm prone areas b) build stronger houses. I am not saying that this will happen just that it could dampen the cost growth. If there were to be a global recession on the scale of 1929 this could affect building rates dramatically and thus there would be less wealth to be destroyed. These are just the objection I came up with without doing any research - I am sure an expert in this field could come up with many other factors.--NHSavage 13:00, 1 January 2006 (UTC)
More people means more homes [4]. Not linearly perhaps, but i think it's reasonable to assume one new home for say, every four new people. So, figure Florida alone has 1.5 million more homes than it had in 1980. Most of florida's booming development has been along its lovely coastline. Figure at minimum, 500,000 new homes near the coastline. Figure only one in ten of those homes has been destroyed by 'extreme weather' over the last twenty years (excluding 2000-2005, since pop. figures are not included for that in the referenced data). Figure a conservative $100,000 to replace each of those 50,000 homes. There's $5,000,000,000 in costs over and above what would have been the costs without the population growth, based on an extremely conservative guestimate. If we're less conservative....median home price in Florida between 1980 and 2000 ranged from a low of about $240,000 to a high of about $360,000 [5]. So let's say, 20% of those homes were destroyed, at a median of $300,000 - there you're talking $30B in extreme weather costs. just for Florida. These factors *could* be taken into account by the graph - by adjusting for the increased population, increased construction, and in general, wildly increasing value of real estate. It would require a lot of hard work, but it would make for a graph that has some probative value, rather than this simplistic graph, that ignores these factors. And the above doesn't even take into account commercial real estate, which could further significantly inflate the costs of damage. Anastrophe 19:01, 1 January 2006 (UTC)

user Nrcprm2026 is now insisting on changing the title of this section from "explicit admission of POV pushing by Nrcprm2026" to "Allegations of POV pushing by Nrcprm2026". most recent edit accompanied by summary "These allegations are baseless". Let me repeat what Nrcprm2026 stated, which I maintain is an explicit admission of POV pushing: "I am replacing the 2nd revision of the graph, because the fact that people like you exist mean that a more serious warning needs to be published". That statement, in a nutshell, says that he's choosing which graph to publish based upon how 'dire' the warning is that he believes the graph suggests. He's saying he's going to choose to publish the graph that aligns most with the point of view he wishes to promulgate. doesn't get much clearer than that. Anastrophe 00:21, 2 January 2006 (UTC)

My statement means that given two graphs which at the time seemed equally plausable, the existence of knee-jerk reactions against the lowest of the two because it was "two high" leaves me editorial discretion to choose the higher prediction of the two. When that statement was made, I believed that the 2005 figure would amount to $120 billion. Now we know the preliminary estimate from ENS is $200 billion, making the greater of the two estimates -- the very extrapolation which Anastrophe complains of here -- the more accurate by far. —James S. 23:00, 2 January 2006 (UTC)
so now we're taking preliminary estimates as data for your graph. isn't that great! your CI pretty much gives you such absurd latitude that you'll "always be right". 'extreme weather costs could be between 30 billion and a trillion dollars next year!!!!!!' well, sure. and it's true. and it also has no informational value. Anastrophe 18:01, 3 January 2006 (UTC)
Retrieved from "https://en.wikipedia.org/w/index.php?title=Talk:Climate_change/Archive_10&oldid=1248792712"