Bishop Hill Official sceptics go gambling
Jun 15, 2015 Climate: Surface The Committee for Skeptical Inquiry have issued a challenge to the Heartland Institute - one of those climate bets that come along to enliven things for us from time to time. I must say, I think it would have helped if they had got someone who spoke English to write the text for them, because I'm not entirely clear what the bet is.
The Committee for Skeptical Inquiry (CSI) hereby presents to the Heartland Institute a challenge as to whether the Earth’s climate will set a new record high temperature this year. The challenge will be settled using the NASA GISS mean global land surface temperatures for the conventional climate averaging period (defined by the World Meteorological Organization as 30 years) ending on December 31, 2015. If the global average temperature does not exceed the mean temperature for an equal period ending on the same date in any previous year for which complete data exist, CSI will donate $25,000 to a nonprofit to be designated by Heartland. Otherwise, Heartland will be asked to donate $25,000 to a science education nonprofit designated by CSI. It is CSI’s intent to repeat this challenge every year for the next 30 years.
I think they are saying that they expect the 30 year mean to be higher each year than any other 30 year mean on record. If so then the bet is obviously piffle. The annual temperature could fall dramatically this year and the 30-year mean would still increase.
Reader Comments (68)
SMS:
Sam Ting famously bet $10 on there being no resonance close to 3 GeV when he was working on the J/ψ particle: that was of course precisely the energy range he was exploring, with the modern accepted value for rest mass being 3.0969 GeV.
This is a bogus bet. They win even if the warming trend is 0.01C per decade, just so long as natural variability doesnt disturb the trend, and they have the pre-2000 increases locked in. They lose only if there is persistent climate variability that cools. Why 30 year and not 15 year? Why GISS and not RSS or UAH?
Oh that's right, they'be done enough GISS fiddling to keep things warming as much as possible.
"GISS mean global land surface " - the easiest ones to fiddle with and to get fooled via UHI effect. No dice. Use satellites and the whole globe. RSS.
Here's a better better bet: If the decadal trend is above 0.2C on the average of 4 main datasets (UAH, RSS, HADcrut GISS,), meaning 2015 is at least 0.2C above the 2000-2010 average, showing that the warmists are right, they get the money. If the decadal trend is below what the warmists are predicting will happen, they lose.
The warmists would be losing for 70 years straight by now so I dont think they'd take it up.
"This is a bogus bet. They win even if the warming trend is 0.01C per decade, just so long as natural variability doesnt disturb the trend, and they have the pre-2000 increases locked in."
How do you work that out?
"Here's a better better bet"
There are all sorts of better bets, if your intention is just to win. But if you offer them a "fixed" bet, they'll point out the fix just as we would, and you'll get dragged down to their level. The real prize here is not the piles of cash, but the impact on the public image and respectability of the two organisations.
Personally, I think it comes off as a rather silly and childish stunt, a "Gotcha!" attempt with a trick question. It would suit the dignity of the GWPF to politely decline to get involved in "public bets", but maybe to express surprise that experienced scientific sceptics would come up with such a badly-designed bet. Taking part in a scientific debate is a better way of deciding the matter than making bets, but if CSI do want to discuss publishing a range of falsifiable predictions of the climate models in conjunction with GWPF, that distinguish the opposing hypotheses and so could be used to properly test the AGW hypothesis and validate (or invalidate) the models, then that would be a scientifically useful thing to do. I'm sure GWPF would be pleased to take part in a debate or a joint publication of falsifiable predictions with anyone qualified to do so productively, and I'm sure CSI are perfectly capable.
How do you work that out?"
temps rose in the 80s and 90s until about 2001, and have since been in a relative plataeu. So for the next 15 years, the 'before' will have 1985-2000 numbers and the 'after' will have 2000-2015 numbers. This biases the result in their favor.
Suppose I were bet the stock market tomorrow would go up ... relative to where it was in 2009 in the middle of the crisis and recession.
They are locked in to 'win' just so long as things dont go 'backwards' for them too much in the future. Even if Co2 has zero impact, its a more than even chance they would win.
A fair bet would be based on CO2 warming *trends*, hence my proposal of 0.2C/decade. the only way to be on a path to 2C of warming by 2100 is to have 0.2C/decade warming.
A fiar bet would use global mean temp and not *land* temp.
It is a silly stunt, but a good response would be to expose it as such.
No, I meant the 0.01 C/decade bit. Not that I doubt you, but it wasn't immediately obvious to me how you'd estimated it.
Oh, that's simple. They win if the 30 year average is monotonically increasing.
Even if the trend is a mere 0.01C/decade, so long as temps are monotonically increasing each year (slightly), they win.
Actually, because of the built bias from prior years, they win even if this doesnt happen. They could win so long as
temps simply dont go down too much for too long.
"Even if the trend is a mere 0.01C/decade, so long as temps are monotonically increasing each year (slightly), they win."
Yes, but how did you work that out?
The values vary randomly from year to year on either side of the trend line. If the rise due to the trend is much smaller than the variation, then the noise dominates and the odds of them winning drop precipitately. If the rise due to the trend is bigger than the random variation, the odds get strongly skewed in their favour. If the rise over 10 years matched the variation, then they ought to win roughly every 10 years, and so on.
Having a quick look at GISSTEMP's 30-year moving average, it looks like the noise standard deviation is about 0.015 C in any given year (and double that to get the 95% interval), so I think you're probably right that the trend needed to push the odds in their favour is far less than the trend the models predict, but I'm not sure that 0.001 C/year trend rise would show up against a +/- 0.03 C spread.
Or was you ignoring the noise and saying that *any* positive trend would let them win? I don't think the implications of their suggestion are at all obvious, or easy to connect with the hypotheses people are actually putting forward - one reason why I thought it was a very odd statistic for them to pick. I assume they did it primarily for the publicity stunt, and haven't actually thought the mathematics through properly.
I did it in my head! yes, Here's another way to look at it:
1. It's a 30 year average, right? Last years average is 1985-2014, this year 1986-2015, etc. So basically, they 'win' so long as year 31 in the series is *always* greater than year 0, ie they win in 2015 so long as 2015 temps are higher than 1985, because next years 30-year average will subtract 1985 and add 2015. So long as you are adding a larger numbers and substracting a smaller one, you are increasing the value of the 30-year average. In short, you could recast the best as
"We bet that temperatures will be higher than they were 30 years prior, and we will make that bet each year for the next 30 years."
Does that sound easier to win.
2. Even 0.01C/decade will give them a value 0.03C higher for temp vs 30 years prior. I was ignoring the noise. If your estimate of 0.015C 'noise' annually is correct, then given the trend of 0.01C/decade, year 31 will be .03C+/- .03C at the 95% interval. So given those parameters they win 95% of the time.
3. If 0.01/decade is cutting it close we could double that to 0.02C/decade and make it a 'sure bet' for them. 99.997% or so? Given the last 20 year trend has been about 0.05C/decade, and 0.11C/decade last 70 years, reality seems to be more of a trend than that.
So, bottom line, this is 'sure bet' for them based on the fact that 2015 temps are higher than 1985 temps by enough that its unlikely we will see any temps in the next decade or more be less than temps of 30 years prior. They win so long as the 30 year trend > annual variability, which is true even for small trend values.
Mmm. Their wording is confusing, isn't it?
The way I read it, they lose if the average for the latest 30-year period is less than the average for any previous 30-year period, including the one ending in the previous year.
So the 1985-2015 average has to beat the 1984-2014 average. To find the difference, you have to look at T2015 - T1985 - T2014 + T1984. So it's not just whether 2015 is warmer than 1985, it's also about whether 1984 is warmer than 2014!
Or to put it another way, it depends on whether the jump from 2014 to 2015 is bigger than the jump from 1984 to 1985. For the first few years of the bet, it won't be, because the temperature was rising rapidly during the 1980s and it isn't now. But from 2029 to 2045, assuming the pause continues, that won't generally be true. The jump from 1998 to 1999 is unlikely to be bigger than the jump from 2028 to 2029, so they would very likely lose.
If the trend increase is much smaller than the noise, it can be ignored, and it will be roughly 50:50 whether the current year's 30-year average beats the previous year's 30-year average. But it's not just got to beat the previous year, but *all* earlier years. So it's got a 50:50 chance against the year before, and a 50:50 chance against the year before that, and so on, back until the point where you can no longer ignore the trend. That could either be the 1998--2028 average, when it is claimed the pause started, or back far enough that the accumulating trend rise is comparable with the year-to-year noise. But you only have to be able to go back three or four years for their odds of beating all three or four of them to be slim.
If they were to assume a TCR rise of 1 C per century = 0.01 C per year, it would take about 3 years for that to exceed the 95% spread of the noise, so they'd win something like one in eight (ballpark) years they played. Use a 3 C per century trend, and they'd probably be winning every year.
This is all back-of-envelope stuff - calculating it properly would be more complicated than I'm prepared to spend the time doing just at the moment - but I would guess the break-even point past 2028 is for a trend around 1-3 C per century. That's possibly not unreasonable.
Up until 2028, it's a bad bet to take - you'd be betting on temperatures falling faster than they rose during the 1980s and 1990s. But we all knew that.
It's an interesting maths problem, but the situation here is really about the PR. Even if it turns out to be not a bad deal after 2028, the point is you would be turning them down for over a decade until then, which certain parties will try to turn into a stick to beat the Heartland Institute (and climate sceptics generally) with for that decade. The main issue with it is that it isn't testing a claim Heartland have actually made (AFAIK). That needs to be pointed out. Other than that it's all very silly.
GISS will "adjust" the record to make sure the results are what they need to be.
I am a former Skeptical Inquirer member. They compromised themselves and their mission by joining the popular mania of climate obsession. Until their leaders return to being a truly independent voice for skepticism they are just another worthless faux science group.
"So the 1985-2015 average has to beat the 1984-2014 average."
" To find the difference, you have to look at T2015 - T1985 - T2014 + T1984."
Not so. The 1985 - 2015 average has added 2015 and removed 1984 from the 1984-2014 average.
So all you have to look at T2015 - T1984.
2014 and 1985 are in both averages.
Since temperatures have been rising naturally since the late 1600s it is very likely records will be set as the years go by. The IPCC notes that not all the observed recent warming is caused by CO2.
"Not so. The 1985 - 2015 average has added 2015 and removed 1984 from the 1984-2014 average."
Sorry, you're right. I wasn't careful enough.
The average period is actually 1986-2015, since it's 30 years and we're running it from the *end* of 1985 to the end of 2015.
So we're adding T2015 and subtracting T1985 to the total.
And you're right - whether 2015 beats 2014 depends on just those two temperatures. But after adding one temperature and removing the other, we divide the total by 30 to get the average. Thus, the contribution of the trend to the change is not thirty years worth of trend, but one year.
And if the random variation about the trend is much bigger than one year's worth of trend, it's still only about 50:50 that you'll beat the previous year. And another 50:50 that you'll beat the year before that as well. And so on.
Dividing by 30 to get an average changes nothing in the logic, just the numbers.
This "it's still only about 50:50 that you'll beat the previous year." is quite wrong.
Let me restate the math logic here:
When you add a larger number and subtract a smaller number from a collection of numbers, you raise the average.
Thus: So long as every years temp going forward is higher than the temp. of 30 years prior, you will get a RISING 30 year average each and every year.
What this means is you could have the temperatures going down monotonically for 15 years and still have the '30 year average' rise! The math works.
At the risk of taking this too far, here is the GISS temp 30 year average, and btw this is surface total, not land. The land would be even easier for them to 'beat'. The anomoly is in .001C units so that 1985 was a 0.15C anomoly and 2014 was 0.68C. The average over the 30 years is 47 (0.47C). The std deviation is 15 (or .15C).
15 (1985)
29
36
24
39
38
19
21
29
43
33
46
62
41
41
53
62
60
52
66
60
63
49
60
67
55
58
60
68 (2014)
Now going forward, I will assume the anomaly goes DOWN by 0.01C per year. Doing that gives you the 30-year average results (2nd column):
67.0 47.2
66.0 48.9
65.0 50.1
64.0 51.0
63.0 52.3
62.0 53.1
61.0 53.9
60.0 55.2
59.0 56.5
58.0 57.5
57.0 57.9
56.0 58.7
55.0 59.0 (2027)
54.0 58.7
53.0 59.1
52.0 59.5
51.0 59.4
50.0 59.0
49.0 58.7
48.0 58.5
The 30 year average monotonically goes UP for 12 years (to 2027), even though the temperature goes DOWN monotonically! Until 2028. 2028 of course is 30 years after 1998, which has a 0.62C anomoly. Warm year. So this 'bet' is won by warmists EVEN IF THERE IS NO WARMING FOR ANOTHER 12 YEARS! The average can go up, even if the anomaly in a future year is below the 30 year average! Its all due to the math of temps in 1985-1997 being colder than right now. the only for them to lose is for temps in 2015-2028 to fall below temps in 1985-1997.
Yes. That's what I said!
The bet is biased in their favour for the first 13 years, because of the pre-pause temperatures still to work their way through.
But once they have, things go the other way! We get 17 years with the odds heavily in our favour, even if there's a small trend. That's the bit I'm talking about.
I did do some R code to illustrate the point. I was going to do some more on it and tidy it up, but got distracted and then lost interest. Perhaps you'd like to play with it and take it further? It looked to me as if the break-even point was somewhere around 0.2 C per century - about what I initially guessed but then revised upwards with my back-of-envelope calculation.
# ###################
# Simulate CSI bet over 30 years assuming a trend in C per century
csitest=function(trnd){
# GISSTEMP land
pasttemp=c(26,41,50,38,52,51,23,27,38,57,48,55,84,60,59,69,80,78,70,89,
78,86,65,79,92,79,78,83,89)/100
# Simulate some future temps
temp=c(pasttemp,rnorm(30,mean=0.8,sd=0.07)+(0:29)*trnd/100)
# 30 year running mean
climate=sapply(0:29,function(offset)mean(temp[seq(1+offset,30+offset)]))
# Warmest so far
records=sapply(1:30,function(end)max(climate[1:end]))
# Which ones beat the record?
csiwin=which(climate == records)
plot(1:59,temp,col="grey")
points(30:59,climate,col="red") # Red for Heartland win
points((30:59)[csiwin],climate[csiwin],col="blue") # CSI win
}
csitest(0)
csitest(0.1)
csitest(0.2)
R code analysis of this bogus bet? Now that's gone too far. A lot more study on it than it deserves. LOL.
...in any previous year for which complete data exist.
What might 'complete data' mean?
Are in-filled data sets complete?
"What might 'complete data' mean?"
It means you don't get to talk about the medieval warm period, or the Holocene optimum.