No Significant Trend Can Be Discerned from Global Tempature Data

LowDown · Dec 8, 2016

Data from weather stations on the earth's surface are available from 1880 to the present day. According to professional statisticians there is no trend in this data that can properly be distinguished from natural random variation. Multiple statisticians agree with this and have published it, but climate scientists say there's a trend anyway.

This follows a pattern I've seen in other fields of science:

1. Investigator does an experiment and collects data.

2. Investigator takes the data to a statistician, who tells him nothing significant can be concluded from the data.

3. Investigator stops talking to the statistician and publishes the data anyway using improper statistics.

A lot of scientists use statistics computer packages. Data can be fed into these programs and they will produce an answer. What most scientists don't understand is that a statistical method has a whole raft of pre-conditions that have to be true for the answer to be valid.

The key issue with temperature data is that it is time series data. Time series analysis is a whole different field, and there is no evidence that any climate scientists understand it. They all use improper statistics. So where they find a trend in the data there is none.

____________________________
Cowpertwait P.S.P., Metcalfe A.V. (2009), Introductory Time Series with R (Springer).

Shumway R.H., Stoffer D.S. (2011), Time Series Analysis and Its Applications (Springer).

longview · Dec 8, 2016

LowDown said:
Data from weather stations on the earth's surface are available from 1880 to the present day. According to professional statisticians there is no trend in this data that can properly be distinguished from natural random variation. Multiple statisticians agree with this and have published it, but climate scientists say there's a trend anyway.

This follows a pattern I've seen in other fields of science:

1. Investigator does an experiment and collects data.

2. Investigator takes the data to a statistician, who tells him nothing significant can be concluded from the data.

3. Investigator stops talking to the statistician and publishes the data anyway using improper statistics.

A lot of scientists use statistics computer packages. Data can be fed into these programs and they will produce an answer. What most scientists don't understand is that a statistical method has a whole raft of pre-conditions that have to be true for the answer to be valid.

The key issue with temperature data is that it is time series data. Time series analysis is a whole different field, and there is no evidence that any climate scientists understand it. They all use improper statistics. So where they find a trend in the data there is none.

____________________________
Cowpertwait P.S.P., Metcalfe A.V. (2009), Introductory Time Series with R (Springer).

Shumway R.H., Stoffer D.S. (2011), Time Series Analysis and Its Applications (Springer).

Discerning a .1 C pattern in a signal with a daily range of 10 C and a semiannual range of 28 C,
would have it's challenges!

Visbek · Dec 8, 2016

LowDown said:
Data from weather stations on the earth's surface are available from 1880 to the present day. According to professional statisticians there is no trend in this data that can properly be distinguished from natural random variation.

You sure about that?

Multiple statisticians agree with this and have published it

Which statisticians?

Where were they published?

How many is "multiple?" 2? 3?

Do you really expect us to believe that professional climatologists never studied statistics?

This follows a pattern I've seen in other fields of science:

Here we go

1. Investigator does an experiment and collects data.
2. Investigator takes the data to a statistician, who tells him nothing significant can be concluded from the data.
3. Investigator stops talking to the statistician and publishes the data anyway using improper statistics.

Yes, on occasion scientists can fudge the data. Sometimes they even get away with it -- it's not always easy to reproduce an experiment.

However, this isn't actually how climatology works. What they're doing is taking the best data they can; they make climate models based on current data; they take more measurements; they adjust the models; and the cycle continues.

And then, they start observing things like glacial retreat, rising ocean levels, shrinking ice sheets, decreased snow cover etc -- all of which are hard data showing that yes, global temperatures are rising.

At this point, denying that climate change is real is sort of like denying that germs exist, or insisting that the moon is made of cheese. Yes, despite the deniers, the evidence is pretty solid. As is the evidence that humans are spewing out record amounts of CO2 into the atmosphere, that CO2 is a greenhouse gas, and so forth.

The key issue with temperature data is that it is time series data. Time series analysis is a whole different field, and there is no evidence that any climate scientists understand it.

WUWT?

Sigh.

Jack Hays · Dec 8, 2016

Climate News
[h=1]Global warming fails the random natural variation contest[/h]Previously on WUWT, I covered this contest. At that time, Doug J. Keenan stated: There have been many claims of observational evidence for global-warming alarmism. I have argued that all such claims rely on invalid statistical analyses. Some people, though, have asserted that the analyses are valid. Those people assert, in particular, that they can…

Jack Hays · Dec 8, 2016

[h=1]How SkepticalScience views global warming – one way only[/h]Guest essay by Sheldon Walker Most people have probably seen the SkepticalScience graph called “The Escalator”. If you haven’t seen it yet, then you can view it here: Source: The Escalator SkepticalScience claims that “Contrarians” inappropriately “cherrypick” short time periods that show a cooling trend. But SkepticalScience uses a linear regression over the full date range (1970…
Continue reading →

countryboy · Dec 8, 2016

LowDown said:
Data from weather stations on the earth's surface are available from 1880 to the present day. According to professional statisticians there is no trend in this data that can properly be distinguished from natural random variation. Multiple statisticians agree with this and have published it, but climate scientists say there's a trend anyway.

This follows a pattern I've seen in other fields of science:

1. Investigator does an experiment and collects data.

2. Investigator takes the data to a statistician, who tells him nothing significant can be concluded from the data.

3. Investigator stops talking to the statistician and publishes the data anyway using improper statistics.

A lot of scientists use statistics computer packages. Data can be fed into these programs and they will produce an answer. What most scientists don't understand is that a statistical method has a whole raft of pre-conditions that have to be true for the answer to be valid.

The key issue with temperature data is that it is time series data. Time series analysis is a whole different field, and there is no evidence that any climate scientists understand it. They all use improper statistics. So where they find a trend in the data there is none.

____________________________
Cowpertwait P.S.P., Metcalfe A.V. (2009), Introductory Time Series with R (Springer).

Shumway R.H., Stoffer D.S. (2011), Time Series Analysis and Its Applications (Springer).

Global warming alarmists love to scream, every day is "the hottest day on record!". What many of the drones fail to realize is, that is utterly meaningless.

LowDown · Dec 8, 2016

Visbek said:
You sure about that?

Which statisticians?

Where were they published?

How many is "multiple?" 2? 3?

Do you really expect us to believe that professional climatologists never studied statistics?

Here we go

Yes, on occasion scientists can fudge the data. Sometimes they even get away with it -- it's not always easy to reproduce an experiment.

However, this isn't actually how climatology works. What they're doing is taking the best data they can; they make climate models based on current data; they take more measurements; they adjust the models; and the cycle continues.

And then, they start observing things like glacial retreat, rising ocean levels, shrinking ice sheets, decreased snow cover etc -- all of which are hard data showing that yes, global temperatures are rising.

At this point, denying that climate change is real is sort of like denying that germs exist, or insisting that the moon is made of cheese. Yes, despite the deniers, the evidence is pretty solid. As is the evidence that humans are spewing out record amounts of CO2 into the atmosphere, that CO2 is a greenhouse gas, and so forth.

WUWT?

Sigh.

You forgot to put the error bars in the graph you posted. Fail.

Surface Detail · Dec 8, 2016

Land-ocean temperature index, 1880 to present, with base period 1951-1980. The solid black line is the global annual mean and the solid red line is the five-year lowess smooth. The blue uncertainty bars (95% confidence limit) account only for incomplete spatial sampling.

I see what you mean. It's very difficult to make out any sort of trend, isn't it?

Jack Hays · Dec 8, 2016

Surface Detail said:
Land-ocean temperature index, 1880 to present, with base period 1951-1980. The solid black line is the global annual mean and the solid red line is the five-year lowess smooth. The blue uncertainty bars (95% confidence limit) account only for incomplete spatial sampling.

I see what you mean. It's very difficult to make out any sort of trend, isn't it?

You might be interested in the link at #5.

Tim the plumber · Dec 9, 2016

LowDown said:
Data from weather stations on the earth's surface are available from 1880 to the present day. According to professional statisticians there is no trend in this data that can properly be distinguished from natural random variation. Multiple statisticians agree with this and have published it, but climate scientists say there's a trend anyway.

This follows a pattern I've seen in other fields of science:

1. Investigator does an experiment and collects data.

2. Investigator takes the data to a statistician, who tells him nothing significant can be concluded from the data.

3. Investigator stops talking to the statistician and publishes the data anyway using improper statistics.

A lot of scientists use statistics computer packages. Data can be fed into these programs and they will produce an answer. What most scientists don't understand is that a statistical method has a whole raft of pre-conditions that have to be true for the answer to be valid.

The key issue with temperature data is that it is time series data. Time series analysis is a whole different field, and there is no evidence that any climate scientists understand it. They all use improper statistics. So where they find a trend in the data there is none.

____________________________
Cowpertwait P.S.P., Metcalfe A.V. (2009), Introductory Time Series with R (Springer).

Shumway R.H., Stoffer D.S. (2011), Time Series Analysis and Its Applications (Springer).

That was my impression of all of geography past O level (taken at 16 years). A level (18 years old exam) geography was either obvious, wrong or both. We were taken on a day trip to the local university where we were given a day's teaching by lecturers and under grads and I had to argue, and failed to get my point across, that a result of -2 in a statistical analysis forumlae means you have used the wrong one. And that the result of 0.5 also means that the idea is wrong.

Steve Case · Dec 9, 2016

Here's what I wrote over at What's Up With That yesterday:

It’s fairly easy to write some random formulas in Excel and more or less duplicate
the random walk WIKI chart at the top. With a little bit of tuning so that the size of
the random steps are in line with the empirical record you will find that charts that
nearly look like HADCRUT or whatever time series you choose drop out quite often.
If you cheat and add a constant to reflect the warm-up over the last 166 years it’s
really quite astonishing how often a near twin to the observed record is generated.

I wrote that little Excel ditty nearly ten years ago. Without the addition of a constant to
fudge the results, here's an example of what I got to drop out after not all that many trials:

HADCRUT five year running average in Blue and Random Generation in Red

What that tells me is the HADCRUT time series could just be random variation.

I also wrote over at What's Up With That:

If you smoke a little dope you can see a pattern in almost anything.

Tim the plumber · Dec 9, 2016

Steve Case said:
Here's what I wrote over at What's Up With That yesterday:

It’s fairly easy to write some random formulas in Excel and more or less duplicate
the random walk WIKI chart at the top. With a little bit of tuning so that the size of
the random steps are in line with the empirical record you will find that charts that
nearly look like HADCRUT or whatever time series you choose drop out quite often.
If you cheat and add a constant to reflect the warm-up over the last 166 years it’s
really quite astonishing how often a near twin to the observed record is generated.

I wrote that little Excel ditty nearly ten years ago. Without the addition of a constant to
fudge the results, here's an example of what I got to drop out after not all that many trials:

HADCRUT five year running average in Blue and Random Generation in Red

What that tells me is the HADCRUT time series could just be random variation.

I also wrote over at What's Up With That:

If you smoke a little dope you can see a pattern in almost anything.

You need to do better than that though.

What degree of randomness are you using and why is there that much? If you have good answers to these you have a point but otherwise not.

Steve Case · Dec 9, 2016

Tim the plumber said:
You need to do better than that though.

What degree of randomness are you using and why is there that much? If you have good answers to these you have a point but otherwise not.

The HADCRUT data looked like this:

Code:

1850	-0.447
1851	-0.292
1852	-0.294
1853	-0.338
and so on

Cell D1 contained the 1850 value -0.447
Cell D2 contained the formula: =D1+(RAND()-0.5)*0.15
and was copied on down Column D

Code:

-0.447
=D1+(RAND()-0.5)*0.15
=D2+(RAND()-0.5)*0.15
=D3+(RAND()-0.5)*0.15
=D4+(RAND()-0.5)*0.15
=D5+(RAND()-0.5)*0.15
and so on

Cell E5 Contained the formlula =MEDIAN(D1 : D5) *
and was copied on down column E

I found that the Median for the random generation worked better than the average.
Column C contained the five year running HADCRUT averages and Column A the years
Columns A,C and E were plotted out.

So what I wound up with was a comparison of the five year running Average of HADCRUT data with a five year running Median of random data.

You asked about degree of randomness I hope that formula =D1+(RAND()-0.5)*0.15 and the five year running median answered your question.

*Space on either side of the colon so it didn't look like this: =MEDIAN(D1

5)

LowDown · Dec 9, 2016

Surface Detail said:
Land-ocean temperature index, 1880 to present, with base period 1951-1980. The solid black line is the global annual mean and the solid red line is the five-year lowess smooth. The blue uncertainty bars (95% confidence limit) account only for incomplete spatial sampling.

I see what you mean. It's very difficult to make out any sort of trend, isn't it?

If you put the error bars in it becomes obvious that no trend can be discerned. Also, a random time series is very different from a random series. A random time series easily resembles the temperature record we have.

countryboy · Dec 9, 2016

Surface Detail said:
Land-ocean temperature index, 1880 to present, with base period 1951-1980. The solid black line is the global annual mean and the solid red line is the five-year lowess smooth. The blue uncertainty bars (95% confidence limit) account only for incomplete spatial sampling.

I see what you mean. It's very difficult to make out any sort of trend, isn't it?

With a hundred and forty years of data, on a global scale? No, it's not difficult to discern a pattern, it's impossible.

Steve Case · Dec 9, 2016

Visbek said:
You sure about that?

Which statisticians?

Where were they published?

How many is "multiple?" 2? 3?

Do you really expect us to believe that professional climatologists never studied statistics?

Here we go

Yes, on occasion scientists can fudge the data.
Sometimes they even get away with it -- it's not
always easy to reproduce an experiment.

However, this isn't actually how climatology works.
What they're doing is taking the best data they can;
they make climate models based on current data; they
take more measurements; they adjust the models; and
the cycle continues.

And then, they start observing things like glacial retreat,
rising ocean levels, shrinking ice sheets, decreased snow
cover etc -- all of which are hard data showing that yes,
global temperatures are rising.

At this point, denying that climate change is real is
sort of like denying that germs exist, or insisting
that the moon is made of cheese. Yes, despite the deniers,
the evidence is pretty solid. As is the evidence that
humans are spewing out record amounts of CO2 into the
atmosphere, that CO2 is a greenhouse gas, and so forth.

WUWT?

Sigh.

From above:

Yes, on occasion scientists can fudge the data.

When they do, does it look like this?

That snarky Red and Blue Above and Below the line trick really works well.

LowDown · Dec 9, 2016

Steve Case said:
The HADCRUT data looked like this:

Code:

1850 -0.447 1851 -0.292 1852 -0.294 1853 -0.338 and so on

Cell D1 contained the 1850 value -0.447
Cell D2 contained the formula: =D1+(RAND()-0.5)*0.15
and was copied on down Column D

Code:

-0.447 =D1+(RAND()-0.5)*0.15 =D2+(RAND()-0.5)*0.15 =D3+(RAND()-0.5)*0.15 =D4+(RAND()-0.5)*0.15 =D5+(RAND()-0.5)*0.15 and so on

Cell E5 Contained the formlula =MEDIAN(D1 : D5) *
and was copied on down column E

I found that the Median for the random generation worked better than the average.
Column C contained the five year running HADCRUT averages and Column A the years
Columns A,C and E were plotted out.

So what I wound up with was a comparison of the five year running Average of HADCRUT data with a five year running Median of random data.

You asked about degree of randomness I hope that formula =D1+(RAND()-0.5)*0.15 and the five year running median answered your question.

*Space on either side of the colon so it didn't look like this: =MEDIAN(D15)

Yes, this is correct. You have the formula for a random time series, the overall variation of which can be much bigger than the variation that you get from a series of random numbers.

Steve Case · Dec 9, 2016

LowDown said:
Yes, this is correct. You have the formula for a random time series, the overall variation of which can be much bigger than the variation that you get from a series of random numbers.

View attachment 67210968

Gee that line labeled Random Time Series looks very familiar |-:

Those four examples I posted were the closest to HADCRUT that I found.
The numbers, upper left, are an indicator of how close they were, low
numbers being closer. I did that nearly ten years ago, and I don't remember
how many trials I made to get those four, 50 or 100 something like that,
the point being that a time series generated from random numbers can
look very much like the time series we see for world average temperature.

Does that mean world average temperature is random? I'm not saying that.

LowDown · Dec 9, 2016

Steve Case said:
Gee that line labeled Random Time Series looks very familiar |-:

Those four examples I posted were the closest to HADCRUT that I found.
The numbers, upper left, are an indicator of how close they were, low
numbers being closer. I did that nearly ten years ago, and I don't remember
how many trials I made to get those four, 50 or 100 something like that,
the point being that a time series generated from random numbers can
look very much like the time series we see for world average temperature.

Does that mean world average temperature is random? I'm not saying that.

No, it just demonstrates how hard it is to get a trend out of a random time series. You can't average the first half of the graph and compare it to the second half of the graph. You can't do a linear regression, that's completely invalid. Or, that is to say, you can do a linear regression on such data, and a lot of scientists do exactly that, but the results will be invalid. Instead, you have to consider the first derivative. When doing that you find that my red plot is not significantly different from zero. You can generate a random time series that appears to have a definite positive slope, and a linear regression will "prove" it's a positive slope, and yet it's not significantly different from zero.

uncleray · Dec 9, 2016

Arguing Global Warming is a waste of time. We live in a PFTP (Post Fact Time Period) People see what they want to see.

Steve Case · Dec 9, 2016

uncleray said:
Arguing Global Warming is a waste of time. We live in a PFTP (Post Fact Time Period) People see what they want to see.

Considering the fact that Climate Change policy is poised to suck $Trillions
out of the world's economy, I would say it's not a waste of time.

KLATTU · Dec 9, 2016

Surface Detail said:
Land-ocean temperature index, 1880 to present, with base period 1951-1980. The solid black line is the global annual mean and the solid red line is the five-year lowess smooth. The blue uncertainty bars (95% confidence limit) account only for incomplete spatial sampling.

I see what you mean. It's very difficult to make out any sort of trend, isn't it?

Since earth temperatures are always going either up or down( statsis being impossible) it's easy to find a trend one way or the other. you just have to cherry pick endpoints.
Int thi case , the AGW conveniently choose 1880, while conveniently ignoring the trend prior to that. Of course we really have no way of knowing those temperatures, do we?

"bbbb bbb bbbut the proxies are accurate. "the scientists"' ( with the Clinton/Kaine bumper stickers ) all said so! "- AGW illuminati { snicker}

Steve Case · Dec 9, 2016

LowDown said:
No, it just demonstrates how hard it is to get a trend out of a random time series. You can't average the first half of the graph and compare it to the second half of the graph. You can't do a linear regression, that's completely invalid. Or, that is to say, you can do a linear regression on such data, and a lot of scientists do exactly that, but the results will be invalid. Instead, you have to consider the first derivative. When doing that you find that my red plot is not significantly different from zero. You can generate a random time series that appears to have a definite positive slope, and a linear regression will "prove" it's a positive slope, and yet it's not significantly different from zero.

Oh for sure, flip a coin: Heads, add one, tails subtract one. Eventually it will become evident that there isn't a trend.

jmotivator · Dec 9, 2016

Visbek said:
You sure about that?

A statistically derived data set without calculated and verifiable error rates isn't useful for proving trends.

jmotivator · Dec 9, 2016

KLATTU said:
Since earth temperatures are always going either up or down( statsis being impossible) it's easy to find a trend one way or the other. you just have to cherry pick endpoints.
Int thi case , the AGW conveniently choose 1880, while conveniently ignoring the trend prior to that. Of course we really have no way of knowing those temperatures, do we?

"bbbb bbb bbbut the proxies are accurate. "the scientists"' ( with the Clinton/Kaine bumper stickers ) all said so! "- AGW illuminati { snicker}

In short: Trends derived from an arbitrarily selected window of time are also arbitrary.

No Significant Trend Can Be Discerned from Global Tempature Data

LowDown

Curmudgeon

longview

Visbek

Stuck In The Circle

Jack Hays

Traveler

Jack Hays

Traveler

countryboy

LowDown

Curmudgeon

Surface Detail

Jack Hays

Traveler

Tim the plumber

Steve Case

Tim the plumber

Steve Case

LowDown

Curmudgeon

countryboy

Steve Case

LowDown

Curmudgeon

Steve Case

LowDown

Curmudgeon

uncleray

Active member

Steve Case

KLATTU

Steve Case

jmotivator

Computer Gaming Nerd

jmotivator

Computer Gaming Nerd