Friday, September 12, 2008

Dr Mark Thoma is a Dweeb!

I had posted a couple of times on his thread at: Arctic Ice Loss. His email response was the following:
By the way, it’s not my graph – but I did check it out and I was confident of the source. Two minutes of your time – just click through – shows this (another reason your assertion is incorrect):

So please quit saying I was misleading. In any case, the links and full sourcing were there.

From: Mark Thoma []
Sent: Thursday, September 11, 2008 4:50 PM
To: ''

I don’t remember what you said, but it either lacked respect, or I thought it was wrong.

Dissension is fine, but crossing certain lines of discourse, or making false and misleading claims, is not.

I removed your comment again (the implication that it was removed to stifle dissent is false). If you want to try again without the complaint about the previous moderation, go ahead.
Actually this makes it even worse that he is actually lying since he only linked to another blog {Fasted recorded August rate of Arctic ice loss} and not to the original source.

But I do credit him for now providing the link and information about why they choose not to answer my questions:
What are the error bars for your images?
NSIDC does not have error bars on the timeseries plot shown in the “Daily Image Update” and the daily timeseries plot (usually labeled “Figure 2”) because we strive to keep the images concise and easy to read. Plus, the error bars would be quite small compared to the total extent values in the images.

We estimate error based on accepted knowledge of the sensor capabilities and analysis of the amount of “noise,” or daily variations not explained by changes in weather variables. For average relative error, or error relative to other years, the error is approximately 20,000 to 30,000 square kilometers (7,700 to 11,600 square miles), a small fraction of the total existing sea ice. For average absolute error, or the amount of ice that the sensor measures compared to actual ice on the ground, the error is approximately 50 thousand to 1 million square kilometers (19,300 to 386,100 square miles).

The absolute error values may seem high, but it is important to note that each year has roughly the same absolute error value, so the decline over the long term remains clear. NSIDC has high confidence in sea ice trend statistics and the comparison of sea ice extent between years.
But the questions I raised are still valid. The data is ignored on the error bands and then compared between averages and true numbers, I still consider a sloppy way of presenting the information. Getting any two years to be above or below the average is not hard especially in out of sample data sets.

Here we also see that estimation errors are also high which makes people come to the wrong conclusions based on a simplistic chart. I would also want to see the standard deviations as they change across the months. It seems that these change over the year, with smaller SDs in the beginning of the year and expand out.

Next post, in case that one disappears also:
The graph compares averages over in an in sample with specific measures over out of sample. Thus not sure why error bands are not used to see if it falls outside normal ranges or not. Thanks for the link to: {Above blockquote.}
"Concise and easy to read" sounds like a big excuse to me. This also brings up whether the variance of the data changes over the year. From the graph we see that differences in early May are in quite a narrow band while September is quite wide. So it seems that when presenting such a graph that more care should be done to ferret out such inconsistencies.