Friday, July 17, 2009

Is there such a thing as variable bias?

This was written July 13th for posting later

Let's look at the strange behavior of the daily temperatures of two Illinois towns, just 24 miles apart. One would expect that two towns that close together the temperatures would move in sync. But they don't. Below is the difference between the temperature at Carlinville, IL and Hillsboro, IL. I put a 6-month moving average on it.



One can see that there is an approximately 3 year variation in temperature between these two towns. They don't go up and down together. One can do one of two things with this kind of data. One can say that this is erroneous as there is no meteorlological phenomenon which can cause this kind of variation over a 24 mile distance; or one can say "EUREKA, I HAVE FOUND A NEW METEOROLOGICAL PHENOMENON!!! HOT AND COLD BLOBS OF AIR THAT REMAIN STATIONARY FOR MONTHS AND YEARS!!"

As much as one would like to say the latter, one knows that this can't be the case. It is silly to think that this kind of variation can actually happen. It would be like having a local heater or air-conditioner over the entire town.

So, what we see there is not a new meteorological phenomenon, it is bad data, and the amount of variation seen above is the amount of intrinsic error in the data. That means that the data is no more accurate than +/- 3 degrees F. And that means that when the global warming hysteriacs tell you that they know that the world has warmed by 1 degree over the past century you can tell them that they don't know that because the error in the data is greater than 1 degree. If the error is +/- 3 degrees, they can't possibly know that the world has warmed by 1 degree. That is statistical lunacy.

Let's look at a couple of detailed areas of this data. This is the data between 1973-1977



One can see that from 1973 to 1975 Carlinville was generally hotter than the town 24 miles away, hotter by as much as 2.5 degrees F for months at a time. Then for 1976 it swung the other way except for July and August. So, do we have blobs or hot and cold air? Or is this data just bad--too bad to be used to determine global warming?

Let's look at another detailed area, only this time lets only use a 30-day moving average on the temperature.



Here we have the crappy database saying that Carlinville was colder than Hillsboro for 1984. Then 1985 and 1986 in general Carlinville was hotter than Hillsboro and it stayed that way mostly until 1990.



Again we see silly patterns in the data

Hagiograph has wanted me to use statistical tools on a geographical grid to deal with the variations and problems I see in the raw data. The problem is that if one is to use statistical tools, the problem one seeks to fix MUST BE STATISTICAL.

If one has a station which is consistently hotter than the other, or if the average of the differences between the two towns is not zero, that is the bias. One fixes this problem by removing the bias.

But the difference between Carlinville, Illinois and Hillsboro, Illinois, two towns just 24 miles apart is not one of mere statistical variation. It is variable bias.

One can see that the temperature is biased one direction for several years and then the other for several years. That is not a normal kind of statistical bias. It is a sinusoidal wave not a bias. Temperature varies over time. It goes up and down one short and long scales. The problem with the kind of data shown above is that if one removes the sinusoidal signal seen above, then how is one to know if he is removing actual data? If you have an actual DC shift, a bias, one knows that it should be removed, but should one remove a sinusoid with a ~3 year period?

Because of thse problems, geographical manipulation of the data is not valid.

1 comment:

  1. Variable bias occurs when a model is created which incorrectly leaves out one or more important causal factors.

    ReplyDelete