Friday, May 13, 2011

Cherry Picking, Trends and Climate Catfights - For Your BS Dectector

For those of you that don't know, I have a basic bullshit detector that I occasionally add to. There is a lot of bullshit, so the detector has a lot of room to grow. I have been meaning to add a climate science filter for the BS detector, but it is difficult to simplify things enough for most folks to understand. Since Tamino and Bob Tisdale are in a cheery picking cat fight, now is as good a time as any.

A climate trend is generally accepted to be 30 years. That is because there is weather and there is climate. Thirty year trends make it somewhat easy to see past trends but it makes it damn hard to predict trends. Even past trends can be somewhat controversial. It is difficult to assign with complete confidence the factor that forced the trend. The 1910 to 1940 trend is one that is often compared to the 1980 to 2010 trend (you can fudge a few years either way as long as the period is at least thirty years long).

Since I have been accused of cherry picking, I try to follow the rules, still if I mention the 1910 to 1940 period, "CHERRY PICKING!!" is the first thing I hear on most of the warmer sites. Tamino did his mathematical magic a year or so ago and came up with what should be the minimum time period for a statistically significant trend. His math indicated that roughly 14.7 years is the minimum and that even then it depends, if it can be statistically significant. I am no statistically wizard, but what he did made sense.

That is why I picked 15 years as the regression length for my 15 30 stacked regression method. With each regression 15 years long and 30 years of regressions it makes it a little harder to call it cherry picking. It is right at the edge though.

The minimum thirty year trends cuts both ways. If the actual measured trend falls outside of the modeled trend after thirty years, that makes people question the validity of the models. That makes the modelers reconsider their models and change the uncertainty of their model accuracy. The length of the period also makes it virtually impossible to predict trends. Waiting until after something happens to predict it happened doesn't impress most folks.

Wagering on climate changes takes cherry picking out of the picture. Had it not been for the wager that the 2015 temperature will be higher than the 2008 temperature I would never even thought of using such a short period of time. If it had not been for the cherry picking issue, I would not have created my stacked 15 30 regression method. Just like the bullshit detector, the stacked regression method is a work in progress.

The spat between Tamino and Tisdale is over use of an "apparent" trend in the ARGO ocean heat content data starting in 2003. That is only eight years of data, so it is not a trend, it can be called an apparent trend developing, but not a trend. Since Mr. Tisdale did not phrase his post answer as a question, he can't win the trend thing. "What could be a developing trend starting in 2003...", would have avoid the trend argument but since cherry picking is more ideological than rational, there is no way to avoid that, other than a friendly wager.

Tamino is pretty math savvy, but is not infallible mathematically. He has face planted trying to defend an inappropriate use of a particular statistical method in the past, being human, he is likely to screw up again. We are all human. Wagering is a fun way to reduce the tension of the cat fighting. I am not talking mega buck wagering, that is a fool's way to attempt to win an argument. If someone fails to take his outrageous wager, a fool thinks he can crow. There is always someone with more money and more savvy on any particular subject. Unless you have the nuts, don't bet big and if you do have the nuts bet wisely to get the most out of it.

I am posting this only so I can add the link to the BS detector. I doubt that Mr. Tisdale and Tamino will make a wager, but if they do, I will try to use the stacked regression method to estimate the odds.

No comments: