New Computer Fund

Friday, March 11, 2016

How to make history disappear as if by magic

Greg Goodman has a nice post on basic issues with Ordinary Least Squares regression analysis at Dr. Curry's place.  Someone asked about what impact it would paleo reconstructions of sea level.  While I don't have a specific sea level example I do have a sea surface temperature example.

The yellow curve is from the Mohtadi et al. 2010 paper reconstructing temperatures of the Indo-Pacific Warm Pool which has an sample rate of about 400 years and the blue curve is from Oppo et al. 2009 from the same area but it has a sample rate of about 50 years.  With more samples and higher resolution you get a clearer picture.  The Mohtadi reconstruction was one of many reconstructions used in the seriously flawed Marcott et al. paper of 2013.

If you regress Oppo with respect to Mohtadi you would have oranges on the x axis and apples on the y axis.  If you just average the two, the coarser Mohtadi would smooth out the information in the finer Oppo.  Either way you end up with a flatter than it should be past history and a sudden pop, either up or down when the influence of the coarse Mohtadi data ends.  If you pick coarse data or make higher resolution data coarse by inappropriate or "novel" averaging, you can make the details of the past disappear, as if by magic.  Even though Oppo et al. 2009 was available for Marcott and company it was not included in their "ground breaking" paper.

This is like the most basic of basics screw ups, so someone with a bit of knowledge would assume ignoring the obvious has to be deceptive instead of frigging stupid, if the mistake is made by a "professional" and published in a peer reviewed journal.  Unfortunately, since nearly everyone has access to canned statistic packages, stupidly using extremely powerful statistical tools is more likely than intentional deception.

With only the options of dishonest or stupid, tact becomes a bit of an issue.  Most engineers are not know for excessive amounts of tact, generally expect professionals to know what they are doing and have close to anal attention to detail, so they lean toward the dishonest accusation.  Hey!  It is an honest mistake and no one really likes being called stupid.

Back in the day, scientists had plenty of time to ponder prior to responding through snail mail or journals, so they were a lot more creative in parsing their insults.  Now a days, time is money and profanity is more socially acceptable.  Deal with it.

No comments:

Post a Comment