AuthorTopic: Homeland Security to compile database of Journalists and Bloogers  (Read 256 times)

Offline Palloy2

  • Administrator
  • Sous Chef
  • *****
  • Posts: 6097
    • View Profile
    • Palloy's Blog
Quote
AG: Expect Palloy to take some Ad Hom, and totally unjustified, pot shots at Sam Carana.

Carana cheats AGAIN - projecting using HIS best fit curve.

You can't use a "best fit" curve for predictions unless you can say WHY that 3rd order polynomial is what applies to the problem.  Without trying, I am confident I could find a 4th order polynomial (one extra turn) that fits the data and turns downwards in the future.

Also predictions should include a range and give the confidence limits, for example: +4C 1C (70%).  Here the confidence limits would be 0%, which is why Canara left it out.
"The State is a body of armed men."

Offline K-Dog

  • Administrator
  • Sous Chef
  • *****
  • Posts: 2696
    • View Profile
    • K-Dog
Re: Homeland Security to compile database of Journalists and Bloogers
« Reply #1 on: April 07, 2018, 06:03:16 PM »
Quote
AG: Expect Palloy to take some Ad Hom, and totally unjustified, pot shots at Sam Carana.

Carana cheats AGAIN - projecting using HIS best fit curve.

You can't use a "best fit" curve for predictions unless you can say WHY that 3rd order polynomial is what applies to the problem.  Without trying, I am confident I could find a 4th order polynomial (one extra turn) that fits the data and turns downwards in the future.

Also predictions should include a range and give the confidence limits, for example: +4C 1C (70%).  Here the confidence limits would be 0%, which is why Canara left it out.

Now I have to throw in my two cents.  A second order polynomial is normally good'nuff for government work.  A third order polynomial may fit better but have less predictive value, and a fourth would be playing games.

The difference between a third and second order fit is relevant.  The third can force overall graph smoothness which over regions of interest and now is probably on one of those regions of the graph, could enhance short term predictive value.  If the difference shows strong improvement in fit then fit is perhaps enhanced over predictive value.  Detailed study of the actual model coefficients would reveal where in the future the third order fit would produce inflection as it must.  A second order fit would be most honest of all.

The only Ad Hom you will get for me is I hope you don't suggest we are all freezing.


Under ideal conditions of temperature and pressure the organism will grow without limit.

Offline Palloy2

  • Administrator
  • Sous Chef
  • *****
  • Posts: 6097
    • View Profile
    • Palloy's Blog
Re: Homeland Security to compile database of Journalists and Bloogers
« Reply #2 on: April 07, 2018, 08:03:56 PM »
Quote
K-dog: The difference between a third and second order fit is relevant. ... If the difference shows strong improvement in fit then fit is perhaps enhanced over predictive value.

No, how good the fit is is irrelevant. What matters is how closely the formula for the fit actually fits the problem you are looking at. If you can't say WHY it is a good fit, then you can't use it for prediction at all.

We know that temperature varies with time, but we also know it the result of a complex set of issues (how much CO2 is produced, how much CH4, how much SF6, the earth's albedo, which in turn varies locally, how much ocean currents move the heat around, etc).  So the answer isn't going to be a simple 3rd order equation, so the 3rd order equation isn't a good fit by design, it is by luck and will be hopeless at predicting into the future.
"The State is a body of armed men."

 

Related Topics

  Subject / Started by Replies Last post
7 Replies
1393 Views
Last post March 19, 2013, 04:34:45 PM
by Mark N
Homeland

Started by RE History

2 Replies
807 Views
Last post February 27, 2016, 05:28:24 PM
by RE
10 Replies
735 Views
Last post April 07, 2018, 05:04:32 PM
by agelbert