2011-09-02 02:24:56 | Roy Spencer-- can you spot the big lie? | |

Albatross Julian Brimelow stomatalaperture@gmail... 199.126.232.206 |
This is the latest diatribe from Spencer. Read it and weep. Can you spot the big lie? "Obscurity" (moi) takes Roy to task.
| |

2011-09-02 02:46:44 | ||

dana1981 Dana Nuccitelli dana1981@yahoo... 69.230.106.125 |
Here's the post in question. And the graphic in question: Worse than Spencer's use of Loehle (2007) is his equating science and statistics with religion:
And of course it's full of the usual Spencer conspiracy theories, as Alby notes. I see that he's still been writing ignorant blog posts about economics, too. | |

2011-09-02 02:49:15 | Out of curiousity... | |

John Hartz John Hartz john.hartz@hotmail... 98.122.98.161 |
Has Spencer posted any comments about SkS articles that have taken him to task? | |

2011-09-02 02:58:11 | ||

Albatross Julian Brimelow stomatalaperture@gmail... 199.126.232.206 |
Thanks Dana! | |

2011-09-02 03:11:06 | ||

Ari JokimÃ¤ki arijmaki@yahoo... 91.154.106.251 |
With just a quick peek, here's one of his smaller lies: "Besides, as a meteorologist I must question the whole idea of 100-year event. Since even the longest weather station datasets only go back about 100 years, it is questionable whether we can even say what constitutes a 100-year event." Hmm... longest weather station datasets extend at least back to 1700's, I think. Surely back to 1800's. | |

2011-09-02 03:41:07 | ||

Albatross Julian Brimelow stomatalaperture@gmail... 199.126.232.206 |
Ari, I'm no statistician, but it is my hunderstanding that one does not necessarily need 100 years or more of data to determine the likelihood of a 100-year event. Dikran or KR would be the persons best equipped to address this. Maybe someone shouold ask Tamino...? | |

2011-09-02 03:59:22 | ||

Dikran Marsupial Gavin Cawley gcc@cmp.uea.ac... 139.222.14.107 |
No, you don't need necessarily 100 years worth of data, the branch of statistics you need is called "extreme value theory", but there are even more basic ways you can make a sensible estimate. If you assume that the statistic of interest is Gaussian and you have (say) 20 samples from which to estimate a mean and variance, then there is no reason that you shouldn't use that statistical model to estimate the maximum value you would expect to see from 100 samples, or 1000 or 1000000. As the extreme becomes more extreme, the more sensitive it is likely to be to the uncertainty in estimating the parameters, but you can always take a Bayesian approach and integrate out that source of uncertainty. Nuclear reactors are designed to resist one in a (large number) year events (such as earthquakes or tsinami etc.), but they didn't have (large number) years worth of data to use. As a meteorologist, perhap he should have asked a statistician (Tamino would be a good choice as it is time series data) | |

2011-09-02 04:02:58 | ||

Albatross Julian Brimelow stomatalaperture@gmail... 199.126.232.206 |
Thanks Dikran! You are probably not interested (and I don't blame you), but it might be worth (for what it is worth) pointing this out to Roy. It is quite the glaring misunderstanding/error on his part...embarrassing for him really, but even more so when it is highlighted by someone. | |

2011-09-02 04:12:28 | ||

Dikran Marsupial Gavin Cawley gcc@cmp.uea.ac... 139.222.14.107 |
I'm afraid my comments on Roys blog no longer seem to make it through moderation. However you could always point him towards the excellent introduction by Stewart Coles http://www.amazon.co.uk/Introduction-Statistical-Modeling-Springer-Statistics/dp/1849968748 | |

2011-09-02 04:31:26 | ||

Albatross Julian Brimelow stomatalaperture@gmail... 199.126.232.206 |
Hi Dikran, That is odd-- re your posts not apearing. I know hw has a very aggressive "spam" filter, so one can only post one URL. But perhaps he does not like being challanged by someone in the know. Would it be OK for me to paraphrase what you said at 3:59 am and post the link that you provided? | |

2011-09-02 04:45:02 | ||

Dikran Marsupial Gavin Cawley gcc@cmp.uea.ac... 139.222.14.107 |
yes, go ahead, though I should add the caveat that I am not expert on time-series statistics - Tamino would be a better advisor. However, perhaps you should ask Roy to comment on the plot on page 139 of Coles' book that gives the distribution of 100-year return level for minimum temperatures calculated from a five year temperature time series (Wooster, Ohio - presumably like myself Coles is a fan of P.G. Wodehouse ;o), however I haven't enough time to read the chapter properly today. ;o) The thing Roy doesn't understand is that there is never too little data to be able to make some statistical inference, the only thing that changes is the certainty with which you make the inference. If you have zero data there is infinite uncertainty, if you have an infinite amount of data there is zero uncertainty, most of the time we are somewhere in the middle and you need to look at the credible interval (error bars) on the prediction to know whether it is plausibly worthwhile/useful. | |

2011-09-02 05:01:54 | ||

Albatross Julian Brimelow stomatalaperture@gmail... 199.126.232.206 |
Thanks again Dikran, much appreciated! Maybe I'll ask Grant Foster, just to make sure. | |

2011-09-02 06:03:24 | ||

dana1981 Dana Nuccitelli dana1981@yahoo... 69.230.106.125 |
badger - to my knowledge, Spencer has never said a word about our SkS posts. | |

2011-09-02 08:56:26 | ||

Albatross Julian Brimelow stomatalaperture@gmail... 199.126.232.206 |
OK, i've emailed Tamino. Will let you know what he says if he has time to respond-- i'm sure he gets many requests each and every day. | |

2011-09-02 09:43:06 | Comment | |

Robert Way robert_way19@hotmail... 142.162.9.162 |
Haha it's funny to see this here but I saw the post on spencer's blog and was wondering about the same thing but thought i'd leave it be |