Hi David,
Indeed I had read that post, but hadn't connected the two. You are
right and your post below highlights the point exactly. To think, the
poster to whom your later message was replying was ready to change an
opinion with neither subject matter knowledge nor statistical
expertise. Wide availability of software and data is both exciting
and frightening!
On Mon, Dec 21, 2009 at 6:37 AM, David Reilly <[log in to unmask]> wrote:
> To your point .... perhaps you failed to read a prior post of mine . Hope
> this helps .
>
>
> From: David Reilly [mailto:[log in to unmask]]
> Sent: Monday, December 07, 2009 12:05 PM
> To: 'John McKellar'; [log in to unmask]
> Cc: [log in to unmask]
> Subject: RE: Analysis of temperature on Earth
>
>
> I am a statistician and have specialized in time series analysis for a
> number of years. I have often been "too busy" to participate in these kind
> of dialogues. Perhaps it is time that I got "unbusy" and deliver some time
> series expertise to the subject at hand. Let me state up front that I am
> totally untrained with respect to the relevant underlying meteorological
> "causes" for this data and simply stated "I walk (talk) where angels fear to
> tread".
>
> I thank others for pointing me to the raw data, 372 months (starting
> 1978/12) of 25 measurements of warming phenomena.
>
> Using 372 monthly values of a single time series to draw inference can be
> flawed for a number of reasons:
>
> 1. The time span is too short to capture longer cycles which might
> explain some of the level shifts that are discussed below.
>
> 2. Using a single series to analyze/predict the future is analogous to
> a car driver using the rear window to forecast future road conditions.
> Tricky business indeed. Single series analysis(ARIMA) suggests somewhat
> naively that the past causes the future while causal models (Transfer
> Functions) embody the impact of user-suggested supporting variables. I did
> not have access to possible causal series data and if any reader can help in
> this regard, I would be willing to incorporate them and report back to the
> list.
>
> 3. Analysts and most software packages often confuse trend with level
> shifts. Level shifts and trends are both intercept changes but reflect
> totally different impacts. It is necessary to clearly make this distinction.
>
> Commentary like " This is not my area of expertise, but if you look at
> the very informative graphs provided, there does seem to be a trend -
> specifically - before 1995 the global temperatures are lower, and after 1995
> they seem to be higher." Are descriptive in nature and often lead to
> spurious conclusions as a before and after analysis/conclusion is not proof
> of a trend but rather a statement of "mean shift".
>
> I have analyzed the data and placed the results/graphs/reports on our web
> site at http://www.autobox.com/warm.zip . The graphs and analyses clearly
> suggest strong seasonal structure. The conclusions are (among others) that
> approximately half the measures suggest a statistical level shift at or
> about the beginning of 1998. These are Level or Step shifts not trends. The
> csv files in the referenced URL support this conclusion.
>
> I hope this helps our greater understanding of this data and I look forward
> to comments on this work.
>
> Dave Reilly
> Automatic Forecasting Systems
> http://www.autobox.com
>
>
> Dave Reilly
> Senior Vice President
> Automatic Forecasting Systems
> www.autobox.com
> 215-675-0652 (office)
> 215-353-7087 (cell)
>
>
>
> -----Original Message-----
> From: A UK-based worldwide e-mail broadcast system mailing list
> [mailto:[log in to unmask]] On Behalf Of Brett Magill
> Sent: Monday, December 21, 2009 1:10 AM
> To: [log in to unmask]
> Subject: Re: SUMMARY: Climate Change: Where are the statisticians?
>
> Moreover David, I would be appalled if even statisticians attempted
> such an analysis with no subject matter knowledge or intelligent
> colleagues with such knowledge to provide advice. A doubly compounded
> error, I should say. Of course, I wouldn't leave it to "intelligent"
> software to make decisions either, however.
You may leave the list at any time by sending the command
SIGNOFF allstat
to [log in to unmask], leaving the subject line blank.
|