Re: [ferret_users] significance test of compsites

["FISSEHA G. BERHANE" <fishcivilmu@xxxxxxxxx>]

Hello Ansley and Williams,

Thank you for your support.

I got a paper by Terray et al., , 2003: Sea surface temperature associations with the late Indian summer.monsoon. Climate Dyn., 21, 593–618  which documents in its Appendix 1 how to test significance of composites.  They recommend using techniques such as bootstrap.

Thank you,
Fisseha


--- On Mon, 11/5/12, William S. Kessler <william.s.kessler@xxxxxxxx> wrote:

From: William S. Kessler <william.s.kessler@xxxxxxxx>
Subject: Re: [ferret_users] significance test of compsites
To: "Ansley Manke" <ansley.b.manke@xxxxxxxx>
Cc: "Ferret users" <ferret_users@xxxxxxxx>
Date: Monday, November 5, 2012, 3:35 PM

Re the description Ansley points to below:

You can make a t-table in fortran (requires a Gamma-function routine, see Numerical Recipes), or use the attached lookup table. Or you can probably find it via google.

But the key thing you need to do this correctly is an estimate of the degrees of freedom. Very roughly this is the number of "bumps" in your time series (including positive and negative), but is formally:

(Number of terms in the time series)/(Independence timescale)

The independence timescale is usually found as the sum (over all lags) of the product of lag correlations (or lag correlation squared for a single time series). This is tricky to do in Ferret (usually requires a REPEAT loop), but the extra axes available in Ferret 6.8 are helpful.

Be careful about the time units! If you are using "time axis index number" as your unit, and the time axis is regular, the calculation is as above. If you do it in, say, days, then the independence timescale is Delta-t (in days) times the sum of correlations, and the "number of terms" above is the total length of the time series in days.

See the discussion and references on p3015 of Kessler et al (1996). Scales of variability in the equatorial Pacific inferred from the Tropical Atmosphere- Ocean (TAO) buoy array. J.Climate, 9, 2999-3024.
Available in pdf at http://faculty.washington.edu/kessler/abstracts/sampling-abstract.html

BK


On Nov 5, 2012, at 11:49 , Ansley Manke wrote:

> Hi,
> Here is a message from the list, which talks a bit about using a correlation to determine significance of a trend line using a Student's t-test.
>
> http://www.pmel.noaa.gov/maillists/tmap/ferret_users/fu_2009/msg00482.html
>
> if this is not similar to what you want to do, then please search the Ferret archives some more for other possible answers , or write back with more information about the data you are working with and what you want to do with it.
>
> Ansley
>
>
> On 11/1/2012 1:12 PM, FISSEHA G. BERHANE wrote:
>>
>> Hello dear Ferreters,
>>
>> I have composite analyses and I want to do  significance test of my results. Can I do it in Ferret?
>>
>> Thank you,
>> Fisseha
>