Re: [ferret_users] shading 95% significance

[Peter Szabo <szabpet83@xxxxxxxxx>]

Hello Nitin,

As Jorg said, you definitely need more than just the average of two periods.
I think you have 10 timesteps since you want to know whether the difference of the JJAS means between the two periods are significant or not (June, July, August and September in the same year would not be independent).

You can see here an example using Welch-test how to do significance test for two 30-years of temperature means.

let var1=xi[l=1:30@var]
let var2=yi[l=1:30@var]
let ave1=xi[l=1:30@ave]
let ave2=yi[l=1:30@ave]

let diffave=ave2-ave1 !difference

let frec=(1/29)*(((var1/30)/(var1/30+var2/30))^2)+(1/29)*(((var2/30)/(var1/30+var2/30))^2)
let f1=int(1/frec) !degrees of freedom

!here is the "built-in" t-table to determine whether it is significant or not at 95% level

!note that you need different values for difference confidence intervals

!note that these values are for 2*30 years from the table

let p1=if f1 eq 30 then 2.042 else p12
let p12=if f1 le 32 then 2.037 else p13
let p13=if f1 le 34 then 2.032 else p14
let p14=if f1 le 37 then 2.026 else p15
let p15=if f1 le 39 then 2.023 else p16
let p16=if f1 le 43 then 2.016 else p17
let p17=if f1 le 46 then 2.013 else p18
let p18=if f1 le 51 then 2.009 else p19
let p19=if f1 le 56 then 2.005 else 2.00

let ttest95_xx=if abs(diffave)/((var1/30+var2/30)^0.5) ge p1 then 1

To plot this over the difference field (diffave) you need to do it in a tricky way with using
go polymark poly and xsequence commands.

Hope this helps!

Regards,

Peter Szabo

On Thu, Jan 28, 2016 at 11:29 AM, Nitin Patil <nitinpatil85@xxxxxxxxx> wrote:

at each grid i have 40 timesteps mean difference. using this can we do?

--

Regards,

Nitin Patil

On Thu, Jan 28, 2016 at 3:56 PM, Kaduk, Jorg D. (Dr.) <jk61@xxxxxxxxxxxxxxx> wrote:

Hello Nitin,
it sounds like a
you havejust the difference of two numbers. Unless you have any kind of distribution that tells you about the variability of the difference between two decades I think you cannot determine a confidence interval. Or maybe I misunderstood.
Best wishes, Jörg


_________ _______________________________
From: owner-ferret_users@xxxxxxxx [owner-ferret_users@xxxxxxxx] on behalf of Nitin Patil [nitinpatil85@xxxxxxxxx]
Sent: 28 January 2016 04:44
To: ferret_users@xxxxxxxx
Subject: [ferret_users] shading 95% significance

Dear Ferret users,

I have gridded nc data for the years JJAS 2001-2010 and 1991-2000.

I am subtracting (2001-2010) minus (1991-2000), I am getting the difference at each grid.

What I have to do is only highlight the difference which is statistically significant at 95%?

Any help will be appreciated.

--
Regards,
Nitin Patil