[ncl-talk] computing confidence interval from ttest

[andrew.kren at noaa.gov]

I basically have two gfs experiments and computed anomaly correlations to see if they are significantly different. I then computed differences in the two and want to use the t test to test this. But ncep uses the 2 sigma standard deviation with a varying z value based on sample size. I am not sure if these are the same or different, and if so, which one to use. Thanks Dennis 

Andrew Kren

From: Dennis Shea
Sent: Tuesday, April 4, 2017 2:26 PM
To: Andrew Kren - NOAA Affiliate
Cc: ncl-talk
Subject: Re: [ncl-talk] computing confidence interval from ttest

I would say so ... However, to be upfront, I am not sure of he original question. 
Note that correlations are bounded [-1 to +1]. Hence, not 'normally distributed.' Usually, the correlations are put into a normal distribution via the Fischer z-transform. Means and std dev are in 'z-space'; the results are transformed back
 
 Look at the example for 
   https://www.ncl.ucar.edu/Document/Functions/Built-in/escorc.shtml
It uses the large sample size z-value1.96 ... you can use the tvalue since tval ==> zval for large sample size
        zlow = z - 1.96*se                ; 95%  (2.58 for 99%)
        zhi  = z + 1.96*se                 
 




On Tue, Apr 4, 2017 at 2:07 PM, Andrew Kren - NOAA Affiliate <andrew.kren at noaa.gov> wrote:
Thanks Dennis,
In general, can the ttest be run on anomaly correlations?

On Tue, Apr 4, 2017 at 2:01 PM, Dennis Shea <shea at ucar.edu> wrote:
Has anybody answered your question?

http://www.itl.nist.gov/div898/handbook/eda/section3/eda352.htm
See the definition of the confidene interval.

   tval_opt =True                                 ; p & t-value
   pt          = ttest(aveX,varX,sX, aveY,varY,sY, iflag, tval_opt)
   print(pt)
The 2nd element is the t-value
HTH




On Fri, Mar 31, 2017 at 3:49 PM, Andrew Kren - NOAA Affiliate <andrew.kren at noaa.gov> wrote:
Dear ncl-talk,

I'm using the function ttest in NCL to compute the two-tailed significance of a set of correlations. I plot the correlations (anomaly) as a time series as most people do. I compute the difference between the two and want to also plot an envelope that encompasses the 95% confidence range for the range of values. Is there a way to determine this from the ttest? 


-- 
Andrew Kren, PhD
Research Scientist I, Global Observing Systems Analysis (GOSA) Group
NOAA ESRL Global Systems Division (Rm 3C515)
325 Broadway, Boulder, CO 80305
(303) 497-5418

_______________________________________________
ncl-talk mailing list
ncl-talk at ucar.edu
List instructions, subscriber options, unsubscribe:
http://mailman.ucar.edu/mailman/listinfo/ncl-talk





-- 
Andrew Kren, PhD
Research Scientist I, Global Observing Systems Analysis (GOSA) Group
NOAA ESRL Global Systems Division (Rm 3C515)
325 Broadway, Boulder, CO 80305
(303) 497-5418


-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mailman.ucar.edu/pipermail/ncl-talk/attachments/20170404/34b5cbda/attachment.html