How to compute p-values associated with regression coefficients from multiple linear regression?

Hi,

I performed a multiple linear regression analysis using NCL function "reg_multlin" to obtain a 2D (lat,lon) map of standardized regression coefficients (6 independent variables, 30 observations).

     N = dimsizes(y)
     M = N_vars
     X = new( (/M+1,N/), "float" )
     X(0,:) = 1.0
     X(1,:) = x1
     X(2,:) = x2
     X(3,:) = x3
     X(4,:) = x4
     X(5,:) = x5
     X(6,:) = x6
     beta = reg_multlin(y, X, False)

Now, I would like to compute the p-value that is associated to each regression coefficient.
I'm thinking about, firstly, computing the t-value by doing the ratio between a regression coefficient and its standard deviation, afterwards I will use the NCL function "betainc" to get the p-value for a Student t-test.

My problem is that I don't know how to obtain the standard deviation of each regression coefficient using NCL. I tried to move the "reg_multlin" function in a loop over the number of observations:

     do nf = 0,nfils-1,1
-> b_nf(:,nf) = reg_multlin(y, X(:,nf), False)
     end do

     fAODbeta(ns,ilat,ilon) = beta(1) + 0.
     fAODbeta_std(ns,ilat,ilon) = stddev(b_nf(1,:)) + 0.

but I get the following error for the line indicated by an arrow (->)

fatal:Number of dimensions in parameter (1) of (reg_multlin) is (1), (2) dimensions were expected

Could you suggest me how to solve my problem?
Or do you know a better way to estimate the p-value associated with each regression coefficients?

Many thanks!

Best regards,
Susanna
_______________________________________________
ncl-talk mailing list
List instructions, subscriber options, unsubscribe:
http://mailman.ucar.edu/mailman/listinfo/ncl-talk
Received on Wed Jan 8 14:26:32 2014