Asked by Anna
on 11 Apr 2013

Hello,

I am trying to output the regression p-values for each cell separately. I've successfully calculated the regression coefficients for each cell using polyfit:

for i=1:695 for j=1:822 summer_trend(1:2,i,j)=polyfit(year,summer(:,i,j),1); end end

However, I would now like to obtain the p-values for the coefficients cell-by-cell. Normally, I would use regstats and select tstat.pval but I'm not sure how to apply this over a matrix. I've tried for example the following code but the problem I have is the functions I've found always return a structure (which is no good when I have a matrix instead of vector...):

for i=1:695 for j=1:822 summer_pval=regstats(summer(:,i,j),year, 'linear',{'tstat'}); end end

Any ideas how I could do this? Many thanks!

*No products are associated with this question.*

Answer by Tom Lane
on 12 Apr 2013

You are right that regstats returns a structure, but it contains numeric values that you can assign into a matrix. For example, something like this:

s = regstats(pop,cdate,'linear'); coefmat(1:2,1) = s.tstat.beta; pvalmat(1:2,1) = s.tstat.pval;

Anna
on 12 Apr 2013

Hi Tom,
Thanks for your reply. However, I'm still not sure how to apply this over a **matrix**. I can't use my [695X822] matrix as the y input of regstats(y,x), as in your example, as that returns the error *"RESPONSES must have a single column"*.

The loop above that I tried using on the other hand returns the error *"The design matrix has more predictor variables than observations"*.

Would you be able to explain how this can be done cell-by-cell using a matrix (as opposed to a vector column) as the y variable? Thanks

Opportunities for recent engineering grads.

## 0 Comments