Skip to content

SEM:Standard error of std.coef estimates?

4 messages · BdeGroot, John Fox

BdeGroot
# 
Hi,

I am currently working with the sem package in R, to create pathway
diagrams. Id like to use the standardized path coeffcients. To get these, I
use std.coef. However, using this yields only the standardized coefficients,
but does not give me the standard error. Does someone know how to get
std.coef to show the standard error of the standardized path coefficients as
well?


Thanks,
Bastiaan


PS: 
When I use std.coef, all I get is this:

std.coef(path.model.SSI4)
          Std. Estimate                       
par1 par1  0.39499      com_veg <--- tempm    
par2 par2  0.35231      SNutBili <--- tempm   
par3 par3 -0.68170      S_SSI4 <--- tempm     
par4 par4 -0.39145      com_veg <--- Wdeficit 
par5 par5 -0.60025      SNutBili <--- Wdeficit
par6 par6 -0.20562      S_SSI4 <--- Wdeficit  
par7 par7  0.14871      SNutBili <--- com_veg 
par8 par8  0.14905      S_SSI4 <--- com_veg   
par9 par9 -0.39164      S_SSI4 <--- SNutBili
John Fox
# 
Dear Bastiaan,

The standard errors of the standardized coefficients aren't simple because
the standard deviations used to standardize the coefficients are also
subject to sampling error. I can think of two ways to get standard errors
for the standardized coefficients: by the delta method and by bootstrapping.
Neither method is implemented in the sem package. Figuring out how to apply
the former would require some work; I'll put it on my to-do list, but may
not get to it. The second approach could easily be implemented via the boot
package.

I hope this helps,
 John
On
I
coefficients,
as
http://www.nabble.com/SEM%3AStandard-error-of-
http://www.R-project.org/posting-guide.html
BdeGroot
# 
Dear John,

Thank you for your reply. I would like to try the bootstrapping method. I
have the boot package, however, I have not used it before. I do not know how
to compute the correct code to calculate the standarized errors. If possible
and easily achievable, could you please tell me what to do specifically or
provide some more specific guidelines?

Thanks,
Bastiaan
John Fox-6 wrote:

  
    

      
      
    

  


  


      

    

    

      
      

        


  

        

      John Fox
    

    

      
      #
    

  


  

      
Dear Bastiaan,

I've written an appendix on bootstrapping regression models, available at
<http://socserv.mcmaster.ca/jfox/Books/Companion/appendix-bootstrapping.pdf>
, which describes generally how to proceed. In outline, you'll write a
function that takes your data matrix (not covariance matrix) as an argument,
along with a vector argument for indexing the data matrix. Your function
will calculate covariances from the indexed data matrix, fit your model to
this covariance matrix, get the standardized coefficients from the model,
and return the standardized coefficients as a vector. 

An alternative would be to adapt the boot.sem() function in the sem package
to return standardized coefficients. That might prove simpler for you.

Regards,
 John

            

      
      
      
    

        
On

            

      
      
      
    

        
how

            

      
      
      
    

        
possible

            

      
      
      
    

        
because

            

      
      
      
    

        
errors

            

      
      
      
    

        
may

            

      
      
      
    

        
[mailto:r-help-bounces at r-project.org]

            

      
      
      
    

        
these,

            

      
      
      
    

        
coefficients

            

      
      
      
    

        

      
      
    

        

      
      
    

            

      
      
      
    

        
http://www.nabble.com/SEM%3AStandard-error-of-

            

      
      
      
    

        

      
      
    

        
http://www.R-project.org/posting-guide.html