Skip to content

conditionally weighted mean with NAs

3 messages · David Winsemius, Joshua Wiley

Joshua Wiley
# 
Dear List,


I am trying to take the weighted average of two numbers (visual acuity
measures from the left and right eye).  For each row, the lowest value
should get a weight of .75, and the highest a weight of .25.  My
problem is, if one value is missing (NA), the remaining one should get
a weight of 1 (i.e., just return the nonmissing value), if both are
missing, NA should be returned.  Below is some example data and the
code I tried to write (Desired is what I actually want).  Any thoughts
or comments would be welcome.

Thanks,

Josh


VA <- cbind(OS = c(.2, 0, 1, -.1, NA, 3, NA),
  OD = c(.3, -.1, .2, -.1, NA, 0, .1),
  Desired = c(0.225, -0.075, 0.4, -0.1, NA, 0.75, 0.1))

## What I tried
weight.combine <- function(left, right) {
  out.r <- right
  out.l <- left
  right <- ifelse(out.r <= out.l, out.r * .75, out.r * .25)
  left <- ifelse(out.r > out.l, out.l * .75, out.l * .25)
  rowSums(cbind(left, right), na.rm = TRUE)
}

## This "works", except it does not handle NAs properly
weight.combine(VA[, "OS"], VA[, "OD"])
David Winsemius
#
On Apr 2, 2011, at 10:20 PM, Joshua Wiley wrote:

            

      
      
      
    

        
VA <- cbind(VA, 0.75*pmin(VA[, "OS"], VA[, "OD"], na.rm=TRUE) +
                   0.25*pmax(VA[, "OS"], VA[, "OD"], na.rm=TRUE) )

 > VA
        OS   OD Desired
[1,]  0.2  0.3   0.225  0.225
[2,]  0.0 -0.1  -0.075 -0.075
[3,]  1.0  0.2   0.400  0.400
[4,] -0.1 -0.1  -0.100 -0.100
[5,]   NA   NA      NA     NA
[6,]  3.0  0.0   0.750  0.750
[7,]   NA  0.1   0.100  0.100

            

      
      
      
    

        
David Winsemius, MD
West Hartford, CT

  
  


  


      

    

    

      
      

        


  

        

      Joshua Wiley
    

    

      
      #
    

  


  

      
On Sat, Apr 2, 2011 at 7:59 PM, David Winsemius <dwinsemius at comcast.net> wrote:

            

      
      
      
    

        
That works wonderfully, and it is so elegant!  Thank you, David.

Josh