Skip to content

Can you get the DEFT from svyratio?

4 messages · Anthony Damico, Thomas Lumley, Chris Webb

Original
Chris Webb
# 
To Dr. Lumley or anyone who may know the answer,

I am trying to obtain ratio estimates from Levy and Lemeshow's Sampling of
Populations 4th ed. page 281. The results in the book are from STATA.
According to the STATA output, the DEFT is 0.830749

I can recreate all of the results except for DEFT. For svytotal and svymean
I can use the option deff="replace" to obtain DEFT results (by taking the
square root), but I get an error when using this option with svyratio. The
problem can be my poor understanding of how to calculate DEFT, but perhaps
it's not implemented for svyratio?


R code to that fails:

library(survey)

# Creating the dataset
df_tbl_10_1 <-
  data.frame(
    center = c(1,1,1,2,2,2,3,3,3,4,4,4,5,5,5),
    nurse = c(rep(c(1,2,3),5)),
    seen = c(58,44,18,42,63,10,13,18,37,16,32,10,25,23,23),
    referred = c(5,6,6,3,19,2,12,6,30,5,14,4,17,9,14)
  )
df_tbl_10_2 <- df_tbl_10_1[c(2,3,4,6,10,11),]

# Defining the cluster sampling design
svy_tbl_10_2 <-
  svydesign(id=~center + nurse,
            data=df_tbl_10_2,
            fpc= ~M + Nbar)

# Ratio estimates
svyratio(~referred, ~seen, svy_tbl_10_2)
confint(svyratio(~referred, ~seen, svy_tbl_10_2), df=degf(svy_tbl_10_2))

# DEFF
deff(svyratio(~referred, ~seen, svy_tbl_10_2, deff=TRUE))

# DEFT (fails)
sqrt(deff(svyratio(~referred, ~seen, svy_tbl_10_2, deff="replace")))

Fail message:
Error in if (deff) deffs <- matrix(ncol = nd, nrow = nn) : argument is not
interpretable as logical


For other individuals, I have included code that will calculate DEFF and
DEFT for svytotal on page 280, This code doesn't fail.

svytotal(~referred, svy_tbl_10_2)
confint(svytotal(~referred, svy_tbl_10_2), df=degf(svy_tbl_10_2))
deff(svytotal(~referred, svy_tbl_10_2, deff=TRUE))
sqrt(deff(svytotal(~referred, svy_tbl_10_2, deff="replace")))



To recap: can you get the DEFT from svyratio?

Sincerely,
Chris Webb
Anthony Damico
# 
hi, your code isn't runnable at

            fpc= ~M + Nbar)
On Wed, Dec 7, 2016 at 5:03 PM, Chris Webb <iknowchris at gmail.com> wrote:

            

      
      
      
    

        

      
      
    

        

  
  


  


      

    

    

      
      

        


  

        

      Thomas Lumley
    

    

      
      #
    

  


  

      
No, you can't (at the moment), though it shouldn't be too hard to extend.


I can't run your example, though. I get:

Error in eval(expr, envir, enclos) : object 'M' not found

   -thomas


Thomas Lumley
Professor of Biostatistics
University of Auckland

        

      
      
    

  
  


  


      

    

    

      
      

        


  

        

      Chris Webb
    

    

      
      #
    

  


  

      
Whoops. I forgot to add the one line that I included later in the script.
(I extracted the prior code from a longer document). The code should run
now. It uses dplyr although this could have been done with alternative
methods.

How would you extend svyratio to included calculations for DEFT?


library(survey)
library(dplyr)

# Creating the dataset
df_tbl_10_1 <-
  data.frame(
    center = c(1,1,1,2,2,2,3,3,3,4,4,4,5,5,5),
    nurse = c(rep(c(1,2,3),5)),
    seen = c(58,44,18,42,63,10,13,18,37,16,32,10,25,23,23),
    referred = c(5,6,6,3,19,2,12,6,30,5,14,4,17,9,14)
  )
df_tbl_10_2 <- df_tbl_10_1[c(2,3,4,6,10,11),]

library(dplyr)

df_tbl_10_2 <-
  df_tbl_10_2 %>%
  mutate(M=5,
         Nbar=3,
         w=2.5)

# Defining the cluster sampling design
svy_tbl_10_2 <-
  svydesign(id=~center + nurse,
            data=df_tbl_10_2,
            fpc= ~M + Nbar)

# Ratio estimates
svyratio(~referred, ~seen, svy_tbl_10_2)
confint(svyratio(~referred, ~seen, svy_tbl_10_2), df=degf(svy_tbl_10_2))

# DEFF
deff(svyratio(~referred, ~seen, svy_tbl_10_2, deff=TRUE))

# DEFT (fails)
sqrt(deff(svyratio(~referred, ~seen, svy_tbl_10_2, deff="replace")))

        
On Wed, Dec 7, 2016 at 4:03 PM, Chris Webb <iknowchris at gmail.com> wrote: