[R] bootstrap CI of the difference between 2 Cramer's V

[Rui Barradas]

Hello,

Here is your code, corrected. It uses a Goodman-Kruskal gamma function 
in package DescTools.
Function G probably doesn't need tryCatch but I had errors in some test 
runs.


library(boot)

G <- function(x, index, cols = 1:2) {
   y <- x[index, cols]    # bootstrapped data
   g <- x$group[index]    # and groups
   # calculate gamma for each group bootstrap sample
   # (trap errors in case one of the groups is empty)
   g <- tryCatch(
     lapply(split(y[1:2], g), \(x) {
       tbl <- table(x)
       DescTools::GoodmanKruskalGamma(tbl)
     }),
     error = function(e) list(NA_real_, NA_real_)
   )
   # calculate difference
   g[[1]] - g[[2]]
}

set.seed(2022)
# use strata parameter in function boot to resample within each group
results <- boot(
   data = f3,
   statistic = G,
   R = 2000,
   strata = as.factor(f3$group),
   cols = 1:2
)

results
boot.ci(results)


Hope this helps,

Rui Barradas


Às 17:21 de 05/06/2022, varin sacha via R-help escreveu:
> Dear Daniel,
> Dear R-experts,
> 
> I really thank you a lot Daniel. Nobody had answered to me offline. So, thanks.
> I have tried in the same vein for the Goodman-Kruskal gamma for ordinal data. There is an error message at the end of the code. Thanks for your help.
> 
> 
> ##############################
> library(ryouready)
> library(boot)
> 
> shopping1<-c("très important","important","pas important","pas important","important","très important","important","pas important","très important","très important","important","pas important","pas important","important","très important","très important","important","pas important","pas important","important","très important","très important","important","pas important","pas important","important","très important","très important","important","pas important","pas important","important","très important","très important","important","pas important","pas important","important","très important","important")
> 
> statut1<-c("riche","pas riche","moyennement riche","moyennement riche","riche","pas riche","moyennement riche","moyennement riche","riche","pas riche","moyennement riche","riche","pas riche","pas riche","riche","moyennement riche","riche","pas riche","pas riche","pas riche","riche","riche","moyennement riche","riche","riche","moyennement riche","moyennement riche","moyennement riche","pas riche","pas riche","riche","pas riche","riche","pas riche","riche","moyennement riche","riche","pas riche","moyennement riche","riche")
> 
> shopping2<-c("important","pas important","très important","très important","important","très important","pas important","important","pas important","très important","important","important","important","important","pas important","très important","très important","important","pas important","très important","pas important","très important","pas important","très important","important","très important","important","pas important","pas important","important","pas important","très important","pas important","pas important","important","important","très important","très important","pas important","pas important")
> 
> statut2<-c("moyennement riche","pas riche","riche","moyennement riche","moyennement riche","moyennement riche","pas riche","riche","riche","pas riche","moyennement riche","riche","riche","riche","riche","riche","pas riche","moyennement riche","moyennement riche","pas riche","moyennement riche","pas riche","pas riche","pas riche","moyennement riche","riche","moyennement riche","riche","pas riche","riche","moyennement riche","blue","moyennement riche","pas riche","pas riche","riche","riche","pas riche","pas riche","pas riche")
> 
> f1 <- data.frame(shopping=shopping1,statut=statut1,group='grp1')
> f2 <- data.frame(shopping=shopping2,statut=statut2,group='grp2')
> f3 <- rbind(f1,f2)
> 
> G <- function(x, index) {
>     
> # calculate goodman for group 1 bootstrap sample
>     g1 <-x[index,][x[,3]=='grp1',]
>     goodman_g1 <- cor(data[index,][1,2])
>    
>   # calculate goodman for group 2 bootstrap sample
>     g2 <-x[index,][x[,3]=='grp2',]
>     goodman_g2 <- cor(data[index,][3,4])
>    
>   # calculate difference
>     goodman_g1-goodman_g2
>     }
>   
> 
> # use strata parameter in function boot to resample within each group
> results <- boot(data=f3,statistic=G, strata=as.factor(f3$group),R=2000)
> 
> results
> boot.ci(results)
> ##############################
> 
> 
> 
> Le samedi 4 juin 2022 à 09:31:36 UTC+2, Daniel Nordlund <djnordlund using gmail.com> a écrit :
> 
> 
> 
> 
> 
> On 5/28/2022 11:21 AM, varin sacha via R-help wrote:
>> Dear R-experts,
>>
>> While comparing groups, it is better to assess confidence intervals of those differences rather than comparing confidence intervals for each group.
>> I am trying to calculate the CIs of the difference between the two Cramer's V and not the CI to the estimate of each group’s Cramer's V.
>>
>> Here below my toy R example. There are error messages. Any help would be highly appreciated.
>>
>> ##############################
>> library(questionr)
>> library(boot)
>>
>> gender1<-c("M","F","F","F","M","M","F","F","F","M","M","F","M","M","F","M","M","F","M","F","F","F","M","M","M","F","F","M","M","M","F","M","F","F","F","M","M","F","M","F")
>> color1<-c("blue","green","black","black","green","green","blue","blue","green","black","blue","green","blue","black","black","blue","green","blue","green","black","blue","blue","black","black","green","green","blue","green","black","green","blue","black","black","blue","green","green","green","blue","blue","black")
>>
>> gender2<-c("F","F","F","M","M","F","M","M","M","F","F","M","F","M","F","F","M","M","M","F","M","M","M","F","F","F","M","M","M","F","M","M","M","F","F","F","M","F","F","F")
>> color2<-c("green","blue","black","blue","blue","blue","green","blue","green","black","blue","black","blue","blue","black","blue","blue","green","blue","black","blue","blue","black","black","green","blue","black","green","blue","green","black","blue","black","blue","green","blue","green","green","blue","black")
>>
>> f1=data.frame(gender1,color1)
>> tab1<-table(gender1,color1)
>> e1<-cramer.v(tab1)
>>
>> f2=data.frame(gender2,color2)
>> tab2<-table(gender2,color2)
>> e2<-cramer.v(tab2)
>>
>> f3<-data.frame(e1-e2)
>>
>> cramerdiff=function(x,w){
>> y<-tapply(x[w,1], x[w,2],cramer.v)
>> y[1]-y[2]
>> }
>>
>> results<-boot(data=f3,statistic=cramerdiff,R=2000)
>> results
>>
>> boot.ci(results,type="all")
>> ##############################
>>
>>    
>>
>> ______________________________________________
>> R-help using r-project.org mailing list -- To UNSUBSCRIBE and more, see
>> https://stat.ethz.ch/mailman/listinfo/r-help
>> PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
>> and provide commented, minimal, self-contained, reproducible code.
> 
> I don't know if someone responded offline, but if not, there are a
> couple of problems with your code.   First, the f3 dataframe is not what
> you think it is.  Second, your cramerdiff function isn't going to
> produce the results that you want.
> 
> I would put your data into a single dataframe with a variable
> designating which group data came from.  Then use that variable as the
> strata variable in the boot function to resample within groups.  So
> something like this:
> 
> f1 <- data.frame(gender=gender1,color=color1,group='grp1')
> f2 <- data.frame(gender=gender2,color=color2,group='grp2')
> f3 <- rbind(f1,f2)
> 
> cramerdiff <- function(x, ndx) {
>     # calculate cramer.v for group 1 bootstrap sample
>     g1 <-x[ndx,][x[,3]=='grp1',]
>     cramer_g1 <- cramer.v(table(g1[,1:2]))
>     # calculate cramer.v for group 2 bootstrap sample
>     g2 <-x[ndx,][x[,3]=='grp2',]
>     cramer_g2 <- cramer.v(table(g2[,1:2]))
>     # calculate difference
>     cramer_g1-cramer_g2
>     }
> # use strata parameter in function boot to resample within each group
> results <- boot(data=f3,statistic=cramerdiff,
> strata=as.factor(f3$group),R=2000)
> 
> results
> boot.ci(results)
> 
> 
> 
> Hope this is helpful,
> 
> Dan
>