The DHS Program User Forum
Discussions regarding The DHS Program data and results
Home » Countries » India » k fold cross validation for logistic regression in R (stuck up using cross validation using surveyCV package)
k fold cross validation for logistic regression in R [message #26199] Thu, 16 February 2023 03:53 Go to next message
dhivvyajp@am.amrita.edu is currently offline  dhivvyajp@am.amrita.edu
Messages: 4
Registered: January 2023
Member
Dear Experts,

I am working on DHS7 dataset. I was able to do logistic regression for 70% training and 30% testing data. But When I am trying to do k fold cross validation instead of 70: 30 split up, came across surveyCV package. I am getting the following error. Kindly let me know how can I fix this issue.
> set.seed(2023)
> svylogistic <- svyglm(formula = InternetUsage~RuralOrUrban+AgeGroup+WealthIndex+SchoolingCom pleted+Religion+Caste+MaritalStatus+Occupation+Gender+Litera cy+OwnsMobile, design=my_design, family=quasibinomial())
> cv.svyglm(svylogistic, nfolds=3, na.rm = FALSE)
Error in if (clusterID %in% c("0", "1")) { : the condition has length > 1


Re: k fold cross validation for logistic regression in R [message #26205 is a reply to message #26199] Thu, 16 February 2023 08:25 Go to previous messageGo to next message
Bridgette-DHS is currently offline  Bridgette-DHS
Messages: 3199
Registered: February 2013
Senior Member

Following is a response from Senior DHS staff member, Tom Pullum:

If you were using Stata, I would want to be sure that your svyset command includes "singleunit(centered)" (or one of the other options in the parentheses). The India surveys are huge, but it is still possible that singleunit is needed when you are dividing the sample up. I hope R has something equivalent to singleunit (I do not use R myself). Beyond that, I have no suggestions. Other users may be able to help.
Re: k fold cross validation for logistic regression in R [message #26409 is a reply to message #26199] Fri, 17 March 2023 03:45 Go to previous messageGo to next message
dhivvyajp@am.amrita.edu is currently offline  dhivvyajp@am.amrita.edu
Messages: 4
Registered: January 2023
Member
I tried the following also. But not able to fix the error.

> cv.svy(train, formulae = " InternetUsage~RuralOrUrban+AgeGroup+WealthIndex+SchoolingCom pleted+Religion+Caste+MaritalStatus+Occupation+Gender+Litera cy+OwnsMobile ", method = "logistic", nfolds=3, strataID = train$strata, clusterID = train$Cluster, nest = T, weightsID = train$samplewt)
....................................Error in .subset2(x, i, exact = exact) : no such index at level 1[/color]
> cv.svy(train, formulae = " InternetUsage~RuralOrUrban+AgeGroup+WealthIndex+SchoolingCom pleted+Religion+Caste+MaritalStatus+Occupation+Gender+Litera cy+OwnsMobile ", method = "logistic", nfolds=3, strataID = train$strata, clusterID = train$Cluster, nest = F, weightsID = train$samplewt)
....................................Error in .subset2(x, i, exact = exact) : no such index at level 1
> cv.svyglm(svylogistic, nfolds=3,na.rm=FALSE)
....................................Error in if (clusterID %in% c("0", "1")) { : the condition has length > 1

Can anyone help me in fixing this error?
Re: k fold cross validation for logistic regression in R [message #26502 is a reply to message #26409] Mon, 27 March 2023 08:05 Go to previous message
Bridgette-DHS is currently offline  Bridgette-DHS
Messages: 3199
Registered: February 2013
Senior Member
Following is a response from DHS Senior Analysis & Research Manager, Shireen Assaf:

# install and load the packages you need
install.packages("survey")
library(survey)

# setting your survey design
# To identify the survey design, you need three variables: weight, psu, and strata

# creating the sampling weight variable. 
IRdata$wt <- IRdata$v005/1000000

mysurvey<-svydesign(id=IRdata$v021, data=IRdata, strata=IRdata$v022,  weight=IRdata$wt, nest=T)
options(survey.lonely.psu="adjust")

#now you can use the svy commands in the survey package and use the "mysurvey" sample design object. Check the commands you can use in the survey package. 

# for example

# table of variable v313 (FP use) this is after you attach your data. You can also use svyby, svyglm, etc. 

svytable(~v313, mysurvey)


[Updated on: Mon, 27 March 2023 08:05]

Report message to a moderator

Previous Topic: Fertility preference (table 4.19.1)
Next Topic: Norms variables [[URGENT]]
Goto Forum:
  


Current Time: Sat Nov 30 02:51:45 Coordinated Universal Time 2024