Hi,
i am trying to learn something about multilevel analysis using a great
"Discovering statistics using R". I constructed some sample data and then
tried to fit a model. Generally model fits well, however when trying to fit
the same model using z-score (standarizded) variables i got an error:
Error in lme.formula(z_wyn ~ z_IQ + Kasa, data = la, random = ~z_IQ | :
nlminb problem, convergence error code = 1
message = iteration limit reached without convergence (10)
Below i attach all the code required to reproduce. My question is : is it a
bug or a feature, in other words: is it normal for a model not to converge
using z-scores ? If so, how to estimate standarized coefficents ? Or maybe
i messed up something to begin with?
Here is what i do:
library(nlme)
# I am creating random data
# df of 900 rows
la <- data.frame(1:900)
# municipality: the variable i want to base my
multilevel analysis on
la$municipality[1:300] <- "Warszawa" # name of
municipality: "Warsaw"
la$municipality[301:600] <- "PÅock" #
la$municipality[601:900] <- "Gniezno" #
# money that each municipality spends on schools
(per student)
la$money_per_municipality[1:300] <- 500 # name of
municipality: "Warsaw"
la$money_per_municipality[301:600] <- 300 #"PÅock" #
la$money_per_municipality[601:900] <- 200
#"Gniezno" #
# In my model, i want to estimate student's
achievments,
#so i pick some random data, but I also want to
#differentiate between the municipalities
la$students_score[la$municipality=="Warszawa"] <-
rnorm(n=nrow(la[la$municipality=="Warszawa",]),mean=33,sd=4)
la$students_score[la$municipality=="PÅock"] <-
rnorm(n=nrow(la[la$municipality=="PÅock",]),mean=22,sd=4)
la$students_score[la$municipality=="Gniezno"]
<-rnorm(n=nrow(la[la$municipality=="Gniezno",]),mean=5,sd=4)
### I assume IQ ~ achievement relation 2x + stderr
la$IQ <- la$students_score * 2 +
rnorm(n=nrow(la),mean=0,sd=30)
with(la,cor(students_score,IQ))
#### [1] 0.6202439
## convert factor/string to integer:
la$mun <- as.integer(factor(la$municipality))
### Z SCORES for IQ and achievements
la$z_IQ <- (la$IQ-mean(la$IQ))/sd(la$IQ)
la$z_sscore <-
(la$students_score-mean(la$students_score))/sd(la$students_score)
## Estimate a model with random slopes based on IQ
for each municipality.
##Again: z_IQ is plain IQ, money_per_municipiality
does not very between municipialities
# THIS runs without an error
studentModel<- lme(z_sscore ~ IQ +
money_per_municipality, data=la,random=~IQ|mun,method="ML")
# THIS causes an error. The only difference is one
variable (z_IQ instead of IQ)
studentModel<- lme(z_sscore ~ z_IQ +
money_per_municipality, data=la,random=~z_IQ|mun,method="ML")
Output again:
Error in lme.formula(z_sscore ~ z_IQ +
money_per_municipality, data = la, :
nlminb problem, convergence error code = 1
message = iteration limit
reached without convergence (10)
Best,
MikoÅaj Hnatiuk, Educational Research Institute, Warsaw, Poland
[[alternative HTML version deleted]]
______________________________________________
R-help@r-project.org mailing list
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.
