Hey,
I want to do research on the question how a climate of psychological safety emerges. Therefore I collected data of 2 groups over 9 measurment times. Now I want to assess the presence of consensus within my data. Therefore I found the text of Lang et. al. 2019, who suggest and develop the consensus emergence model (CEM) and even supply a "r" code for calulating it. Since I wasn't able to run the r code with r studio and the authors state, that the model can be easily executed in stata I want to ask for help to use the CEM.
Maybe one of you can help me to transform the r code into a stata code, taking into account that I have 2 groups, 7 variables for psychological safety and 9 measurement times? Thanks a lot in advance.
This is the "r" code written in the study:
R code for the army data
library(nlme)
#################### PREPARE DATA #######################
## get and prepare data
library(multilevel)
data(univbct)
#prepare variables
univbct2<-univbct
univbct2$UNIT<-paste(univbct2$BTN,univbct2$COMPANY,sep="")
RMEANS<-aggregate(READY1 ~ UNIT,univbct2[univbct2$TIME==0,],mean)
names(RMEANS)<-c("UNIT","CREAD")
univbct2<-merge(univbct2,RMEANS,by="UNIT")
#delete persons with single observations
univbct2<-univbct2[rowSums(sapply(subset(univbct2,select=c(JOBSAT1,JO BSAT2,JOBSAT3)),is.na))<2,]
# at least three
members<-table(univbct2$UNIT)/3
univbct2<-univbct2[univbct2$UNIT %in% names(members[members>2]), ]
################### ANALYSIS ##############################
## control settings for lme
csettings=lmeControl(opt="nlminb") # default setting
#alternative
#csettings=lmeControl(maxIter=3000,msMaxIter=3000, opt="optim",optimMethod="Nelder-Mead")
# 3-Level null model. The higher-level unit identifier is UNIT, the person identifier is SUBNUM
# The identifier used first in the list command refers to the highest level (here level-3).
# pdSymm specifies the nature of the random effects covariance matrix and refers to a symmetric
# matrix
# univbct2 is the dataset
M2a<-lme(JSAT ~ TIME, random = list(UNIT=pdSymm(~TIME),SUBNUM=pdSymm(~1)),
data = univbct2,na.action=na.omit,control=csettings)
# 3-level CEM model. Change in the residual variance over time is added to the null model
M2b<-update(M2a,weights=varExp( form = ~ TIME))
anova(M2a,M2b)
#view additional model details with summary()
summary(M2b)
# 3-level with standardized group-level predictor
# scale(..) centers and standardizes the predictor
# scale(..,scale=FALSE) centers but does not standardize the predictor
# each predictor in the residual variance part of the model needs a separate varExp command
# otherwise R just estimates one coefficient for all predictors
m3a<-lme(JSAT ~ TIME*scale(CREAD), random = list(UNIT=pdSymm(~TIME),SUBNUM=pdSymm(~1)),
data = univbct2,na.action=na.omit,control=csettings,weigh ts=varExp( form = ~ TIME))
m3b<-update(m3a,weights=varComb(varExp( form = ~ TIME),varExp( form = ~ scale(CREAD))))
m3c<-update(m3b,weights=varComb(varExp( form = ~ TIME),varExp( form = ~ scale(CREAD)),
varExp( form = ~ scale(CREAD)*TIME)))
anova(m3a,m3b,m3c)
# 2-level with unstandardized group-level predictor
# scale(..,scale=FALSE) centers but does not standardize the predictor
m3a2<-lme(JSAT ~ TIME*scale(CREAD,scale=FALSE), random = list(UNIT=pdSymm(~TIME)),
data = univbct2,na.action=na.omit,control=csettings,weigh ts=varExp( form = ~ TIME))
m3b2<-update(m3a2,weights=varComb(varExp( form = ~ TIME),varExp( form = ~
scale(CREAD,scale=FALSE))))
m3c2<-update(m3b2,weights=varComb(varExp( form = ~ TIME),varExp( form = ~
scale(CREAD,scale=FALSE)),
varExp( form = ~ scale(CREAD,scale=FALSE)*TIME)))
anova(m3a2,m3b2,m3c2)
Literature:
Lang, J. W., Bliese, P. D., & de Voogt, A. (2018). Modeling consensus emergence in groups using longitudinal multilevel methods. Personnel Psychology, 71(2), 255-281.
I want to do research on the question how a climate of psychological safety emerges. Therefore I collected data of 2 groups over 9 measurment times. Now I want to assess the presence of consensus within my data. Therefore I found the text of Lang et. al. 2019, who suggest and develop the consensus emergence model (CEM) and even supply a "r" code for calulating it. Since I wasn't able to run the r code with r studio and the authors state, that the model can be easily executed in stata I want to ask for help to use the CEM.
Maybe one of you can help me to transform the r code into a stata code, taking into account that I have 2 groups, 7 variables for psychological safety and 9 measurement times? Thanks a lot in advance.
This is the "r" code written in the study:
R code for the army data
library(nlme)
#################### PREPARE DATA #######################
## get and prepare data
library(multilevel)
data(univbct)
#prepare variables
univbct2<-univbct
univbct2$UNIT<-paste(univbct2$BTN,univbct2$COMPANY,sep="")
RMEANS<-aggregate(READY1 ~ UNIT,univbct2[univbct2$TIME==0,],mean)
names(RMEANS)<-c("UNIT","CREAD")
univbct2<-merge(univbct2,RMEANS,by="UNIT")
#delete persons with single observations
univbct2<-univbct2[rowSums(sapply(subset(univbct2,select=c(JOBSAT1,JO BSAT2,JOBSAT3)),is.na))<2,]
# at least three
members<-table(univbct2$UNIT)/3
univbct2<-univbct2[univbct2$UNIT %in% names(members[members>2]), ]
################### ANALYSIS ##############################
## control settings for lme
csettings=lmeControl(opt="nlminb") # default setting
#alternative
#csettings=lmeControl(maxIter=3000,msMaxIter=3000, opt="optim",optimMethod="Nelder-Mead")
# 3-Level null model. The higher-level unit identifier is UNIT, the person identifier is SUBNUM
# The identifier used first in the list command refers to the highest level (here level-3).
# pdSymm specifies the nature of the random effects covariance matrix and refers to a symmetric
# matrix
# univbct2 is the dataset
M2a<-lme(JSAT ~ TIME, random = list(UNIT=pdSymm(~TIME),SUBNUM=pdSymm(~1)),
data = univbct2,na.action=na.omit,control=csettings)
# 3-level CEM model. Change in the residual variance over time is added to the null model
M2b<-update(M2a,weights=varExp( form = ~ TIME))
anova(M2a,M2b)
#view additional model details with summary()
summary(M2b)
# 3-level with standardized group-level predictor
# scale(..) centers and standardizes the predictor
# scale(..,scale=FALSE) centers but does not standardize the predictor
# each predictor in the residual variance part of the model needs a separate varExp command
# otherwise R just estimates one coefficient for all predictors
m3a<-lme(JSAT ~ TIME*scale(CREAD), random = list(UNIT=pdSymm(~TIME),SUBNUM=pdSymm(~1)),
data = univbct2,na.action=na.omit,control=csettings,weigh ts=varExp( form = ~ TIME))
m3b<-update(m3a,weights=varComb(varExp( form = ~ TIME),varExp( form = ~ scale(CREAD))))
m3c<-update(m3b,weights=varComb(varExp( form = ~ TIME),varExp( form = ~ scale(CREAD)),
varExp( form = ~ scale(CREAD)*TIME)))
anova(m3a,m3b,m3c)
# 2-level with unstandardized group-level predictor
# scale(..,scale=FALSE) centers but does not standardize the predictor
m3a2<-lme(JSAT ~ TIME*scale(CREAD,scale=FALSE), random = list(UNIT=pdSymm(~TIME)),
data = univbct2,na.action=na.omit,control=csettings,weigh ts=varExp( form = ~ TIME))
m3b2<-update(m3a2,weights=varComb(varExp( form = ~ TIME),varExp( form = ~
scale(CREAD,scale=FALSE))))
m3c2<-update(m3b2,weights=varComb(varExp( form = ~ TIME),varExp( form = ~
scale(CREAD,scale=FALSE)),
varExp( form = ~ scale(CREAD,scale=FALSE)*TIME)))
anova(m3a2,m3b2,m3c2)
Literature:
Lang, J. W., Bliese, P. D., & de Voogt, A. (2018). Modeling consensus emergence in groups using longitudinal multilevel methods. Personnel Psychology, 71(2), 255-281.
