Skip to content
Snippets Groups Projects
Commit dd690355 authored by Ben Anderson's avatar Ben Anderson
Browse files

added mixed effects models for CER in R code

parent 46758e5e
No related branches found
No related tags found
No related merge requests found
Show whitespace changes
Inline Side-by-side
Census2022-CER-mixed_model_0910.R 0 → 100644
+ 435
− 0
View file @ dd690355
####################################################################################
## Data Analysis for ESRC Transformative project
## - Using the Commission for Energy Regulation (CER)'s Irish Smart Meter Trial data
## - http://www.ucd.ie/issda/data/commissionforenergyregulationcer/
## This file is used to produce midweek (Tues-Thurs) and weekend (Sat-Sun) mixed models which test the ability of
## various power (electricity consumption) variables to predict the household attributes of interest
## The energy data is for 28 days (4 weeks) of Oct 2009 & 2010
## October 2009 was before the smart meter trials started and
## was also close to the pre-trial survey date
## Comparativ emodels also run for 2010
## For the latest version of this code go to: https://github.com/dataknut/Census2022
## This work was funded by RCUK through the ESRC's Transformative Social Science Programme via the
## "Census 2022: Transforming Small Area Socio-Economic Indicators through 'Big Data'" Project
## - http://gtr.rcuk.ac.uk/project/2D2CD798-4F04-4399-B1AF-D810A233DD21
## - http://www.energy.soton.ac.uk/tag/census2022/
## Copyright (C) 2014 University of Southampton
## Author: Sharon Lin (X.Lin@soton.ac.uk)
## [Energy & Climate Change, Faculty of Engineering & Environment, University of Southampton]
## This program is free software; you can redistribute it and/or modify
## it under the terms of the GNU General Public License as published by
## the Free Software Foundation; either version 2 of the License
## (http://choosealicense.com/licenses/gpl-2.0/), or (at your option) any later version.
## This program is distributed in the hope that it will be useful,
## but WITHOUT ANY WARRANTY; without even the implied warranty of
## MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
## GNU General Public License for more details.
## #YMMV - http://en.wiktionary.org/wiki/YMMV
rm(list=ls())
library(lme4)
## load in energy data
setwd("//soton.ac.uk/ude/PersonalFiles/Users/xl25g12/mydocuments/census2022/CER_analysis/data") ## the file is in working order
## egdata = read.table("OctHH_midwk_long", header=T, stringsAsFactors=FALSE) ## read in R exported txt file, no need sep="t"; long form of mid week data for all Oct 2009 and 2010 data
egdata = read.table("CER_OctHH_wkend_long", header=T, stringsAsFactors=FALSE) ## read in R exported txt file, no need sep="t"; long form of mid week data for all Oct 2009 and 2010 data
eg09 = egdata[egdata$DateOct<365,]
eg10 = egdata[egdata$DateOct>365,]
## load in HH survey data, note survey content for pre-trial and post-trial are different
data1 = read.table("HH09_pretrial.txt",header=T,stringsAsFactors=FALSE,sep="\t") ## read in txt file, need sep="t", HHpresurvey data
HH09FA = read.table("HH_floor_pretrial_survey.txt",header=T,stringsAsFactors=FALSE,sep="\t") ## read in txt file, need sep="t"; pretrial floor area data
## HH10 = read.table("HH10_posttrial.txt",header=T,stringsAsFactors=FALSE,sep="\t") ## read in txt file, need sep="t", HHpresurvey data
HH09 = merge(data1,HH09FA)
## merge energy and HH data together
## data for 09 and 10 separately
mydata09=merge(eg09,HH09)
##mydata10=merge(eg10,HH10)
########## mixed effects models are run on these separate files ###################
## check on mixed model with income variables using 09 data
## make a comparison with 0910 data results using 0910
mydata09$dummy_HHchild = NA
mydata09 = within(mydata09,dummy_HHchild[HHtype_y09==1]<-0)
mydata09 = within(mydata09,dummy_HHchild[HHtype_y09==2]<-0)
mydata09 = within(mydata09,dummy_HHchild[HHtype_y09==3]<-1)
ddply(mydata09, .(dummy_HHchild),nrow)
ddply(mydata09, .(HHtype_y09),nrow)
## make a new dummy vriable column D_unemployed
mydata09$dummy_unemp_R = NA
mydata09 = within(mydata09,dummy_unemp_R[employment_y09==1]<-0)
mydata09 = within(mydata09,dummy_unemp_R[employment_y09==4]<-1)
mydata09 = within(mydata09,dummy_unemp_R[employment_y09==5]<-1)
mydata09 = within(mydata09,dummy_unemp_R[employment_y09==2]<-NA)
mydata09 = within(mydata09,dummy_unemp_R[employment_y09==3]<-NA)
mydata09 = within(mydata09,dummy_unemp_R[employment_y09==6]<-NA)
ddply(mydata09, .(dummy_unemp_R),nrow)
## make a second new dummy vriable column dum_em_mixed, (1-3)
## prepare dummy variable for "number of residents"
mydata09["dum_noResident"] = NA
mydata09 = within(mydata09,dum_noResident[HHtype_y09==1]<-0)
mydata09 = within(mydata09,dum_noResident[HHtype_y09==3&HHchild_y09==1]<-0)
mydata09 = within(mydata09,dum_noResident[HHtype_y09==2&HHadult_y09==1]<-0)
mydata09 = within(mydata09,dum_noResident[HHadult_y09>1]<-1)
mydata09 = within(mydata09,dum_noResident[HHchild_y09>1]<-1)
ddply(mydata09, .(dum_noResident),nrow)
names(mydata09)
## change income to approximated mid ranged numbers
mydata09$income_num = mydata09$income ## add a new column
mydata09 = within(mydata09,income_num[income_num==1]<-7500)
mydata09 = within(mydata09,income_num[income_num==2]<-22500)
mydata09 = within(mydata09,income_num[income_num==3]<-40000)
mydata09 = within(mydata09,income_num[income_num==4]<-62500)
mydata09 = within(mydata09,income_num[income_num==5]<-92500)
mydata09 = within(mydata09,income_num[income_num==6]<-NA)
## change dataframe structure
mydata09 = within(mydata09,HHtype_y09<-as.factor(HHtype_y09))
mydata09 = within(mydata09,dummy_unemp_R<-as.factor(dummy_unemp_R))
mydata09 = within(mydata09,dummy_HHchild<-as.factor(dummy_HHchild))
mydata = mydata09
## the following code are used for weekend data
final =0
for (i in c(11,9,12, 3, 5, 7,16,17) )
{
energy = mydata[,i]
## C1: test = lmer(energy ~ 1 + (1|HHID) + num_resident + log(income_num) + emp_dummy, data=mydata, na.action="na.omit") ## RE on intercept only
## C2: test = lmer(energy ~ 1 + (1|HHID) + num_resident + log(income_num), data=mydata, na.action="na.omit") ## RE on intercept only
## C3: test = lmer( energy ~ 1 + (1|HHID) + num_resident + log(income_num) + HHtype , data=mydata, na.action="na.omit") ## RE on intercept only
## C4: test = lmer( energy ~ 1 + (1|HHID) + num_resident + log(income_num) + dummy_HHchild , data=mydata, na.action="na.omit") ## RE on intercept only
## C5: test = lmer(energy ~ 1 + (1|HHID) + num_resident + log(income_num) + dummy_HHchild + dummy_unemp_R , data=mydata, na.action="na.omit") ## RE on intercept only
## C4: test = lmer(energy ~ 1 + (1|ID) + dum_noResident + log(income_num) + dummy_HHchild, data=mydata, na.action="na.omit") ## RE on intercept only
## C3: test = lmer(energy ~ 1 + (1|ID) + dum_noResident + log(income_num), data=mydata, na.action="na.omit") ## RE on intercept only
## C1:
test = lmer(energy ~ 1 + (1|ID) + dum_noResident + log(income_num)+dummy_unemp_R, data=mydata, na.action="na.omit") ## RE on intercept only
## Extract fixed effect coefficients
out_FE = as.data.frame(cbind(as.data.frame(coef(summary(test)))[,1],as.data.frame(coef(summary(test)))[,3]))
colnames(out_FE)[1] = c(get("i"))
colnames(out_FE)[2] = c("zsta")
## for 1 RE only: Extract random effect varaince and percentage of ICC (intercorrelation between RE in intercept and residuals)
RE1 = as.data.frame(VarCorr(test))$vcov[1] ## variance of HHID ## as.data.frame(VarCorr(test))$sdcor[1] for std. dev.
RE2 = as.data.frame(VarCorr(test))$vcov[2] ## varaince of residual ##
# alternative code
## as.numeric(summary(test)@REmat[1,3]) ## variance of HHID;
## as.numeric(summary(test)@REmat[1,4]) ## std error of HHID
## as.numeric(summary(test)@REmat[2,4]) ## std error of residuals
## k = 2
## tmp11 = RE1 / (RE1 + RE2) ## % of RE effects in HH
## RE3 = as.numeric(format(round(tmp11,k),nsmall=k)) ## % of RE effects in HH
## RE = NA ## REtmp and out are of different length
## out = cbind(out_FE,RE)
## out$RE[1]=RE1
## out$RE[2]=RE2
## out$RE[3]=RE3
testlmer = as.vector(test@pp$X %*% fixef(test)) ## fitted values ## length of fitted values is 4099, no adjustment is made
VarF=var(testlmer)
## VarV = variance (intercept) + varaince(Slope) + 2 cov.correlation * stdev(intercept)stdev(slope) ## variation generated from RE
## RE_sdcor = as.data.frame(VarCorr(test))$sdcor ## stdev and correlation of two RE effects, and stdev of residuals
## VarV = RE_sdcor[1]^2
## VarR = RE_sdcor[2]^2 ## variation from residuals
VarV=RE1
VarR = RE2
R2.m = VarF/(VarF + VarV + VarR) ## Var(FE)/[Var(FE)+Var(RE) + Var(Res)]
R2.c = (VarF+VarV)/(VarF + VarV + VarR)
R2.r = VarR / (VarF + VarV + VarR)
R2 = NA ## REtmp and out are of different length
out = cbind(out_FE,R2)
out$R2[1]=R2.m
out$R2[2]=R2.c
out$R2[3]=R2.r
# ## for 2 REs in the model: extract RE effects from the model
# RE_sdcor = as.data.frame(VarCorr(test))$sdcor ## stdev and correlation of two RE effects, and stdev of residuals
# ## final_FE = cbind(final_FE,out_FE)
# ## ## final_RE = cbind(final_RE,RE_sdcor)
# ## R squared for D1-D5 models, for mixed with RE in intercept and slope
# testlmer = as.vector(test@pp$X %*% fixef(test)) ## fitted values ## length of fitted values is 4099, no adjustment is made
# VarF=var(testlmer)
# ## VarV = variance (intercept) + varaince(Slope) + 2 cov.correlation * stdev(intercept)stdev(slope) ## variation generated from RE
# ## or VarV = RE_sdcor[1]^2 + 2*RE_sdcor[1]*RE_sdcor[2]*RE_sdcor[3] + RE_sdcor[2]^2 ## variation generated from RE
# VarR = RE_sdcor[4]^2 ## variation from residuals
# R2.m = VarF/(VarF + VarV + VarR) ## Var(FE)/[Var(FE)+Var(RE) + Var(Res)]
# R2.c = (VarF+VarV)/(VarF + VarV + VarR)
# R2.r = VarR / (VarF + VarV + VarR)
# R2 = NA ## REtmp and out are of different length
# out = cbind(out_FE,R2)
# out$R2[1]=R2.m
# out$R2[2]=R2.c
# out$R2[3]=R2.r
#
final = cbind(final,out)
}
## RE1 working codes are as above.
###### models with T1 - T2 (without income variables)
final=0
for (i in c(3,7,9,11, 12, 16, 17) ) ## mean, q975, ammean, pktime, base, LF (ECF has been consistent not converging, taken out)
## i = 16, ECF does not converge
{
energy = (mydata09[,i])
## D1: test = lmer(energy ~ num_resident + income_num.c + dummy_unemp_R + (1+income_num.c|HHID) , data=mydata, na.action="na.omit") ## RE on intercept and slope
## D2 test = lmer(energy ~ num_resident + income_num.c + (1+income_num.c|HHID), data=mydata, na.action="na.omit") ## RE on intercept and slope
## D3: test = lmer(energy ~ num_resident + income_num.c + HHtype + (1+income_num.c|HHID) , data=mydata, na.action="na.omit") ## RE on intercept and slope
## D4: test = lmer(energy ~ num_resident + income_num.c + dummy_HHchild + (1+income_num.c|HHID) , data=mydata, na.action="na.omit") ## RE on intercept and slope
## T1 = lmer(energy ~ 1 + dum_noResident + dum_em13_46 + (1|ID), data=mydata0910, na.action="na.omit") ## RE on intercept only
T2 = lmer(energy ~ 1 + dum_noResident + (1|ID), data=mydata09, na.action="na.omit") ## RE on intercept only
## T2a = lmer( energy ~ 1 + dum_noResident + dum_em + (1|ID), data=mydata09, na.action="na.omit") ## RE on intercept only
## T3 = lmer(energy ~ 1 + dum_noResident + dum_HHtype2 + dum_HHtype3 + (1|ID), data=mydata09, na.action="na.omit")
## T4 = lmer( energy ~ 1 + dum_noResident + dum_child + (1|ID), data=mydata09, na.action="na.omit") ## RE on intercept only
## T4a = lmer( energy ~ 1 + dum_child + (1|ID), data=mydata09, na.action="na.omit") ## RE on intercept only
## T5 = lmer( energy ~ 1 + dum_noResident + dum_child + dum_em + (1|ID), data=mydata09, na.action="na.omit") ## RE on intercept only
## T6 = lmer( energy ~ 1 + dum_child + dum_noResident + dum_child*dum_noResident + (1|ID), data=mydata09, na.action="na.omit") ## RE on intercept only
## T7 = lmer( energy ~ 1 + dum_child + dum_noResident + dum_child*dum_noResident + dum_em + (1|ID), data=mydata09, na.action="na.omit") ## RE on intercept only
T2 = T2
## Extract fixed effect coefficients
out = as.data.frame(cbind(as.data.frame(coef(summary(T2)))[,1],as.data.frame(coef(summary(T2)))[,3]))
colnames(out)[1] = c(get("i"))
colnames(out)[2] = c("zsta")
## for 1 RE only: Extract random effect varaince and percentage of ICC (intercorrelation between RE in intercept and residuals)
## command: as.data.frame(VarCorr(test)) gives variance-covariance of the model: as.data.frame(VarCorr(test))$vcov; std.error of the model: as.data.frame(VarCorr(test))$sdcor
RE1 = as.data.frame(VarCorr(T2))$vcov[1] ## variance of RE in the intercept "ID"
RE2 = as.data.frame(VarCorr(T2))$vcov[2] ## varaince of residual ##
testlmer = as.vector(T2@pp$X %*% fixef(T2)) ## fitted values ## length of fitted values is 4099, no adjustment is made
VarF=var(testlmer)
## VarV = variance (intercept) + varaince(Slope) + 2 cov.correlation * stdev(intercept)stdev(slope) ## variation generated from RE
VarV =RE1
VarR = RE2 ## variation from residuals
R2.m = VarF/(VarF + VarV + VarR) ## Var(FE)/[Var(FE)+Var(RE) + Var(Res)]
R2.c = (VarF+VarV)/(VarF + VarV + VarR)
R2.r = VarR / (VarF + VarV + VarR)
out["R2"]=NA
out$R2[1] = R2.m
out$R2[2] = R2.c
out$R2[3] = R2.r
final= cbind(final,out)
}
######## end of mixed effects models on sepearte 09 and 10 periods ###################
################# beginning of mixed models on both Oct 09 and Oct 10 data #######################
######## data for 09 and 10 together #############
HH0910 = read.table("HH_both_survey", header=T, stringsAsFactors=FALSE) ## HH info.
tmpHH09=HH0910[HH0910$index==0,] ## HH info in Oct09
tmpHH10=HH0910[HH0910$index==1,] ## HH info in Oct10
tdata09=merge(eg09,tmpHH09) ## HH and enregy infoin Oct 09 where HH has both info (HH participated in both pre and post-survey)
tdata10=merge(eg10,tmpHH10) ## HH and enregy infoin Oct 10 where HH has both info (HH participated in both pre and post-survey)
mydata0910 = rbind(tdata09,tdata10)
mydata0910["dum_em"]=NA ## in survey data, 1 is employed; 4 is unemployed
mydata0910 = within(mydata0910,dum_em[employment==1]<-0)
mydata0910 = within(mydata0910,dum_em[employment==4]<-1)
ddply(mydata0910, .(dum_em),nrow)
names(mydata0910)
mydata0910["dum_child"] = NA
mydata0910 = within(mydata0910,dum_child[HHtype==1]<-0)
mydata0910 = within(mydata0910,dum_child[HHtype==2]<-0)
mydata0910 = within(mydata0910,dum_child[HHtype==3]<-1)
ddply(mydata0910, .(dum_child),nrow)
ddply(mydata0910, .(HHtype),nrow)
names(mydata0910)
## prepare the new columne, count for the number of resdients (0/1),
mydata0910["dum_noResident"] = NA
mydata0910 = within(mydata0910,dum_noResident[HHtype==1]<-0)
mydata0910 = within(mydata0910,dum_noResident[HHtype==3&HHchild==1]<-0)
mydata0910 = within(mydata0910,dum_noResident[HHtype==2&HHadult==1]<-0)
mydata0910 = within(mydata0910,dum_noResident[HHadult>1]<-1)
mydata0910 = within(mydata0910,dum_noResident[HHchild>1]<-1)
ddply(mydata0910, .(dum_noResident),nrow)
names(mydata0910)
##### prepare for dummy_em, with 1-3 value of employment variable as 0, 4-6 as 1 ######
mydata0910["dum_em13_46"] = NA
mydata0910 = within(mydata0910,dum_em13_46[employment<4]<-0) ## 0 for employed people: 1: employee, 2:sel-employed with employees, 3:self-employed without employees;
mydata0910 = within(mydata0910,dum_em13_46[employment>3]<-1) ## 1 for unemployed people: , 4: unemployed actively seeking work; 5 unemployed not seeking work, 6: retired, 7: carer: looking after relative family)
ddply(mydata0910, .(dum_em13_46),nrow)
## prepare dummy for HHtype, HHtype2 = 1 if HHtype = 2, otherwise 0; HHtype3 = 1 if HHtype = 3, otherwise 0;
mydata0910["dum_HHtype2"] = NA
mydata0910 = within(mydata0910,dum_HHtype2[HHtype==2]<-1) ## 0 for employed people: 1: employee, 2:sel-employed with employees, 3:self-employed without employees;
mydata0910 = within(mydata0910,dum_HHtype2[HHtype==1]<-0) ## 0 for employed people: 1: employee, 2:sel-employed with employees, 3:self-employed without employees;
mydata0910 = within(mydata0910,dum_HHtype2[HHtype==3]<-0) ## 0 for employed people: 1: employee, 2:sel-employed with employees, 3:self-employed without employees;
ddply(mydata0910, .(dum_HHtype2),nrow)
mydata0910["dum_HHtype3"] = NA
mydata0910 = within(mydata0910,dum_HHtype3[HHtype==3]<-1) ## 0 for employed people: 1: employee, 2:sel-employed with employees, 3:self-employed without employees;
mydata0910 = within(mydata0910,dum_HHtype3[HHtype==1]<-0) ## 0 for employed people: 1: employee, 2:sel-employed with employees, 3:self-employed without employees;
mydata0910 = within(mydata0910,dum_HHtype3[HHtype==2]<-0) ## 0 for employed people: 1: employee, 2:sel-employed with employees, 3:self-employed without employees;
ddply(mydata0910, .(dum_HHtype3),nrow)
## change dataframe structure
mydata0910 = within(mydata0910,MF<-as.factor(MF))
mydata0910 = within(mydata0910,HHtype<-as.factor(HHtype))
mydata0910 = within(mydata0910,age<-as.factor(age))
mydata0910 = within(mydata0910,employment<-as.factor(employment))
mydata0910 = within(mydata0910,socialclass<-as.factor(socialclass))
mydata0910 = within(mydata0910,HHadult<-as.factor(HHadult))
mydata0910 = within(mydata0910,dayault<-as.factor(dayault))
mydata0910 = within(mydata0910,dum_child<-as.factor(dum_child))
mydata0910 = within(mydata0910,daychild<-as.factor(daychild))
## mydata0910 = within(mydata0910,bedroom<-as.factor(bedroom))
## mydata0910 = within(mydata0910,heating<-as.factor(heating))
## mydata0910 = within(mydata0910,income<-as.factor(income))
mydata0910 = within(mydata0910,dum_em<-as.factor(dum_em))
mydata0910 = within(mydata0910,dum_HHtype2<-as.factor(dum_HHtype2))
mydata0910 = within(mydata0910,dum_HHtype3<-as.factor(dum_HHtype3))
mydata0910 = within(mydata0910,ID<-as.factor(ID))
mydata0910 = within(mydata0910,dum_child<-as.factor(dum_child))
mydata0910 = within(mydata0910,dum_noResident<-as.factor(dum_noResident))
mydata0910 = within(mydata0910,dum_em13_46<-as.factor(dum_em13_46))
########## mixed models for RE effect in the intercept ############################
mydata = mydata0910[order(mydata0910$ID),] ## sort data accoring to HH ID.
A1 = lmer( dailyammean ~ 1 + (1|ID), data=mydata0910, na.action="na.omit") ## RE on intercept only
A2 = lmer( dailyammean ~ 1 + dum_noResident + (1|ID), data=mydata0910, na.action="na.omit") ## RE on intercept only
A3 = lmer( dailyammean ~ 1 + dum_noResident + dum_em + (1|ID), data=mydata0910, na.action="na.omit") ## RE on intercept only
A4 = lmer( dailyammean ~ 1 + dum_child + (1|ID), data=mydata0910, na.action="na.omit") ## RE on intercept only
A5 = lmer( dailyammean ~ 1 + dum_child + dum_em + (1|ID), data=mydata0910, na.action="na.omit") ## RE on intercept only
A6 = lmer( dailyammean ~ 1 + dum_child + dum_noResident + dum_child*dum_noResident + (1|ID), data=mydata0910, na.action="na.omit") ## RE on intercept only
A7 = lmer( dailyammean ~ 1 + dum_child + dum_noResident + dum_child*dum_noResident + dum_em + (1|ID), data=mydata0910, na.action="na.omit") ## RE on intercept only
############### automation to extract tables from RE ##########################
########## loop through one energy variable at a time ##########
final=0
for (i in c(3,7,9,11, 12, 17) ) ## mean, q975, ammean, pktime, base, LF (ECF has been consistent not converging, taken out)
## i = 16, ECF does not converge
{
energy = (mydata0910[,i])
## D1: test = lmer(energy ~ num_resident + income_num.c + dummy_unemp_R + (1+income_num.c|HHID) , data=mydata, na.action="na.omit") ## RE on intercept and slope
## D2 test = lmer(energy ~ num_resident + income_num.c + (1+income_num.c|HHID), data=mydata, na.action="na.omit") ## RE on intercept and slope
## D3: test = lmer(energy ~ num_resident + income_num.c + HHtype + (1+income_num.c|HHID) , data=mydata, na.action="na.omit") ## RE on intercept and slope
## D4: test = lmer(energy ~ num_resident + income_num.c + dummy_HHchild + (1+income_num.c|HHID) , data=mydata, na.action="na.omit") ## RE on intercept and slope
## T1 = lmer(energy ~ 1 + dum_noResident + dum_em13_46 + (1|ID), data=mydata0910, na.action="na.omit") ## RE on intercept only
## T2 = lmer(energy ~ 1 + dum_noResident + (1|ID), data=mydata0910, na.action="na.omit") ## RE on intercept only
## T2a = lmer( energy ~ 1 + dum_noResident + dum_em + (1|ID), data=mydata0910, na.action="na.omit") ## RE on intercept only
## T3 = lmer(energy ~ 1 + dum_noResident + dum_HHtype2 + dum_HHtype3 + (1|ID), data=mydata0910, na.action="na.omit")
T4 = lmer( energy ~ 1 + dum_noResident + dum_child + (1|ID), data=mydata0910, na.action="na.omit") ## RE on intercept only
## T4a = lmer( energy ~ 1 + dum_child + (1|ID), data=mydata0910, na.action="na.omit") ## RE on intercept only
T5 = lmer( energy ~ 1 + dum_noResident + dum_child + dum_em + (1|ID), data=mydata0910, na.action="na.omit") ## RE on intercept only
## T6 = lmer( energy ~ 1 + dum_child + dum_noResident + dum_child*dum_noResident + (1|ID), data=mydata0910, na.action="na.omit") ## RE on intercept only
## T7 = lmer( energy ~ 1 + dum_child + dum_noResident + dum_child*dum_noResident + dum_em + (1|ID), data=mydata0910, na.action="na.omit") ## RE on intercept only
T2 = T5
## Extract fixed effect coefficients
out = as.data.frame(cbind(as.data.frame(coef(summary(T2)))[,1],as.data.frame(coef(summary(T2)))[,3]))
colnames(out)[1] = c(get("i"))
colnames(out)[2] = c("zsta")
## for 1 RE only: Extract random effect varaince and percentage of ICC (intercorrelation between RE in intercept and residuals)
## command: as.data.frame(VarCorr(test)) gives variance-covariance of the model: as.data.frame(VarCorr(test))$vcov; std.error of the model: as.data.frame(VarCorr(test))$sdcor
RE1 = as.data.frame(VarCorr(T2))$vcov[1] ## variance of RE in the intercept "ID"
RE2 = as.data.frame(VarCorr(T2))$vcov[2] ## varaince of residual ##
testlmer = as.vector(T2@pp$X %*% fixef(T2)) ## fitted values ## length of fitted values is 4099, no adjustment is made
VarF=var(testlmer)
## VarV = variance (intercept) + varaince(Slope) + 2 cov.correlation * stdev(intercept)stdev(slope) ## variation generated from RE
VarV =RE1
VarR = RE2 ## variation from residuals
R2.m = VarF/(VarF + VarV + VarR) ## Var(FE)/[Var(FE)+Var(RE) + Var(Res)]
R2.c = (VarF+VarV)/(VarF + VarV + VarR)
R2.r = VarR / (VarF + VarV + VarR)
out["R2"]=NA
out$R2[1] = R2.m
out$R2[2] = R2.c
out$R2[3] = R2.r
final= cbind(final,out)
}
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Please register or to comment