- Rotina no Markdown: MODELO_LINEAR_MISTO_TRANS_NOVO
require(tidyverse)
require(lme4)
require(redres) #residuos condicionais
require(geoR)
require(lmerTest) #exibe os valor-p no modelo lmer
require(gtools) #função combinations
require(performance) #função R2_nakagawa
require(buildmer) #função converged
require(DT) #função datatable
- Utilizando o scale para que a média seja igual a 0 (colocar as variaveis na mesma escala para auxiliar na convergencia)
dados = read.table("banco.csv",sep=";",dec=",",header=T)
l1=0.2953672 #este valor foi encontrado no boxcoxfit abaixo:
l2=0
matrix=dados %>%
dplyr::select(ano,raz_2020,pbf,jovens_14_24,tx_desocup,tx_pres,perc_hom,dens1) %>%
as.matrix()
bc <- boxcoxfit(dados$tx_latrc,matrix, lambda2 = TRUE)
l1 <- bc$lambda[1]
l2 <- bc$lambda[2]
l1
## lambda
## 0.2953672
l2
## lambda2
## 0
vars
## function (...)
## {
## quos(...)
## }
## <bytecode: 0x0000000018fe0448>
## <environment: namespace:dplyr>
#Box cox
#http://www.biostat.jhsph.edu/~iruczins/teaching/jf/ch8.pdf
dados1 = dados %>%
mutate(
ytrans = ((tx_latrc + l2)^l1-1)/1, #Box Cox - não utilizado
ylog = log(tx_latrc),
l_latrocinio=log(latrocinio),
s_gini=scale(gini_ibge),
s_pbf=scale(pbf),
s_tx_desligamentos = scale(tx_desligamentos),
s_dens1 = scale(dens1),
s_jov_14_24 = scale(jovens_14_24),
s_raz_2020=scale(raz_2020),
s_raz_1040 = scale(raz_1040),
s_tx_pres=scale(tx_pres),
s_perc_hom = scale(perc_hom),
s_tx_desocup=scale(tx_desocup),
ano_2 = ano^2,
pop_ativa_cont=pop*pop_ativa,
ano1=poly(ano,1)[,1]) #transformação de ano em polinomio ortogonal
hist(dados1$ytrans)
plot(dados1$tx_latrc,dados1$ytrans,type='l')
lines(dados1$tx_latrc,log(dados1$tx_latrc),type='l',col=2)
shapiro.test(dados1$ytrans)
##
## Shapiro-Wilk normality test
##
## data: dados1$ytrans
## W = 0.9935, p-value = 0.4672
shapiro.test(log(dados1$tx_latrc))
##
## Shapiro-Wilk normality test
##
## data: log(dados1$tx_latrc)
## W = 0.98967, p-value = 0.1244
hist(log(dados1$tx_latrc))
qqnorm(log(dados1$tx_latrc))
qqline(log(dados1$tx_latrc))
Modelo linear com efeitos mistos
- sem fazer o scale(), que significa padronizar as variáveis, o algoritmo não converge
- para a variável ano, foi utilizado a transformação poly(ano,1), que significa criar um polinômio ortogonal, também para facilitar a convergência do algoritmo.
- A rotina abaixo consiste em testar todas as combinações de variáveis,
- Também poderia utilizar o método stepwise com critério AIC - disponíveis nos pacotes buildmer & stepcAIC, no entanto, tivemos problemas pois algumas combinações não apresentam convergência e fazem o algoritmo parar,
- pendente pesquisar melhor sobre os pacotes que executam stepwise para modelos mistos.
- Optou-se por não exibir os modelos com taxa de presos junto a proporção de homens, por serem muito correlacionadas;
formula=c()
modelos=list()
AIC=c()
R2=c()
matriz_singular=c()
convergencia=c()
resultados=c()
shapiro_res=c()
#shapiro_ef=c()
base=c("ytrans ~ ano + (1 | UF)", #os estados diferem por uma constante
"ytrans ~ ano + (ano | UF)", #os estados diferem por uma constante + coeficiente linear no tempo
"ytrans ~ poly(ano,2) + (poly(ano,2) | UF)", #os estados diferem por uma constante + coeficiente linear no tempo + coeficiente quadrático no tempo
"ytrans ~ poly(ano,3) + (poly(ano,3) | UF)")
base="ytrans ~ poly(ano,1) + (poly(ano,1) | UF)"
base="ylog ~ poly(ano,1) + (poly(ano,1) | UF)"
base="l_latrocinio ~ poly(ano,1) + (poly(ano,1) | UF)"
#base="l_latrocinio ~ ano + (ano | UF)"
vars=c("+ s_raz_2020","+ s_pbf","+ s_jov_14_24","+ s_tx_desocup","+ s_tx_pres","+ s_perc_hom", "+ s_dens1","+ s_tx_desligamentos","+ s_gini", "+ s_raz_1040")
#vars=c("+ s_raz_2020","+ s_pbf","+ s_jov_14_24","+ s_tx_desocup","+ s_tx_pres","+ s_perc_hom")
#vars=c("+ raz_2020","+ pbf","+ jovens_14_24","+ tx_desocup","+ tx_pres","+ perc_hom", "+ dens1")
aux=c()
aux1=c()
aux2=c()
aux3=c()
aux4=c()
nvar=1
f=list()
while(nvar<=length(vars)){
c=combinations(n=length(vars),r=nvar,vars,repeats.allowed=FALSE)
f[[nvar]]=paste0(base,apply(c,1,paste,collapse=""))
#aux[[nvar]]=
nvar=nvar+1
}
f=unlist(f)
i=1
while(i<=length(f)){
aux[i]=ifelse(length(grep(pattern="tx_desligamentos",x=f[i]))>0 & length(grep(pattern="tx_desocup",x=f[i]))>0,1,0) #escolher qual variável utilizar, não pode entrar as duas ao mesmo tempo.
aux1[i]=ifelse(length(grep(pattern="raz_2020",x=f[i]))>0 & length(grep(pattern="raz_1040",x=f[i]))>0,1,0)
aux2[i]=ifelse(length(grep(pattern="raz_2020",x=f[i]))>0 & length(grep(pattern="gini",x=f[i]))>0,1,0)
aux3[i]=ifelse(length(grep(pattern="raz_1040",x=f[i]))>0 & length(grep(pattern="gini",x=f[i]))>0,1,0)
aux4[i]=ifelse(length(grep(pattern="perc_hom",x=f[i]))>0 & length(grep(pattern="tx_pres",x=f[i]))>0,1,0)
i=i+1
}
formula=data.frame(f,aux,aux1,aux2,aux3)%>%
filter(aux+aux1+aux2+aux3+aux4==0)%>%
select(f)%>%
as.matrix()
head(formula)
## f
## [1,] "l_latrocinio ~ poly(ano,1) + (poly(ano,1) | UF)+ s_dens1"
## [2,] "l_latrocinio ~ poly(ano,1) + (poly(ano,1) | UF)+ s_gini"
## [3,] "l_latrocinio ~ poly(ano,1) + (poly(ano,1) | UF)+ s_jov_14_24"
## [4,] "l_latrocinio ~ poly(ano,1) + (poly(ano,1) | UF)+ s_pbf"
## [5,] "l_latrocinio ~ poly(ano,1) + (poly(ano,1) | UF)+ s_perc_hom"
## [6,] "l_latrocinio ~ poly(ano,1) + (poly(ano,1) | UF)+ s_raz_1040"
i=1
modelos=list()
while(i<=length(formula)){
# while(i<=2){
modelos[[i]]=lmer(formula[i],data=dados1,offset=log(pop_ativa_cont))
AIC[i]=AIC(modelos[[i]])
R2[i]=r2_nakagawa(modelos[[i]])$R2_marginal
matriz_singular[i]=isSingular(modelos[[i]])
convergencia[i]=converged(modelos[[i]])[1]
shapiro_res[i]=shapiro.test(compute_redres(modelos[[i]]))[2]
resultados=rbind(resultados,c(i,formula[i],AIC[i],R2[i],matriz_singular[i],convergencia[i],shapiro_res[i]))
i=i+1
}
## Loading required namespace: testthat
resultados = resultados%>%
data.frame()%>%
`colnames<-`(c("i","modelo","AIC","R2","matriz_singular","convergencia","shapiro_res"))
resultados1=resultados%>%
filter(matriz_singular=="FALSE")%>%
filter(convergencia=="TRUE")%>%
dplyr::select(-convergencia)%>%
dplyr::select(-matriz_singular)%>%
dplyr::select(-modelo)%>%
filter(shapiro_res>0.1)
datatable(resultados1)
Top 5 modelos pelo criterio R2 marginal (os maiores R2)
- Nos modelos onde entram PBF e pobreza, temos uma interpretação oposta (quando cresce bolsa familia cresce os latrocinios); quando cresce pobreza , decresce os latrocinios- verificar o porque;
- o pbf está muito correlacionado com densidade urbana! Justificar no relatorio
- o R2 marginal é quanto os efeitos fixos explicam
- o R2 condicional é quanto os efeitos fixos + aleatorios explicam
i=175
#https://easystats.github.io/performance/reference/r2_nakagawa.html
formula[[i]]
## [1] "l_latrocinio ~ poly(ano,1) + (poly(ano,1) | UF)+ s_dens1+ s_pbf+ s_tx_desocup+ s_tx_pres"
summary(modelos[[i]])
## Linear mixed model fit by REML. t-tests use Satterthwaite's method [
## lmerModLmerTest]
## Formula: formula[i]
## Data: dados1
## Offset: log(pop_ativa_cont)
##
## REML criterion at convergence: 184
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -2.96545 -0.55343 0.00567 0.55299 2.65445
##
## Random effects:
## Groups Name Variance Std.Dev. Corr
## UF (Intercept) 0.09776 0.3127
## poly(ano, 1) 8.05749 2.8386 -0.17
## Residual 0.07932 0.2816
## Number of obs: 216, groups: UF, 27
##
## Fixed effects:
## Estimate Std. Error df t value Pr(>|t|)
## (Intercept) -11.01015 0.06315 23.68440 -174.350 < 2e-16 ***
## poly(ano, 1) -1.83419 0.73459 49.97823 -2.497 0.0159 *
## s_dens1 -0.15536 0.08372 32.74901 -1.856 0.0725 .
## s_pbf 0.19344 0.07465 35.58896 2.591 0.0138 *
## s_tx_desocup 0.23228 0.04242 190.72832 5.476 1.36e-07 ***
## s_tx_pres 0.06982 0.05042 92.87362 1.385 0.1694
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr) pl(,1) s_dns1 s_pbf s_tx_d
## poly(ano,1) -0.123
## s_dens1 0.000 0.161
## s_pbf 0.000 -0.004 -0.558
## s_tx_desocp 0.000 -0.502 -0.291 0.092
## s_tx_pres 0.000 -0.179 0.182 0.117 0.000
r2_nakagawa(modelos[[i]])
## # R2 for Mixed Models
##
## Conditional R2: 0.708
## Marginal R2: 0.212
converged(modelos[[i]])[1]
## [1] TRUE
- Análise de resíduos condicionais para o modelo i:
shapiro.test(compute_redres(modelos[[i]]))[2]
## $p.value
## [1] 0.8962107
plot(compute_redres(modelos[[i]]),main="resíduos condicionais versus índices")
plot_resqq(modelos[[i]])
- Análise dos efeitos aleatórios para o modelo i:
random <- ranef(modelos[[i]])
aleatorio1 = random[["UF"]][["(Intercept)"]]
aleatorio2 = random[["UF"]][[2]] #É o termo do poly(ano,1)
plot(aleatorio1,main="efeitos aleatórios versus índices")
plot(aleatorio2,main="efeitos aleatórios versus índices")
plot_ranef(modelos[[i]])
shapiro.test(aleatorio1)
##
## Shapiro-Wilk normality test
##
## data: aleatorio1
## W = 0.98337, p-value = 0.9292
shapiro.test(aleatorio2)
##
## Shapiro-Wilk normality test
##
## data: aleatorio2
## W = 0.94053, p-value = 0.1256
i=167
#https://easystats.github.io/performance/reference/r2_nakagawa.html
formula[[i]]
## [1] "l_latrocinio ~ poly(ano,1) + (poly(ano,1) | UF)+ s_dens1+ s_pbf+ s_perc_hom+ s_tx_desocup"
summary(modelos[[i]])
## Linear mixed model fit by REML. t-tests use Satterthwaite's method [
## lmerModLmerTest]
## Formula: formula[i]
## Data: dados1
## Offset: log(pop_ativa_cont)
##
## REML criterion at convergence: 184
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -3.01229 -0.55097 0.00157 0.57054 2.78089
##
## Random effects:
## Groups Name Variance Std.Dev. Corr
## UF (Intercept) 0.09188 0.3031
## poly(ano, 1) 8.11907 2.8494 -0.19
## Residual 0.07983 0.2825
## Number of obs: 216, groups: UF, 27
##
## Fixed effects:
## Estimate Std. Error df t value Pr(>|t|)
## (Intercept) -11.01015 0.06142 24.18334 -179.256 < 2e-16 ***
## poly(ano, 1) -1.52564 0.72960 47.86471 -2.091 0.0419 *
## s_dens1 -0.11614 0.09086 46.30211 -1.278 0.2076
## s_pbf 0.13982 0.07801 44.91533 1.792 0.0798 .
## s_perc_hom 0.06839 0.04720 189.56117 1.449 0.1490
## s_tx_desocup 0.23053 0.04236 189.83421 5.442 1.61e-07 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr) pl(,1) s_dns1 s_pbf s_prc_
## poly(ano,1) -0.133
## s_dens1 0.000 0.231
## s_pbf 0.000 -0.027 -0.661
## s_perc_hom 0.000 0.120 0.466 -0.372
## s_tx_desocp 0.000 -0.508 -0.277 0.089 -0.029
r2_nakagawa(modelos[[i]])
## # R2 for Mixed Models
##
## Conditional R2: 0.697
## Marginal R2: 0.205
i=247
#https://easystats.github.io/performance/reference/r2_nakagawa.html
formula[[i]]
## [1] "l_latrocinio ~ poly(ano,1) + (poly(ano,1) | UF)+ s_dens1+ s_jov_14_24+ s_pbf+ s_tx_desocup+ s_tx_pres"
summary(modelos[[i]])
## Linear mixed model fit by REML. t-tests use Satterthwaite's method [
## lmerModLmerTest]
## Formula: formula[i]
## Data: dados1
## Offset: log(pop_ativa_cont)
##
## REML criterion at convergence: 187.3
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -2.95093 -0.53293 0.01909 0.56185 2.67375
##
## Random effects:
## Groups Name Variance Std.Dev. Corr
## UF (Intercept) 0.10433 0.3230
## poly(ano, 1) 8.09341 2.8449 -0.15
## Residual 0.07885 0.2808
## Number of obs: 216, groups: UF, 27
##
## Fixed effects:
## Estimate Std. Error df t value Pr(>|t|)
## (Intercept) -11.01015 0.06503 21.94915 -169.301 < 2e-16 ***
## poly(ano, 1) -2.10982 0.84880 78.76767 -2.486 0.0150 *
## s_dens1 -0.17031 0.08990 39.02452 -1.895 0.0656 .
## s_jov_14_24 -0.04241 0.06595 138.18011 -0.643 0.5213
## s_pbf 0.22482 0.09199 64.39154 2.444 0.0173 *
## s_tx_desocup 0.24227 0.04528 192.28706 5.351 2.48e-07 ***
## s_tx_pres 0.07822 0.05226 100.23945 1.497 0.1376
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr) pl(,1) s_dns1 s__14_ s_pbf s_tx_d
## poly(ano,1) -0.092
## s_dens1 0.000 0.284
## s_jov_14_24 0.000 0.498 0.301
## s_pbf 0.000 -0.281 -0.611 -0.553
## s_tx_desocp 0.000 -0.579 -0.361 -0.342 0.269
## s_tx_pres 0.000 -0.257 0.103 -0.205 0.209 0.072
r2_nakagawa(modelos[[i]])
## # R2 for Mixed Models
##
## Conditional R2: 0.714
## Marginal R2: 0.201
i=68
#https://easystats.github.io/performance/reference/r2_nakagawa.html
formula[[i]]
## [1] "l_latrocinio ~ poly(ano,1) + (poly(ano,1) | UF)+ s_dens1+ s_pbf+ s_tx_desocup"
summary(modelos[[i]])
## Linear mixed model fit by REML. t-tests use Satterthwaite's method [
## lmerModLmerTest]
## Formula: formula[i]
## Data: dados1
## Offset: log(pop_ativa_cont)
##
## REML criterion at convergence: 181.8
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -2.9148 -0.5497 0.0174 0.5786 2.6618
##
## Random effects:
## Groups Name Variance Std.Dev. Corr
## UF (Intercept) 0.09355 0.3059
## poly(ano, 1) 8.10743 2.8474 -0.16
## Residual 0.08011 0.2830
## Number of obs: 216, groups: UF, 27
##
## Fixed effects:
## Estimate Std. Error df t value Pr(>|t|)
## (Intercept) -11.01015 0.06193 24.28161 -177.779 < 2e-16 ***
## poly(ano, 1) -1.65094 0.72486 47.13957 -2.278 0.0273 *
## s_dens1 -0.17528 0.08098 31.81505 -2.164 0.0380 *
## s_pbf 0.18001 0.07312 35.63349 2.462 0.0188 *
## s_tx_desocup 0.23210 0.04250 191.43291 5.461 1.46e-07 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr) pl(,1) s_dns1 s_pbf
## poly(ano,1) -0.118
## s_dens1 0.000 0.200
## s_pbf 0.000 0.018 -0.594
## s_tx_desocp 0.000 -0.510 -0.298 0.088
r2_nakagawa(modelos[[i]])
## # R2 for Mixed Models
##
## Conditional R2: 0.696
## Marginal R2: 0.199
i=261
#https://easystats.github.io/performance/reference/r2_nakagawa.html
formula[[i]]
## [1] "l_latrocinio ~ poly(ano,1) + (poly(ano,1) | UF)+ s_dens1+ s_pbf+ s_raz_1040+ s_tx_desocup+ s_tx_pres"
summary(modelos[[i]])
## Linear mixed model fit by REML. t-tests use Satterthwaite's method [
## lmerModLmerTest]
## Formula: formula[i]
## Data: dados1
## Offset: log(pop_ativa_cont)
##
## REML criterion at convergence: 186.4
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -2.95024 -0.56898 -0.00864 0.53863 2.63052
##
## Random effects:
## Groups Name Variance Std.Dev. Corr
## UF (Intercept) 0.10106 0.3179
## poly(ano, 1) 8.46792 2.9100 -0.18
## Residual 0.07817 0.2796
## Number of obs: 216, groups: UF, 27
##
## Fixed effects:
## Estimate Std. Error df t value Pr(>|t|)
## (Intercept) -11.01015 0.06407 23.66744 -171.844 < 2e-16 ***
## poly(ano, 1) -1.85591 0.74369 48.52292 -2.496 0.0160 *
## s_dens1 -0.13387 0.08633 33.45528 -1.551 0.1304
## s_pbf 0.20545 0.07586 37.79235 2.708 0.0101 *
## s_raz_1040 -0.06757 0.05028 175.62238 -1.344 0.1807
## s_tx_desocup 0.25270 0.04483 193.37214 5.637 6.06e-08 ***
## s_tx_pres 0.07601 0.05072 99.50728 1.499 0.1371
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr) pl(,1) s_dns1 s_pbf s__104 s_tx_d
## poly(ano,1) -0.129
## s_dens1 0.000 0.153
## s_pbf 0.000 -0.008 -0.526
## s_raz_1040 0.000 0.017 -0.190 -0.106
## s_tx_desocp 0.000 -0.471 -0.205 0.127 -0.333
## s_tx_pres 0.000 -0.178 0.192 0.123 -0.078 0.027
r2_nakagawa(modelos[[i]])
## # R2 for Mixed Models
##
## Conditional R2: 0.713
## Marginal R2: 0.197
Top 5 modelos pelo AIC (os menores AIC)
datatable(resultados1)
i=9
formula[[i]]
## [1] "l_latrocinio ~ poly(ano,1) + (poly(ano,1) | UF)+ s_tx_desocup"
summary(modelos[[i]])
## Linear mixed model fit by REML. t-tests use Satterthwaite's method [
## lmerModLmerTest]
## Formula: formula[i]
## Data: dados1
## Offset: log(pop_ativa_cont)
##
## REML criterion at convergence: 180.7
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -3.05777 -0.59973 0.01066 0.60272 2.61967
##
## Random effects:
## Groups Name Variance Std.Dev. Corr
## UF (Intercept) 0.10468 0.3235
## poly(ano, 1) 7.94060 2.8179 0.03
## Residual 0.08069 0.2841
## Number of obs: 216, groups: UF, 27
##
## Fixed effects:
## Estimate Std. Error df t value Pr(>|t|)
## (Intercept) -11.01015 0.06520 25.95350 -168.878 < 2e-16 ***
## poly(ano, 1) -1.51229 0.69922 42.71817 -2.163 0.0362 *
## s_tx_desocup 0.21085 0.04126 187.32007 5.110 7.89e-07 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr) pl(,1)
## poly(ano,1) 0.021
## s_tx_desocp 0.000 -0.483
r2_nakagawa(modelos[[i]])
## # R2 for Mixed Models
##
## Conditional R2: 0.681
## Marginal R2: 0.122
i=42
formula[[i]]
## [1] "l_latrocinio ~ poly(ano,1) + (poly(ano,1) | UF)+ s_perc_hom+ s_tx_desocup"
summary(modelos[[i]])
## Linear mixed model fit by REML. t-tests use Satterthwaite's method [
## lmerModLmerTest]
## Formula: formula[i]
## Data: dados1
## Offset: log(pop_ativa_cont)
##
## REML criterion at convergence: 180.1
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -3.13399 -0.56102 0.00438 0.56616 2.82249
##
## Random effects:
## Groups Name Variance Std.Dev. Corr
## UF (Intercept) 0.09108 0.3018
## poly(ano, 1) 8.04234 2.8359 -0.07
## Residual 0.08021 0.2832
## Number of obs: 216, groups: UF, 27
##
## Fixed effects:
## Estimate Std. Error df t value Pr(>|t|)
## (Intercept) -11.01015 0.06119 25.45365 -179.926 < 2e-16 ***
## poly(ano, 1) -1.45835 0.69890 41.96989 -2.087 0.0430 *
## s_perc_hom 0.09750 0.04170 130.68318 2.338 0.0209 *
## s_tx_desocup 0.22075 0.04069 182.58896 5.425 1.82e-07 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr) pl(,1) s_prc_
## poly(ano,1) -0.049
## s_perc_hom 0.000 0.034
## s_tx_desocp 0.000 -0.468 0.102
r2_nakagawa(modelos[[i]])
## # R2 for Mixed Models
##
## Conditional R2: 0.668
## Marginal R2: 0.137
i=24
formula[[i]]
## [1] "l_latrocinio ~ poly(ano,1) + (poly(ano,1) | UF)+ s_gini+ s_tx_desocup"
summary(modelos[[i]])
## Linear mixed model fit by REML. t-tests use Satterthwaite's method [
## lmerModLmerTest]
## Formula: formula[i]
## Data: dados1
## Offset: log(pop_ativa_cont)
##
## REML criterion at convergence: 182.3
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -3.05877 -0.56044 -0.00191 0.54674 2.63550
##
## Random effects:
## Groups Name Variance Std.Dev. Corr
## UF (Intercept) 0.11216 0.3349
## poly(ano, 1) 8.31569 2.8837 0.03
## Residual 0.07876 0.2806
## Number of obs: 216, groups: UF, 27
##
## Fixed effects:
## Estimate Std. Error df t value Pr(>|t|)
## (Intercept) -11.01015 0.06722 24.85818 -163.791 < 2e-16 ***
## poly(ano, 1) -1.62189 0.71029 42.43961 -2.283 0.0275 *
## s_gini -0.08232 0.05074 143.02885 -1.622 0.1069
## s_tx_desocup 0.23705 0.04412 194.27190 5.373 2.2e-07 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr) pl(,1) s_gini
## poly(ano,1) 0.025
## s_gini 0.000 0.093
## s_tx_desocp 0.000 -0.476 -0.362
r2_nakagawa(modelos[[i]])
## # R2 for Mixed Models
##
## Conditional R2: 0.689
## Marginal R2: 0.094
i=37
formula[[i]]
## [1] "l_latrocinio ~ poly(ano,1) + (poly(ano,1) | UF)+ s_pbf+ s_tx_desocup"
summary(modelos[[i]])
## Linear mixed model fit by REML. t-tests use Satterthwaite's method [
## lmerModLmerTest]
## Formula: formula[i]
## Data: dados1
## Offset: log(pop_ativa_cont)
##
## REML criterion at convergence: 182.8
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -3.00924 -0.59927 0.01059 0.59135 2.58433
##
## Random effects:
## Groups Name Variance Std.Dev. Corr
## UF (Intercept) 0.10422 0.3228
## poly(ano, 1) 8.01556 2.8312 -0.04
## Residual 0.08037 0.2835
## Number of obs: 216, groups: UF, 27
##
## Fixed effects:
## Estimate Std. Error df t value Pr(>|t|)
## (Intercept) -11.01015 0.06505 24.86986 -169.248 < 2e-16 ***
## poly(ano, 1) -1.36181 0.71018 44.86485 -1.918 0.0615 .
## s_pbf 0.08007 0.06183 35.54221 1.295 0.2037
## s_tx_desocup 0.20659 0.04122 190.23771 5.011 1.23e-06 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr) pl(,1) s_pbf
## poly(ano,1) -0.026
## s_pbf 0.000 0.167
## s_tx_desocp 0.000 -0.486 -0.086
r2_nakagawa(modelos[[i]])
## # R2 for Mixed Models
##
## Conditional R2: 0.702
## Marginal R2: 0.179
i=50
formula[[i]]
## [1] "l_latrocinio ~ poly(ano,1) + (poly(ano,1) | UF)+ s_tx_desocup+ s_tx_pres"
summary(modelos[[i]])
## Linear mixed model fit by REML. t-tests use Satterthwaite's method [
## lmerModLmerTest]
## Formula: formula[i]
## Data: dados1
## Offset: log(pop_ativa_cont)
##
## REML criterion at convergence: 183.3
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -3.09540 -0.57813 0.02067 0.58518 2.63823
##
## Random effects:
## Groups Name Variance Std.Dev. Corr
## UF (Intercept) 0.10876 0.3298
## poly(ano, 1) 7.89518 2.8098 0.01
## Residual 0.08005 0.2829
## Number of obs: 216, groups: UF, 27
##
## Fixed effects:
## Estimate Std. Error df t value Pr(>|t|)
## (Intercept) -11.01015 0.06632 25.32308 -166.005 < 2e-16 ***
## poly(ano, 1) -1.74870 0.72204 48.13406 -2.422 0.0193 *
## s_tx_desocup 0.21484 0.04134 188.96262 5.196 5.24e-07 ***
## s_tx_pres 0.06276 0.04995 94.10367 1.256 0.2121
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr) pl(,1) s_tx_d
## poly(ano,1) 0.009
## s_tx_desocp 0.000 -0.485
## s_tx_pres 0.000 -0.259 0.073
r2_nakagawa(modelos[[i]])
## # R2 for Mixed Models
##
## Conditional R2: 0.688
## Marginal R2: 0.121
Sugestão 1: entre os Top 10 dos melhores R2, pegar o menor AIC (verificar no datatable de resultados1, que o modelo com i=36:
- É equivalente a escolher o melhor R2 entre os 10 menores AIC.
datatable(resultados1)
i=68
formula[[i]]
## [1] "l_latrocinio ~ poly(ano,1) + (poly(ano,1) | UF)+ s_dens1+ s_pbf+ s_tx_desocup"
summary(modelos[[i]])
## Linear mixed model fit by REML. t-tests use Satterthwaite's method [
## lmerModLmerTest]
## Formula: formula[i]
## Data: dados1
## Offset: log(pop_ativa_cont)
##
## REML criterion at convergence: 181.8
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -2.9148 -0.5497 0.0174 0.5786 2.6618
##
## Random effects:
## Groups Name Variance Std.Dev. Corr
## UF (Intercept) 0.09355 0.3059
## poly(ano, 1) 8.10743 2.8474 -0.16
## Residual 0.08011 0.2830
## Number of obs: 216, groups: UF, 27
##
## Fixed effects:
## Estimate Std. Error df t value Pr(>|t|)
## (Intercept) -11.01015 0.06193 24.28161 -177.779 < 2e-16 ***
## poly(ano, 1) -1.65094 0.72486 47.13957 -2.278 0.0273 *
## s_dens1 -0.17528 0.08098 31.81505 -2.164 0.0380 *
## s_pbf 0.18001 0.07312 35.63349 2.462 0.0188 *
## s_tx_desocup 0.23210 0.04250 191.43291 5.461 1.46e-07 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr) pl(,1) s_dns1 s_pbf
## poly(ano,1) -0.118
## s_dens1 0.000 0.200
## s_pbf 0.000 0.018 -0.594
## s_tx_desocp 0.000 -0.510 -0.298 0.088
r2_nakagawa(modelos[[i]])
## # R2 for Mixed Models
##
## Conditional R2: 0.696
## Marginal R2: 0.199
- Análise de resíduos condicionais para o modelo i=45:
shapiro.test(compute_redres(modelos[[i]]))[2]
## $p.value
## [1] 0.9450177
plot(compute_redres(modelos[[i]]),main="resíduos condicionais versus índices")
plot_resqq(modelos[[i]])
- Análise dos efeitos aleatórios para o modelo i=45:
random <- ranef(modelos[[i]])
aleatorio1 = random[["UF"]][["(Intercept)"]]
aleatorio2 = random[["UF"]][[2]] #É o termo do poly(ano,1)
plot(aleatorio1,main="efeitos aleatórios versus índices")
plot(aleatorio2,main="efeitos aleatórios versus índices")
plot_ranef(modelos[[i]])
shapiro.test(aleatorio1)
##
## Shapiro-Wilk normality test
##
## data: aleatorio1
## W = 0.98386, p-value = 0.9371
shapiro.test(aleatorio2)
##
## Shapiro-Wilk normality test
##
## data: aleatorio2
## W = 0.93713, p-value = 0.1035