Precisão de juízes.
Exercício com kappa. importando banco de dados
library(readxl)
        setwd("~/Dropbox/R Stat/")
        bd <- read_excel("bd_tcfi.xlsx")
        bd$id_rwn <- as.numeric(rownames(bd))

source("http://www.labape.com.br/rprimi/R/sharedcount.R")
        
names(bd)
##  [1] "ord0"     "RaterN"   "Rater"    "ID"       "At.1"     "At.2 "   
##  [7] "At.3"     "At3.01"   "At3.02"   "At3.03"   "At3.04"   "At3.05"  
## [13] "At3.06"   "At3.07"   "At3.08"   "At3.09"   "At3.10"   "Obs."    
## [19] "id_letra" "id"       "id_num_f" "idnum"    "id_rwn"
# Quantos sujeitos
length(unique(bd$id))
## [1] 219
# Quantos juízes e quantas avaliações cada um fez
table(bd$Rater)
## 
##   1   2   3   4   5   6 
## 163  79  78  54  79  78
bd <- bd[!is.na(bd$ID), ]
Organizando banco de dados
library(reshape2)
 
names(bd)
##  [1] "ord0"     "RaterN"   "Rater"    "ID"       "At.1"     "At.2 "   
##  [7] "At.3"     "At3.01"   "At3.02"   "At3.03"   "At3.04"   "At3.05"  
## [13] "At3.06"   "At3.07"   "At3.08"   "At3.09"   "At3.10"   "Obs."    
## [19] "id_letra" "id"       "id_num_f" "idnum"    "id_rwn"
vars <- names(bd[ , c(4, 2, 23, 5:17)]) 
bd_long <- melt(bd[ , vars], 
                id.vars =1:3,
                measured.vars=4:16)

bd_long2 <-  dcast(data = bd_long[ , -3], formula = ID+variable~RaterN, value.var = "value")

# quantos idéias em comum cada juiz avaliou ?         
source("http://www.labape.com.br/rprimi/R/sharedcount.R")
sharedcount(bd_long2[ , c(3:8)])
##           X1       X2 out
## 1  alexandre    andre   0
## 2  alexandre    carla   0
## 3  alexandre  everson 843
## 4  alexandre jurandir   0
## 5  alexandre    marco 604
## 6      andre    carla 898
## 7      andre  everson   0
## 8      andre jurandir   0
## 9      andre    marco 614
## 10     carla  everson   0
## 11     carla jurandir   0
## 12     carla    marco 615
## 13   everson jurandir   0
## 14   everson    marco 623
## 15  jurandir    marco 662
Correlação simples entre os escores
names(bd_long2)
## [1] "ID"        "variable"  "alexandre" "andre"     "carla"     "everson"  
## [7] "jurandir"  "marco"
table(bd_long2$carla, bd_long2$andre)
##    
##       1   2   3   4   5
##   1 102  71  18   2   8
##   2  81 194  56  17   5
##   3  18  67  67  19   3
##   4   4  36  87  37   6
library(psych)
describe(bd_long2[ , 3:8])
##           vars    n mean   sd median trimmed  mad min max range  skew
## alexandre    1  844 3.18 0.75      3    3.19 0.00   2   5     3  0.17
## andre        2  900 2.27 0.98      2    2.18 1.48   1   5     4  0.59
## carla        3  906 2.34 1.03      2    2.30 1.48   1   4     3  0.32
## everson      4  864 2.22 1.08      2    2.10 1.48   1   5     4  0.77
## jurandir     5  702 2.27 1.11      2    2.31 1.48   0   5     5 -0.19
## marco        6 2117 2.07 0.85      2    2.20 0.00   0   4     4 -1.01
##           kurtosis   se
## alexandre    -0.34 0.03
## andre        -0.02 0.03
## carla        -1.03 0.03
## everson       0.02 0.04
## jurandir      0.46 0.04
## marco         1.15 0.02
multi.hist(bd_long2[ , 3:8])

pairs.panels(bd_long2[ , 3:8])

corPlot(bd_long2[ , 3:8],numbers=TRUE,upper=FALSE,diag=FALSE)


library(d3heatmap)
d3heatmap(cor(bd_long2[ , 3:8], use="pair"), 
          symn= TRUE,  symm = TRUE, 
          k_row = 1, k_col =1)

##### Concordância e kappa

names(bd_long2)
## [1] "ID"        "variable"  "alexandre" "andre"     "carla"     "everson"  
## [7] "jurandir"  "marco"
library(irr)
## Loading required package: lpSolve
agree(bd_long2[, c(8, 3)])
##  Percentage agreement (Tolerance=0)
## 
##  Subjects = 604 
##    Raters = 2 
##   %-agree = 28.3
agree(bd_long2[ , c(8, 3)], tolerance = 1)
##  Percentage agreement (Tolerance=1)
## 
##  Subjects = 604 
##    Raters = 2 
##   %-agree = 81
icc(bd_long2[, c(8, 3)], model="twoway", type="consistency")
##  Single Score Intraclass Correlation
## 
##    Model: twoway 
##    Type : consistency 
## 
##    Subjects = 604 
##      Raters = 2 
##    ICC(C,1) = 0.308
## 
##  F-Test, H0: r0 = 0 ; H1: r0 > 0 
##  F(603,603) = 1.89 , p = 4.37e-15 
## 
##  95%-Confidence Interval for ICC Population Values:
##   0.234 < ICC < 0.379