roe_pres

pacckages

if(!require(googlesheets4)){install.packages("googlesheets4")}
if(!require(AICcmodavg)){install.packages("AICcmodavg")}
if(!require(ggplot2)){install.packages("ggplot2")}
#
library(googlesheets4); gs4_deauth()
library(AICcmodavg)
library(ggplot2)

base de datos link

sheet="factoriesgo"
range="B4:J400"
roe <- read_sheet("https://docs.google.com/spreadsheets/d/1fBIBUpp8gCwPIyo_Hr1wPXshPDIWXYHrDXY-UcF-s7I/edit?usp=sharing",
                  col_names = TRUE,
                  sheet=sheet,
                  range=range,
                  col_types = NULL,
                  na= "NA")

Reading from "Copia de Factores de riesgo"
Range "'factoriesgo'!B4:J400"

dim(roe)

[1] 396   9

head(roe)

Explore data

roe$piso <- as.factor(roe$piso)
roe$pared <- as.factor(roe$pared)
roe$rest.alim <- as.factor(roe$rest.alim)
roe$rest.alim.F <- as.factor(roe$rest.alim)
roe$techo <- as.factor(roe$techo)
roe$sitio <- as.factor(roe$sitio)
roe$pres.roe.F <- as.factor(roe$pres.roe)
roe$pres.dog.F <- as.factor(roe$pres.dog)
roe$pres.cat.F <- as.factor(roe$pres.cat)
#
dim(roe)

[1] 396  13

names(roe)

 [1] "hh.id"       "piso"        "pared"       "rest.alim"   "techo"       "sitio"      
 [7] "pres.roe"    "pres.dog"    "pres.cat"    "rest.alim.F" "pres.roe.F"  "pres.dog.F" 
[13] "pres.cat.F" 

summary(roe)

     hh.id              piso            pared     rest.alim             techo     sitio  
 Min.   :  1.00   A.Tyles : 42   1.Block   :198   No  : 91   conc_zinc_teja: 11   A:148  
 1st Qu.: 99.75   Concrete:183   adobe     : 43   Si  :288   zinc          :385   B: 83  
 Median :198.50   Dirt    :171   Wood-Cane : 50   NA's: 17                        C:165  
 Mean   :198.50                  Wood_other: 54                                          
 3rd Qu.:297.25                  zinc_other: 51                                          
 Max.   :396.00                                                                          
    pres.roe         pres.dog        pres.cat      rest.alim.F pres.roe.F pres.dog.F
 Min.   :0.0000   Min.   :0.000   Min.   :0.0000   No  : 91    0:277      0:120     
 1st Qu.:0.0000   1st Qu.:0.000   1st Qu.:0.0000   Si  :288    1:119      1:276     
 Median :0.0000   Median :1.000   Median :0.0000   NA's: 17                         
 Mean   :0.3005   Mean   :0.697   Mean   :0.4141                                    
 3rd Qu.:1.0000   3rd Qu.:1.000   3rd Qu.:1.0000                                    
 Max.   :1.0000   Max.   :1.000   Max.   :1.0000                                    
 pres.cat.F
 0:232     
 1:164     
           
           
           
           

Plot_explore

## PLOT Sitio ####
ggplot(aes(x= sitio, y= as.factor(pres.roe) , 
           fill= as.factor(pres.roe) ), data= roe ) +
  geom_col() 

ggplot(aes(x= sitio, y= as.factor(pres.roe) , fill= as.factor(pres.roe) ), 
       data= roe ) + 
  geom_point(size= 0.5,  position= position_jitter(width= 0.2 , height =0.1 ))

#### chiqs.test
a <- chisq.test(table(roe$sitio, roe$pres.roe));a; a$stdres

    Pearson's Chi-squared test

data:  table(roe$sitio, roe$pres.roe)
X-squared = 12.589, df = 2, p-value = 0.001847

   
            0         1
  A -1.025216  1.025216
  B -2.708528  2.708528
  C  3.242188 -3.242188

## Según la tabla de valores esperados de la Chi2 arriba, los sitios B y C tienen más y menos (respectivamente) proporción de lo "esperado" bajo un modelo nulo.

## PLOT restos alimentos ####
ggplot(aes(x= rest.alim.F, y= as.factor(pres.roe) , 
           fill= as.factor(pres.roe) ), data= roe ) +
  geom_col() 

ggplot(aes(x= rest.alim.F, y= as.factor(pres.roe) , fill= as.factor(pres.roe) ), 
       data= roe ) + 
  geom_point(size= 0.5,  position= position_jitter(width= 0.2 , height =0.1 ))

##  PLOT piso    ####

ggplot(aes(x= piso, y= as.factor(pres.roe) , 
           fill= as.factor(pres.roe) ), data= roe ) +
  geom_col() 

ggplot(aes(x= piso, y= as.factor(pres.roe) , fill= as.factor(pres.roe) ), 
       data= roe ) + 
  geom_point(size= 0.5,  position= position_jitter(width= 0.2 , height =0.1 ))

## PLOT pared ####

ggplot(aes(x= pared, y= as.factor(pres.roe) , 
           fill= as.factor(pres.roe) ), data= roe ) +
  geom_col() 

ggplot(aes(x= pared, y= as.factor(pres.roe) , fill= as.factor(pres.roe) ), 
       data= roe ) + 
  geom_point(size= 0.5,  position= position_jitter(width= 0.2 , height =0.1 ))

# chiqs.test
a <- chisq.test(table(roe$pared, roe$pres.roe));a; a$stdres

    Pearson's Chi-squared test

data:  table(roe$pared, roe$pres.roe)
X-squared = 9.639, df = 4, p-value = 0.04697

            
                      0          1
  1.Block    -0.3288179  0.3288179
  adobe      -1.7890592  1.7890592
  Wood-Cane   0.9983159 -0.9983159
  Wood_other  2.3083089 -2.3083089
  zinc_other -1.2022716  1.2022716

## PLOT techo ####

ggplot(aes(x= techo, y= as.factor(pres.roe) , 
           fill= as.factor(pres.roe) ), data= roe ) +
  geom_col() 

ggplot(aes(x= techo, y= as.factor(pres.roe) , fill= as.factor(pres.roe) ), 
       data= roe ) + 
  geom_point(size= 0.5,  position= position_jitter(width= 0.2 , height =0.1 ))

#### Asociaci[on entre la presencia de gatos y el sitio 
a <- chisq.test(table(roe$sitio, roe$pres.cat)); a

    Pearson's Chi-squared test

data:  table(roe$sitio, roe$pres.cat)
X-squared = 43.636, df = 2, p-value = 3.347e-10

a$stdres

   
            0         1
  A  3.646605 -3.646605
  B  3.602760 -3.602760
  C -6.552865  6.552865

GLM

## Sitio ####
m.sitio <- glm( pres.roe ~ sitio, family= binomial, data=roe)
summary(m.sitio)

Call:
glm(formula = pres.roe ~ sitio, family = binomial, data = roe)

Deviance Residuals: 
    Min       1Q   Median       3Q      Max  
-1.0466  -0.8968  -0.6905   1.3141   1.7610  

Coefficients:
            Estimate Std. Error z value Pr(>|z|)    
(Intercept)  -0.7033     0.1747  -4.026 5.66e-05 ***
sitioB        0.3874     0.2827   1.371   0.1705    
sitioC       -0.6089     0.2584  -2.356   0.0185 *  
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

(Dispersion parameter for binomial family taken to be 1)

    Null deviance: 484.14  on 395  degrees of freedom
Residual deviance: 471.49  on 393  degrees of freedom
AIC: 477.49

Number of Fisher Scoring iterations: 4

#### pres.ROe ~ BUILDING  ####
# piso
m.piso <- glm( pres.roe ~ piso, family= binomial, data=roe)  ### piso
summary(m.piso)

Call:
glm(formula = pres.roe ~ piso, family = binomial, data = roe)

Deviance Residuals: 
    Min       1Q   Median       3Q      Max  
-0.8908  -0.8908  -0.8270   1.4942   1.6942  

Coefficients:
             Estimate Std. Error z value Pr(>|z|)   
(Intercept)   -1.1632     0.3623  -3.211  0.00132 **
pisoConcrete   0.2659     0.3973   0.669  0.50326   
pisoDirt       0.4436     0.3972   1.117  0.26416   
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

(Dispersion parameter for binomial family taken to be 1)

    Null deviance: 484.14  on 395  degrees of freedom
Residual deviance: 482.64  on 393  degrees of freedom
AIC: 488.64

Number of Fisher Scoring iterations: 4

# pared
m.pared <- glm( pres.roe ~ pared, family= binomial, data=roe) ### pared
summary(m.pared)

Call:
glm(formula = pres.roe ~ pared, family = binomial, data = roe)

Deviance Residuals: 
    Min       1Q   Median       3Q      Max  
-1.0415  -0.8582  -0.7409   1.4053   1.8930  

Coefficients:
                Estimate Std. Error z value Pr(>|z|)    
(Intercept)      -0.8091     0.1539  -5.257 1.47e-07 ***
paredadobe        0.4806     0.3453   1.392   0.1640    
paredWood-Cane   -0.3436     0.3652  -0.941   0.3468    
paredWood_other  -0.8003     0.3963  -2.020   0.0434 *  
paredzinc_other   0.2878     0.3280   0.878   0.3802    
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

(Dispersion parameter for binomial family taken to be 1)

    Null deviance: 484.14  on 395  degrees of freedom
Residual deviance: 474.14  on 391  degrees of freedom
AIC: 484.14

Number of Fisher Scoring iterations: 4

# techo
m.techo <- glm( pres.roe ~ techo, family= binomial, data=roe) ### techo
summary(m.techo)

Call:
glm(formula = pres.roe ~ techo, family = binomial, data = roe)

Deviance Residuals: 
    Min       1Q   Median       3Q      Max  
-0.9508  -0.8424  -0.8424   1.5546   1.5546  

Coefficients:
            Estimate Std. Error z value Pr(>|z|)
(Intercept)  -0.5596     0.6268  -0.893    0.372
techozinc    -0.2939     0.6366  -0.462    0.644

(Dispersion parameter for binomial family taken to be 1)

    Null deviance: 484.14  on 395  degrees of freedom
Residual deviance: 483.94  on 394  degrees of freedom
AIC: 487.94

Number of Fisher Scoring iterations: 4

# restos de alimento
m.rest.alim <- glm( pres.roe ~ rest.alim, family= binomial, data=roe)   ### rest.alim
summary(m.rest.alim)

Call:
glm(formula = pres.roe ~ rest.alim, family = binomial, data = roe)

Deviance Residuals: 
    Min       1Q   Median       3Q      Max  
-0.9127  -0.8246  -0.8246   1.4676   1.5774  

Coefficients:
            Estimate Std. Error z value Pr(>|z|)   
(Intercept)  -0.6604     0.2212  -2.985  0.00283 **
rest.alimSi  -0.2438     0.2566  -0.950  0.34206   
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

(Dispersion parameter for binomial family taken to be 1)

    Null deviance: 463.54  on 378  degrees of freedom
Residual deviance: 462.65  on 377  degrees of freedom
  (17 observations deleted due to missingness)
AIC: 466.65

Number of Fisher Scoring iterations: 4

m.house <- glm( pres.roe ~ piso + pared + rest.alim + roe$techo, family= binomial, data=roe)
summary(m.house)

Call:
glm(formula = pres.roe ~ piso + pared + rest.alim + roe$techo, 
    family = binomial, data = roe)

Deviance Residuals: 
    Min       1Q   Median       3Q      Max  
-1.1838  -0.9037  -0.7275   1.3689   1.9550  

Coefficients:
                Estimate Std. Error z value Pr(>|z|)  
(Intercept)      -0.8620     0.7357  -1.172   0.2413  
pisoConcrete      0.4638     0.4214   1.101   0.2710  
pisoDirt          0.7126     0.4599   1.550   0.1213  
paredadobe        0.3029     0.3898   0.777   0.4371  
paredWood-Cane   -0.6199     0.4578  -1.354   0.1757  
paredWood_other  -0.9278     0.4313  -2.151   0.0315 *
paredzinc_other  -0.0077     0.3763  -0.020   0.9837  
rest.alimSi      -0.1385     0.2675  -0.518   0.6047  
roe$techozinc    -0.2864     0.6562  -0.436   0.6625  
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

(Dispersion parameter for binomial family taken to be 1)

    Null deviance: 463.54  on 378  degrees of freedom
Residual deviance: 451.04  on 370  degrees of freedom
  (17 observations deleted due to missingness)
AIC: 469.04

Number of Fisher Scoring iterations: 4

### pres.roe ~ DOMESTIC_ANIMAL ####

m.cat <- glm( pres.roe ~ pres.cat, family= binomial, data=roe)    ### gatos
summary(m.cat)

Call:
glm(formula = pres.roe ~ pres.cat, family = binomial, data = roe)

Deviance Residuals: 
    Min       1Q   Median       3Q      Max  
-1.0406  -1.0406  -0.5367   1.3206   2.0044  

Coefficients:
            Estimate Std. Error z value Pr(>|z|)    
(Intercept)  -0.3306     0.1331  -2.483    0.013 *  
pres.cat     -1.5342     0.2650  -5.790 7.04e-09 ***
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

(Dispersion parameter for binomial family taken to be 1)

    Null deviance: 484.14  on 395  degrees of freedom
Residual deviance: 444.66  on 394  degrees of freedom
AIC: 448.66

Number of Fisher Scoring iterations: 4

m.dog <- glm( pres.roe ~ pres.dog, family= binomial, data=roe)    ### perros
summary(m.dog)

Call:
glm(formula = pres.roe ~ pres.dog, family = binomial, data = roe)

Deviance Residuals: 
    Min       1Q   Median       3Q      Max  
-0.9282  -0.8088  -0.8088   1.4490   1.5979  

Coefficients:
            Estimate Std. Error z value Pr(>|z|)   
(Intercept)  -0.6190     0.1914  -3.234  0.00122 **
pres.dog     -0.3305     0.2338  -1.414  0.15745   
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

(Dispersion parameter for binomial family taken to be 1)

    Null deviance: 484.14  on 395  degrees of freedom
Residual deviance: 482.17  on 394  degrees of freedom
AIC: 486.17

Number of Fisher Scoring iterations: 4

m.dom.anim <- glm( pres.roe ~ pres.dog * pres.cat, family= binomial, data=roe)
summary(m.dom.anim)

Call:
glm(formula = pres.roe ~ pres.dog * pres.cat, family = binomial, 
    data = roe)

Deviance Residuals: 
   Min      1Q  Median      3Q     Max  
-1.066  -1.025  -0.553   1.293   2.146  

Coefficients:
                  Estimate Std. Error z value Pr(>|z|)   
(Intercept)        -0.2683     0.2127  -1.261  0.20726   
pres.dog           -0.1021     0.2728  -0.374  0.70813   
pres.cat           -1.9290     0.6446  -2.992  0.00277 **
pres.dog:pres.cat   0.4988     0.7114   0.701  0.48315   
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

(Dispersion parameter for binomial family taken to be 1)

    Null deviance: 484.14  on 395  degrees of freedom
Residual deviance: 444.13  on 392  degrees of freedom
AIC: 452.13

Number of Fisher Scoring iterations: 4

### combinaciones ####
m.pared.cat <- glm( pres.roe ~ pared + pres.cat        , family= binomial, data=roe)
summary(m.pared.cat)

Call:
glm(formula = pres.roe ~ pared + pres.cat, family = binomial, 
    data = roe)

Deviance Residuals: 
    Min       1Q   Median       3Q      Max  
-1.1740  -1.0341  -0.5521   1.2363   2.2317  

Coefficients:
                Estimate Std. Error z value Pr(>|z|)    
(Intercept)     -0.34679    0.17338  -2.000   0.0455 *  
paredadobe       0.33872    0.35988   0.941   0.3466    
paredWood-Cane  -0.09434    0.38499  -0.245   0.8064    
paredWood_other -0.59978    0.41179  -1.457   0.1453    
paredzinc_other  0.20941    0.34239   0.612   0.5408    
pres.cat        -1.45707    0.26913  -5.414 6.16e-08 ***
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

(Dispersion parameter for binomial family taken to be 1)

    Null deviance: 484.14  on 395  degrees of freedom
Residual deviance: 440.31  on 390  degrees of freedom
AIC: 452.31

Number of Fisher Scoring iterations: 4

#
m.pared.sitio <- glm( pres.roe ~ pared +            sitio, family= binomial, data=roe)
summary(m.pared.sitio)

Call:
glm(formula = pres.roe ~ pared + sitio, family = binomial, data = roe)

Deviance Residuals: 
    Min       1Q   Median       3Q      Max  
-1.1512  -0.8797  -0.7207   1.3197   1.9334  

Coefficients:
                Estimate Std. Error z value Pr(>|z|)    
(Intercept)     -0.74987    0.20954  -3.579 0.000345 ***
paredadobe      -0.01518    0.44640  -0.034 0.972875    
paredWood-Cane  -0.02271    0.41078  -0.055 0.955916    
paredWood_other -0.50866    0.42207  -1.205 0.228140    
paredzinc_other  0.25143    0.34591   0.727 0.467314    
sitioB           0.43655    0.37779   1.156 0.247871    
sitioC          -0.44299    0.31267  -1.417 0.156536    
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

(Dispersion parameter for binomial family taken to be 1)

    Null deviance: 484.14  on 395  degrees of freedom
Residual deviance: 469.07  on 389  degrees of freedom
AIC: 483.07

Number of Fisher Scoring iterations: 4

#
m.cat.sitio <- glm( pres.roe ~         + pres.cat + sitio, family= binomial, data=roe)
summary(m.cat.sitio)

Call:
glm(formula = pres.roe ~ +pres.cat + sitio, family = binomial, 
    data = roe)

Deviance Residuals: 
    Min       1Q   Median       3Q      Max  
-1.1685  -0.9390  -0.5017   1.1864   2.0663  

Coefficients:
            Estimate Std. Error z value Pr(>|z|)    
(Intercept)  -0.3632     0.1873  -1.939   0.0525 .  
pres.cat     -1.4184     0.2753  -5.151 2.59e-07 ***
sitioB        0.3421     0.2929   1.168   0.2428    
sitioC       -0.2273     0.2762  -0.823   0.4105    
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

(Dispersion parameter for binomial family taken to be 1)

    Null deviance: 484.14  on 395  degrees of freedom
Residual deviance: 441.33  on 392  degrees of freedom
AIC: 449.33

Number of Fisher Scoring iterations: 4

### MODELO GLOBAL ####
global <- glm( pres.roe ~ pared + pres.cat + sitio, family= binomial, data=roe)
summary(global)

Call:
glm(formula = pres.roe ~ pared + pres.cat + sitio, family = binomial, 
    data = roe)

Deviance Residuals: 
    Min       1Q   Median       3Q      Max  
-1.1712  -0.9846  -0.5302   1.1925   2.2263  

Coefficients:
                Estimate Std. Error z value Pr(>|z|)    
(Intercept)     -0.41783    0.22311  -1.873   0.0611 .  
paredadobe       0.02078    0.46202   0.045   0.9641    
paredWood-Cane  -0.02297    0.42824  -0.054   0.9572    
paredWood_other -0.49926    0.43769  -1.141   0.2540    
paredzinc_other  0.26597    0.36019   0.738   0.4603    
pres.cat        -1.41924    0.27629  -5.137 2.79e-07 ***
sitioB           0.38238    0.39197   0.976   0.3293    
sitioC          -0.05416    0.33325  -0.163   0.8709    
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

(Dispersion parameter for binomial family taken to be 1)

    Null deviance: 484.14  on 395  degrees of freedom
Residual deviance: 439.10  on 388  degrees of freedom
AIC: 455.1

Number of Fisher Scoring iterations: 4

AIC, AICc, Wi

## models

n.= length(roe$pres.roe)
## extracting AICc
extractAIC(global)[2]

[1] 455.0989

k=AICc(global, return.K = TRUE)
AICc.global <- extractAIC(global)[2] + (((2*k)*(k+1))/(n.-k-1))
k1=AICc(global, return.K = TRUE)

extractAIC(m.cat)[2]

[1] 448.6635

k=AICc(m.cat, return.K = TRUE)
AICc.m.cat <- extractAIC(m.cat)[2] + (((2*k)*(k+1))/(n.-k-1))
k2=AICc(m.cat, return.K = TRUE)

extractAIC(m.pared)[2]

[1] 484.1374

k=AICc(m.pared, return.K = TRUE)
AICc.m.pared <- extractAIC(m.pared)[2] + (((2*k)*(k+1))/(n.-k-1))
k3=AICc(m.pared, return.K = TRUE)

extractAIC(m.sitio)[2]

[1] 477.4893

k=AICc(m.sitio, return.K = TRUE)
AICc.m.sitio <- extractAIC(m.sitio)[2] + (((2*k)*(k+1))/(n.-k-1))
k4=AICc(m.sitio, return.K = TRUE)

extractAIC(m.pared.cat)[2]

[1] 452.3076

k=AICc(m.pared.cat, return.K = TRUE)
AICc.m.pared.cat <- extractAIC(m.pared.cat)[2] + (((2*k)*(k+1))/(n.-k-1))
k5=AICc(m.pared.cat, return.K = TRUE)

extractAIC(m.pared.sitio)[2]

[1] 483.0708

k=AICc(m.pared.sitio, return.K = TRUE)
AICc.m.pared.sitio <- extractAIC(m.pared.sitio)[2] + (((2*k)*(k+1))/(n.-k-1))
k6=AICc(m.pared.sitio, return.K = TRUE)

extractAIC(m.cat.sitio)[2]

[1] 449.3316

k=AICc(m.cat.sitio, return.K = TRUE)
AICc.m.cat.sitio <- extractAIC(m.cat.sitio)[2] + (((2*k)*(k+1))/(n.-k-1))
k7=AICc(m.cat.sitio, return.K = TRUE)

########## AIC table #####
aics<- data.frame(paste("m",c("Global", "Gato", "Pared", "Sitio", "pared.gato", "pared.sitio", "gato.sitio"),sep=""),
                  c(AICc.global,AICc.m.cat,AICc.m.pared, AICc.m.sitio, AICc.m.pared.cat, AICc.m.pared.sitio, AICc.m.cat.sitio),
                  c(k1,k2,k3,k4,k5,k6,k7),
                  row.names=NULL) 
colnames(aics) <- c("model","AICc", "k") 
# sort according to AIC
aics<-aics[order(aics$AICc),] 

wi (weights)

for(i in 1:dim(aics)[1]){
  aics$deltaAIC[i]<-aics$AICc[1]-aics$AICc[i]}
wi<-exp(aics$deltaAIC/2)
wi <- wi/sum(wi)
aics$wi <- round(wi/sum(wi), 3)
aics

Delta AICs. Cuánto “pierde” en ajuste (AICc) el modelo completo Gato+Pared+Sitio sacando (ó incluyendo) cada factor?

Deviance Gain Method [ loss of GOF]

###
global$formula; AICc(global)      ###### GATO + PARED 

pres.roe ~ pared + pres.cat + sitio
[1] 455.471

##########################  Contribution
m.pared.sitio$formula; AICc(m.pared.sitio); AICc(global) - AICc(m.pared.sitio)   ### 'GATO' contribution

pres.roe ~ pared + sitio
[1] 483.3595
[1] -27.88845

m.cat.sitio$formula; AICc(m.cat.sitio); AICc(global)- AICc(m.cat.sitio)          ### 'Pared' contribution

pres.roe ~ +pres.cat + sitio
[1] 449.4339
[1] 6.037145

m.pared.cat$formula; AICc(m.pared.cat); AICc(global)- AICc(m.pared.cat)          ### 'Sitio' contribution

pres.roe ~ pared + pres.cat
[1] 452.5235
[1] 2.947494

#
DeltaAics <- as.data.frame(cbind("Model"= c("global","Gato", "Pared", "Sitio"),
      "Delta.AIC"= c(0,AICc(global) - AICc(m.pared.sitio), AICc(global)- AICc(m.cat.sitio),AICc(global)- AICc(m.pared.cat))))

DeltaAics$Delta.AIC <- as.numeric(DeltaAics$Delta.AIC)
DeltaAics$Delta.AIC <- round(DeltaAics$Delta.AIC,2)
DeltaAics

El incluir pared y Sitio de hecho hace que el modelo empeore (aumenta el AICc)