[R-br] Diferença entre métodos

Ari Clecius ari072000 em yahoo.com.br
Sábado Novembro 22 10:06:33 BRST 2014


Bom dia, estou rodando um experimento do tipo Box Behnken e usei a função reg e comparei com o obtido pela função rsm, o problema é que o que os dois consideram significativo difere, aonde posso estar errando? Testei também a função AIC para o modelo que ele fornece e dá diferente pela função AIC().
 Ari Clecius Alves de Lima
Engenheiro Químico
Me. Engenharia Civil
(085)88412345
(085)33669042
#CMR
library(DoE.wrapper)Design.1_2<- bbd.design(nfactors= 3 , ncenter= 4 ,randomize= FALSE,seed=7480,                       factor.names=list( A=c(0,400),B=c(600,2200),C=c(1,10) ) )
res=c(73.450,75.725,63.980,67.670,76.070,77.345,66.005,67.335,84.740,81.595,74.830,70.325,72.460,73.170,73.925,73.800)length(res)#Medidas resumoDesign.1_2<- add.response(Design.1_2,res, replace=FALSE)plancoded<- code.design(Design.1_2)plancoded
LinearModel.1 <- aov(res ~(x1+x2+x3)^2+ I(x1^2) + I(x2^2) + I(x3^2),                      data=plancoded)sreg = step(LinearModel.1,direction="both")
#Melhor modelo#Step:  AIC=14.07#res ~ x1 + x2 + x3 + I(x1^2) + I(x2^2) + I(x3^2)
#          Df Sum of Sq     RSS    AIC#<none>                  16.074 14.074#+ x1:x2    1     0.501  15.574 15.568#+ x2:x3    1     0.462  15.612 15.607#+ x1:x3    1     0.001  16.073 16.073#- x1       1     9.181  25.255 19.303#- I(x2^2)  1     9.310  25.384 19.385#- I(x3^2)  1    36.195  52.269 30.941#- x2       1    79.223  95.297 40.551#- I(x1^2)  1    86.793 102.867 41.774#- x3       1   212.747 228.821 54.566
summary(sreg)
#       Df Sum Sq Mean Sq F value   Pr(>F)    #x1           1   9.18    9.18   5.140  0.04959 *  #x2           1  79.22   79.22  44.357 9.27e-05 ***#x3           1 212.75  212.75 119.118 1.72e-06 ***#I(x1^2)      1  86.79   86.79  48.596 6.53e-05 ***#I(x2^2)      1   9.31    9.31   5.213  0.04831 *  #I(x3^2)      1  36.20   36.20  20.266  0.00149 ** #Residuals    9  16.07    1.79                     #---#Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
Design.1_2.coded<-plancodedrsmModel.2 <- rsm(res ~ FO(x1, x2, x3) + TWI(x1, x2, x3) + PQ(x1, x2, x3),                   data=Design.1_2.coded)summary(rsmModel.2)
#            Estimate Std. Error t value  Pr(>|t|)    #(Intercept) 73.33875    0.79347 92.4275 1.081e-10 ***#x1           1.07125    0.56107  1.9093  0.104810    #x2          -3.14688    0.56107 -5.6087  0.001370 ** #x3          -5.15688    0.56107 -9.1911 9.350e-05 ***#x1:x2        0.35375    0.79347  0.4458  0.671360    #x1:x3        0.01375    0.79347  0.0173  0.986736    #x2:x3       -0.34000    0.79347 -0.4285  0.683246    #x1^2        -4.65813    0.79347 -5.8705  0.001081 ** #x2^2         1.52563    0.79347  1.9227  0.102878    #x3^2         3.00813    0.79347  3.7911  0.009062 ** #---#Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
#Multiple R-squared:  0.9664,    Adjusted R-squared:  0.916 #F-statistic: 19.17 on 9 and 6 DF,  p-value: 0.0009315


LinearModel.2 <- aov(res ~ x1 + x2 + x3 + I(x1^2) + I(x2^2) + I(x3^2),                      data=plancoded)summary(LinearModel.2 )AIC(LinearModel.2)#[1] 61.47997
#valor obtido anteriormente Step:  AIC=14.07
-------------- Próxima Parte ----------
Um anexo em HTML foi limpo...
URL: <http://listas.inf.ufpr.br/pipermail/r-br/attachments/20141122/c6648100/attachment.html>


Mais detalhes sobre a lista de discussão R-br